Sync with NEO/py v1.12

CHANGELOG.rst

 Change History
 ==============
 
+1.12 (2019-04-28)
+-----------------
+
+Most changes in this version focus on the ability to migrate efficiently
+and reliably a big ZODB to NEO, which required changes in the protocol.
+See testSplitAndMakeResilientUsingClone for an example of scenario.
+
+Better cluster management:
+
+- New --new-nid storage option for fast cloning.
+- The number of wanted replicas is now a property of the database, which is
+  modifiable when the cluster is running, and reported by `neoctl print pt`.
+- Better error reporting from the master to neoctl for denied requests.
+- tweak: do not touch cells of nodes that are intended to be dropped.
+- tweak: do not crash when trying to remove all nodes.
+- tweak: new neoctl option to ask the master to simulate.
+- neoctl: better display of full partition tables.
+- master: reject drop/tweak commands that could lead to unwanted status.
+
+Importer:
+
+- Fix possible data loss on writeback.
+- v1.9 broke replication (as source) once the import is finished.
+- Speed up startup when the import is already finished.
+- Fix closure of ZODB, and also do it when the import is finished.
+- Fix hidden "maximum recursion depth exceeded" at startup.
+- Fix resumption when using SQLite.
+- v1.10 broke resumption when there are new transactions since the import
+  started.
+
+MySQL:
+
+- Better support of RocksDB by specifying column families.
+- Fix handling of connection strings (--database) without credentials.
+
 1.11 (2019-03-11)
 -----------------
 

go/neo/client.go

@@ -258,6 +258,37 @@ func (c *Client) talkMaster1(ctx context.Context) (err error) {
 		c.node.MyInfo.UUID = accept.YourUUID
 	}
 
+	wg, ctx := errgroup.WithContext(ctx)	// XXX -> xsync.WorkGroup
+	defer xio.CloseWhenDone(ctx, mlink)()
+
+	// master pushes nodeTab and partTab to us right after identification
+	// XXX merge into -> node.DialMaster?
+
+	// nodeTab
+	mnt := proto.NotifyNodeInformation{}
+	_, err = mlink.Expect1(&mnt)
+	if err != nil {
+		return fmt.Errorf("after identification: %w", err)
+	}
+
+	// partTab
+	mpt := proto.SendPartitionTable{}
+	_, err = mlink.Expect1(&mpt)
+	if err != nil {
+		return fmt.Errorf("after identification: %w", err)
+	}
+	pt := PartTabFromDump(mpt.PTid, mpt.RowList) // TODO handle mpt.NumReplicas
+	log.Infof(ctx, "master initialized us with next parttab:\n%s", pt)
+
+	c.node.StateMu.Lock()
+	c.node.UpdateNodeTab(ctx, &mnt)
+	c.node.PartTab = pt
+	opready := c.updateOperational()
+	c.node.StateMu.Unlock()
+	opready()
+
+
+
 	// set c.mlink and notify waiters
 	c.mlinkMu.Lock()
 	c.mlink = mlink
@@ -266,10 +297,6 @@ func (c *Client) talkMaster1(ctx context.Context) (err error) {
 	c.mlinkMu.Unlock()
 	close(ready)
 
-	wg, ctx := errgroup.WithContext(ctx)	// XXX -> xsync.WorkGroup
-
-	defer xio.CloseWhenDone(ctx, mlink)()
-
 	// when we are done - reset .mlink
 	defer func() {
 		c.mlinkMu.Lock()
@@ -298,21 +325,6 @@ func (c *Client) talkMaster1(ctx context.Context) (err error) {
 func (c *Client) initFromMaster(ctx context.Context, mlink *neonet.NodeLink) (err error) {
 	defer task.Running(&ctx, "init")(&err)
 
-	// query partTab
-	rpt := proto.AnswerPartitionTable{}
-	err = mlink.Ask1(&proto.AskPartitionTable{}, &rpt)
-	if err != nil {
-		return err
-	}
-
-	pt := PartTabFromDump(rpt.PTid, rpt.RowList)
-	log.Infof(ctx, "master initialized us with next parttab:\n%s", pt)
-	c.node.StateMu.Lock()
-	c.node.PartTab = pt
-	opready := c.updateOperational()
-	c.node.StateMu.Unlock()
-	opready()
-
 	// query last_tid
 	lastTxn := proto.AnswerLastTransaction{}
 	err = mlink.Ask1(&proto.LastTransaction{}, &lastTxn)

go/neo/master.go

-// Copyright (C) 2017-2018  Nexedi SA and Contributors.
+// Copyright (C) 2017-2020  Nexedi SA and Contributors.
 //                          Kirill Smelkov <kirr@nexedi.com>
 //
 // This program is free software: you can Use, Study, Modify and Redistribute
@@ -505,7 +505,7 @@ func storCtlRecovery(ctx context.Context, stor *Node, res chan storRecovery) {
 	}
 
 	// reconstruct partition table from response
-	pt := PartTabFromDump(resp.PTid, resp.RowList)
+	pt := PartTabFromDump(resp.PTid, resp.RowList) // TODO handle resp.NumReplicas
 	res <- storRecovery{stor: stor, partTab: pt}
 }
 
@@ -695,8 +695,9 @@ func storCtlVerify(ctx context.Context, stor *Node, pt *PartitionTable, res chan
 
 	// send just recovered parttab so storage saves it
 	err = slink.Send1(&proto.SendPartitionTable{
-		PTid:    pt.PTid,
-		RowList: pt.Dump(),
+		PTid:	     pt.PTid,
+		NumReplicas: 0, // FIXME hardcoded
+		RowList:     pt.Dump(),
 	})
 	if err != nil {
 		return
@@ -898,6 +899,9 @@ func (m *Master) serveClient(ctx context.Context, cli *Node) (err error) {
 	wg, ctx := errgroup.WithContext(ctx)
 	defer xio.CloseWhenDone(ctx, clink)()	// XXX -> cli.CloseLink?
 
+	// FIXME send initial nodeTab and partTab before starting serveClient1
+	// (move those initial sends from keepPeerUpdated to here)
+
 	// M -> C notifications about cluster state
 	wg.Go(func() error {
 		return m.keepPeerUpdated(ctx, clink)
@@ -926,15 +930,6 @@ func (m *Master) serveClient(ctx context.Context, cli *Node) (err error) {
 // serveClient1 prepares response for 1 request from client
 func (m *Master) serveClient1(ctx context.Context, req proto.Msg) (resp proto.Msg) {
 	switch req := req.(type) {
-	case *proto.AskPartitionTable:
-		m.node.StateMu.RLock()
-		rpt := &proto.AnswerPartitionTable{
-			PTid:	 m.node.PartTab.PTid,
-			RowList: m.node.PartTab.Dump(),
-		}
-		m.node.StateMu.RUnlock()
-		return rpt
-
 	case *proto.LastTransaction:
 		// XXX lock
 		return &proto.AnswerLastTransaction{m.lastTid}
@@ -965,6 +960,10 @@ func (m *Master) keepPeerUpdated(ctx context.Context, link *neonet.NodeLink) (er
 		nodeiv[i] = node.NodeInfo
 	}
 
+	ptid := m.node.PartTab.PTid
+	ptnr := uint32(0) // FIXME hardcoded NumReplicas; NEO/py keeps this as n(replica)-1
+	ptv  := m.node.PartTab.Dump()
+
 	// XXX RLock is not enough for subscribe - right?
 	nodech, nodeUnsubscribe := m.node.NodeTab.SubscribeBuffered()
 
@@ -990,6 +989,16 @@ func (m *Master) keepPeerUpdated(ctx context.Context, link *neonet.NodeLink) (er
 		return err
 	}
 
+	err = link.Send1(&proto.SendPartitionTable{
+		PTid:        ptid,
+		NumReplicas: ptnr,
+		RowList:     ptv,
+	})
+	if err != nil {
+		return err
+	}
+
+
 	// now proxy the updates until we are done
 	for {
 		var msg proto.Msg
@@ -1081,8 +1090,6 @@ func (m *Master) identify(ctx context.Context, n nodeCome) (node *Node, resp pro
 	accept := &proto.AcceptIdentification{
 			NodeType:	proto.MASTER,
 			MyUUID:		m.node.MyInfo.UUID,
-			NumPartitions:	1,	// FIXME hardcoded
-			NumReplicas:	0,	// FIXME hardcoded  (neo/py meaning for n(replica) is `n(real-replica) - 1`)
 			YourUUID:	uuid,
 		}
 

go/neo/neo.go

@@ -113,8 +113,9 @@ func (app *NodeApp) Dial(ctx context.Context, peerType proto.NodeType, addr stri
 		UUID:		app.MyInfo.UUID,
 		Address:	app.MyInfo.Addr,
 		ClusterName:	app.ClusterName,
-		DevPath:        nil,                    // XXX stub
 		IdTime:		app.MyInfo.IdTime,	// XXX ok?
+		DevPath:        nil,                    // XXX stub
+		NewNID:         nil,			// XXX stub
 	}
 	accept := &proto.AcceptIdentification{}
 	// FIXME error if peer sends us something with another connID
@@ -323,7 +324,7 @@ func (app *NodeApp) UpdateNodeTab(ctx context.Context, msg *proto.NotifyNodeInfo
 
 // UpdatePartTab applies updates to .PartTab from message and logs changes appropriately.
 func (app *NodeApp) UpdatePartTab(ctx context.Context, msg *proto.SendPartitionTable) {
-	pt := PartTabFromDump(msg.PTid, msg.RowList)
+	pt := PartTabFromDump(msg.PTid, msg.RowList) // FIXME handle msg.NumReplicas
 	// XXX logging under lock
 	log.Infof(ctx, "parttab update: %v", pt)
 	app.PartTab = pt

go/neo/neo_test.go

@@ -107,8 +107,9 @@ func TestMasterStorage(t0 *testing.T) {
 		UUID:		0,
 		Address:	xnaddr("s:1"),
 		ClusterName:	"abc1",
-		DevPath:        nil,
 		IdTime:		proto.IdTimeNone,
+		DevPath:        nil,
+		NewNID:         nil,
 	}))
 
 
@@ -117,8 +118,6 @@ func TestMasterStorage(t0 *testing.T) {
 	tSM.Expect(conntx("m:2", "s:2", 1, &proto.AcceptIdentification{
 		NodeType:	proto.MASTER,
 		MyUUID:		proto.UUID(proto.MASTER, 1),
-		NumPartitions:	1,
-		NumReplicas:	0,
 		YourUUID:	proto.UUID(proto.STORAGE, 1),
 	}))
 
@@ -136,6 +135,7 @@ func TestMasterStorage(t0 *testing.T) {
 	tMS.Expect(conntx("m:2", "s:2", 2, &proto.AskPartitionTable{}))
 	tMS.Expect(conntx("s:2", "m:2", 2, &proto.AnswerPartitionTable{
 		PTid:		0,
+		NumReplicas:	0,
 		RowList:	[]proto.RowInfo{},
 	}))
 
@@ -162,8 +162,9 @@ func TestMasterStorage(t0 *testing.T) {
 
 	tMS.Expect(conntx("m:2", "s:2", 4, &proto.SendPartitionTable{
 		PTid:		1,
+		NumReplicas:    0,
 		RowList:	[]proto.RowInfo{
-			{0, []proto.CellInfo{{proto.UUID(proto.STORAGE, 1), proto.UP_TO_DATE}}},
+			{[]proto.CellInfo{{proto.UUID(proto.STORAGE, 1), proto.UP_TO_DATE}}},
 		},
 	}))
 
@@ -209,8 +210,9 @@ func TestMasterStorage(t0 *testing.T) {
 		UUID:		0,
 		Address:	xnaddr(""),
 		ClusterName:	"abc1",
-		DevPath:        nil,
 		IdTime:		proto.IdTimeNone,
+		DevPath:        nil,
+		NewNID:         nil,
 	}))
 
 	tM.Expect(δnode("m", "", proto.CLIENT, 1, proto.RUNNING, 0.02))
@@ -218,25 +220,10 @@ func TestMasterStorage(t0 *testing.T) {
 	tCM.Expect(conntx("m:3", "c:1", 1, &proto.AcceptIdentification{
 		NodeType:	proto.MASTER,
 		MyUUID:		proto.UUID(proto.MASTER, 1),
-		NumPartitions:	1,
-		NumReplicas:	0,
 		YourUUID:	proto.UUID(proto.CLIENT, 1),
 	}))
 
-	// C asks M about PT and last_tid
-	// NOTE this might come in parallel with vvv "C <- M NotifyNodeInformation C1,M1,S1"
-	tCM.Expect(conntx("c:1", "m:3", 3, &proto.AskPartitionTable{}))
-	tCM.Expect(conntx("m:3", "c:1", 3, &proto.AnswerPartitionTable{
-		PTid:		1,
-		RowList:	[]proto.RowInfo{
-			{0, []proto.CellInfo{{proto.UUID(proto.STORAGE, 1), proto.UP_TO_DATE}}},
-		},
-	}))
-	tCM.Expect(conntx("c:1", "m:3", 5, &proto.LastTransaction{}))
-	tCM.Expect(conntx("m:3", "c:1", 5, &proto.AnswerLastTransaction{lastTid}))
-
 	// C <- M NotifyNodeInformation C1,M1,S1
-	// NOTE this might come in parallel with ^^^ "C asks M about PT"
 	tMC.Expect(conntx("m:3", "c:1", 0, &proto.NotifyNodeInformation{
 		IdTime:		proto.IdTimeNone,	// XXX ?
 		NodeList:	[]proto.NodeInfo{
@@ -246,10 +233,22 @@ func TestMasterStorage(t0 *testing.T) {
 		},
 	}))
 
+	tMC.Expect(conntx("m:3", "c:1", 2, &proto.SendPartitionTable{
+		PTid:		1,
+		NumReplicas:	0,
+		RowList:	[]proto.RowInfo{
+			{[]proto.CellInfo{{proto.UUID(proto.STORAGE, 1), proto.UP_TO_DATE}}},
+		},
+	}))
+
 	tC.Expect(δnode("c", "m:1", proto.MASTER,  1, proto.RUNNING, proto.IdTimeNone))
 	tC.Expect(δnode("c", "s:1", proto.STORAGE, 1, proto.RUNNING, 0.01))
 	tC.Expect(δnode("c", "",    proto.CLIENT,  1, proto.RUNNING, 0.02))
 
+	// C asks M last_tid
+	tCM.Expect(conntx("c:1", "m:3", 3, &proto.LastTransaction{}))
+	tCM.Expect(conntx("m:3", "c:1", 3, &proto.AnswerLastTransaction{lastTid}))
+
 	// ----------------------------------------
 
 	// C asks M about last tid	XXX better master sends it itself on new client connected
@@ -262,8 +261,8 @@ func TestMasterStorage(t0 *testing.T) {
 		}
 	})
 
-	tCM.Expect(conntx("c:1", "m:3", 7, &proto.LastTransaction{}))
-	tCM.Expect(conntx("m:3", "c:1", 7, &proto.AnswerLastTransaction{
+	tCM.Expect(conntx("c:1", "m:3", 5, &proto.LastTransaction{}))
+	tCM.Expect(conntx("m:3", "c:1", 5, &proto.AnswerLastTransaction{
 		Tid: lastTid,
 	}))
 
@@ -295,15 +294,14 @@ func TestMasterStorage(t0 *testing.T) {
 		UUID:		proto.UUID(proto.CLIENT, 1),
 		Address:	xnaddr(""),
 		ClusterName:	"abc1",
-		DevPath:        nil,
 		IdTime:		0.02,
+		DevPath:        nil,
+		NewNID:         nil,
 	}))
 
 	tCS.Expect(conntx("s:3", "c:2", 1, &proto.AcceptIdentification{
 		NodeType:	proto.STORAGE,
 		MyUUID:		proto.UUID(proto.STORAGE, 1),
-		NumPartitions:	1,
-		NumReplicas:	0,
 		YourUUID:	proto.UUID(proto.CLIENT, 1),
 	}))
 

go/neo/neonet/connection.go

@@ -58,7 +58,7 @@
 // object. The Request contains the message received and internally the
 // connection. A response can be sent back via Request.Reply. Then once
 // Request.Close is called the connection object that was accepted is
-// immediately put back into pool for later reuse.
+// immediately put back into pool for later reuse. XXX Expect1
 package neonet
 
 // XXX neonet compatibility with NEO/py depends on the following small NEO/py patch:
@@ -1552,6 +1552,24 @@ func (link *NodeLink) Send1(msg proto.Msg) error {
 	return err
 }
 
+// Expect1 receives notification in 1-1 model.
+//
+// See Conn.Expect for semantic details.
+//
+// See "Lightweight mode" in top-level package doc for overview.
+func (link *NodeLink) Expect1(msgv ...proto.Msg) (which int, err error) {
+	// XXX a bit dup wrt Recv1
+	conn, err := link.Accept()
+	if err != nil {
+		return -1, err
+	}
+
+	// NOTE serveRecv guaranty that when a conn is accepted, there is 1 message in conn.rxq
+	which, err = conn.Expect(msgv...)
+	conn.lightClose()
+	return which, err
+}
+
 
 // Ask1 sends request and receives response in 1-1 model.
 //

go/neo/parttab.go

-// Copyright (C) 2017  Nexedi SA and Contributors.
-//                     Kirill Smelkov <kirr@nexedi.com>
+// Copyright (C) 2017-2020  Nexedi SA and Contributors.
+//                          Kirill Smelkov <kirr@nexedi.com>
 //
 // This program is free software: you can Use, Study, Modify and Redistribute
 // it under the terms of the GNU General Public License version 3, or (at your
@@ -254,7 +254,7 @@ func (pt *PartitionTable) Dump() []proto.RowInfo { // XXX also include .ptid? ->
 			cellv[j] = cell.CellInfo
 		}
 
-		rowv[i] = proto.RowInfo{Offset: uint32(i), CellList: cellv}	// XXX cast?
+		rowv[i] = proto.RowInfo{CellList: cellv}
 	}
 	return rowv
 }
@@ -264,9 +264,8 @@ func PartTabFromDump(ptid proto.PTid, rowv []proto.RowInfo) *PartitionTable {
 	pt := &PartitionTable{}
 	pt.PTid = ptid
 
-	for _, row := range rowv {
-		i := row.Offset
-		for i >= uint32(len(pt.tab)) {
+	for i, row := range rowv {
+		for i >= len(pt.tab) {
 			pt.tab = append(pt.tab, []Cell{})
 		}
 

go/neo/proto/proto.go

@@ -81,7 +81,7 @@ const (
 	// The protocol version must be increased whenever upgrading a node may require
 	// to upgrade other nodes. It is encoded as a 4-bytes big-endian integer and
 	// the high order byte 0 is different from TLS Handshake (0x16).
-	Version = 5
+	Version = 6
 
 	// length of packet header
 	PktHeaderLen = 10 // = unsafe.Sizeof(PktHeader{}), but latter gives typed constant (uintptr)
@@ -138,6 +138,7 @@ var ErrDecodeOverflow = errors.New("decode: buffer overflow")
 type ErrorCode int8
 const (
 	ACK ErrorCode = iota
+	DENIED
 	NOT_READY
 	OID_NOT_FOUND
 	TID_NOT_FOUND
@@ -346,7 +347,6 @@ type CellInfo struct {
 
 //neo:proto typeonly
 type RowInfo struct {
-	Offset   uint32 // PNumber	XXX -> Pid
 	CellList []CellInfo
 }
 
@@ -372,16 +372,16 @@ type RequestIdentification struct {
 	UUID        NodeUUID
 	Address     Address // where requesting node is also accepting connections
 	ClusterName string
-	DevPath     []string // [] of devid
 	IdTime      IdTime
+	// storage
+	DevPath     []string // [] of devid
+	NewNID      []uint32 // [] of PNumber
 }
 
 //neo:proto answer
 type AcceptIdentification struct {
 	NodeType      NodeType // XXX name
 	MyUUID        NodeUUID
-	NumPartitions uint32 // PNumber
-	NumReplicas   uint32 // PNumber
 	YourUUID      NodeUUID
 }
 
@@ -455,23 +455,24 @@ type AnswerLastIDs struct {
 }
 
 // Ask storage node the remaining data needed by master to recover.
-// This is also how the clients get the full partition table on connection.
 //
-//neo:nodes M -> S; C -> M
+//neo:nodes M -> S
 type AskPartitionTable struct {
 }
 
 type AnswerPartitionTable struct {
 	PTid
-	RowList []RowInfo
+	NumReplicas uint32	// PNumber
+	RowList     []RowInfo
 }
 
-// Send the full partition table to admin/storage nodes on connection.
+// Send the full partition table to admin/client/storage nodes on connection.
 //
-//neo:nodes M -> A, S
+//neo:nodes M -> A, C, S
 type SendPartitionTable struct {
 	PTid
-	RowList []RowInfo
+	NumReplicas uint32	// PNumber
+	RowList     []RowInfo
 }
 
 // Notify about changes in the partition table.
@@ -479,7 +480,8 @@ type SendPartitionTable struct {
 //neo:nodes M -> *
 type NotifyPartitionChanges struct {
 	PTid
-	CellList []struct {
+	NumReplicas uint32 // PNumber
+	CellList    []struct {
 		Offset   uint32 // PNumber	XXX -> Pid
 		CellInfo CellInfo
 	}
@@ -804,7 +806,8 @@ type PartitionList struct {
 
 type AnswerPartitionList struct {
 	PTid
-	RowList []RowInfo
+	NumReplicas uint32 // PNumber
+	RowList     []RowInfo
 }
 
 // Ask information about nodes.
@@ -843,11 +846,24 @@ type AddPendingNodes struct {
 //
 //neo:nodes ctl -> A -> M
 type TweakPartitionTable struct {
+	DryRun   bool
 	NodeList []NodeUUID
 
 	// answer = Error
 }
 
+type AnswerTweakPartitionTable struct {
+	Changed  bool
+	RowList  []RowInfo
+}
+
+// Set the number of replicas.
+//
+//neo:nodes ctl -> A -> M
+type SetNumReplicas struct {
+	NumReplicas uint32 // PNumber
+}
+
 // Set the cluster state.
 //
 //neo:nodes ctl -> A -> M

go/neo/proto/proto_test.go

@@ -189,19 +189,21 @@ func TestMsgMarshal(t *testing.T) {
 
 		// PTid, [] (of [] of {UUID, CellState})
 		{&AnswerPartitionTable{
-			PTid:    0x0102030405060708,
+			PTid:        0x0102030405060708,
+			NumReplicas: 34,
 			RowList: []RowInfo{
-				{1, []CellInfo{{11, UP_TO_DATE}, {17, OUT_OF_DATE}}},
-				{2, []CellInfo{{11, FEEDING}}},
-				{7, []CellInfo{{11, CORRUPTED}, {15, DISCARDED}, {23, UP_TO_DATE}}},
+				{[]CellInfo{{11, UP_TO_DATE}, {17, OUT_OF_DATE}}},
+				{[]CellInfo{{11, FEEDING}}},
+				{[]CellInfo{{11, CORRUPTED}, {15, DISCARDED}, {23, UP_TO_DATE}}},
 			},
 			},
 
 			hex("0102030405060708") +
+			hex("00000022") +
 			hex("00000003") +
-				hex("00000001000000020000000b010000001100") +
-				hex("00000002000000010000000b02") +
-				hex("00000007000000030000000b030000000f040000001701"),
+				hex("000000020000000b010000001100") +
+				hex("000000010000000b02") +
+				hex("000000030000000b030000000f040000001701"),
 		},
 
 		// map[Oid]struct {Tid,Tid,bool}
@@ -250,13 +252,14 @@ func TestMsgMarshal(t *testing.T) {
 		},
 
 		// uint32, Address, string, IdTime
-		{&RequestIdentification{CLIENT, 17, Address{"localhost", 7777}, "myname", []string{"room1", "rack234"}, 0.12345678},
+		{&RequestIdentification{CLIENT, 17, Address{"localhost", 7777}, "myname", 0.12345678, []string{"room1", "rack234"}, []uint32{3,4,5} },
 
 			u8(2) + u32(17) + u32(9) +
 			"localhost" + u16(7777) +
 			u32(6) + "myname" +
+			hex("3fbf9add1091c895") +
 			u32(2) + u32(5)+"room1" + u32(7)+"rack234" +
-			hex("3fbf9add1091c895"),
+			u32(3) + u32(3)+u32(4)+u32(5),
 		},
 
 		// IdTime, empty Address, int32

go/neo/proto/zproto-str.go

@@ -9,22 +9,23 @@ func _() {
 	// Re-run the stringer command to generate them again.
 	var x [1]struct{}
 	_ = x[ACK-0]
-	_ = x[NOT_READY-1]
-	_ = x[OID_NOT_FOUND-2]
-	_ = x[TID_NOT_FOUND-3]
-	_ = x[OID_DOES_NOT_EXIST-4]
-	_ = x[PROTOCOL_ERROR-5]
-	_ = x[REPLICATION_ERROR-6]
-	_ = x[CHECKING_ERROR-7]
-	_ = x[BACKEND_NOT_IMPLEMENTED-8]
-	_ = x[NON_READABLE_CELL-9]
-	_ = x[READ_ONLY_ACCESS-10]
-	_ = x[INCOMPLETE_TRANSACTION-11]
+	_ = x[DENIED-1]
+	_ = x[NOT_READY-2]
+	_ = x[OID_NOT_FOUND-3]
+	_ = x[TID_NOT_FOUND-4]
+	_ = x[OID_DOES_NOT_EXIST-5]
+	_ = x[PROTOCOL_ERROR-6]
+	_ = x[REPLICATION_ERROR-7]
+	_ = x[CHECKING_ERROR-8]
+	_ = x[BACKEND_NOT_IMPLEMENTED-9]
+	_ = x[NON_READABLE_CELL-10]
+	_ = x[READ_ONLY_ACCESS-11]
+	_ = x[INCOMPLETE_TRANSACTION-12]
 }
 
-const _ErrorCode_name = "ACKNOT_READYOID_NOT_FOUNDTID_NOT_FOUNDOID_DOES_NOT_EXISTPROTOCOL_ERRORREPLICATION_ERRORCHECKING_ERRORBACKEND_NOT_IMPLEMENTEDNON_READABLE_CELLREAD_ONLY_ACCESSINCOMPLETE_TRANSACTION"
+const _ErrorCode_name = "ACKDENIEDNOT_READYOID_NOT_FOUNDTID_NOT_FOUNDOID_DOES_NOT_EXISTPROTOCOL_ERRORREPLICATION_ERRORCHECKING_ERRORBACKEND_NOT_IMPLEMENTEDNON_READABLE_CELLREAD_ONLY_ACCESSINCOMPLETE_TRANSACTION"
 
-var _ErrorCode_index = [...]uint8{0, 3, 12, 25, 38, 56, 70, 87, 101, 124, 141, 157, 179}
+var _ErrorCode_index = [...]uint8{0, 3, 9, 18, 31, 44, 62, 76, 93, 107, 130, 147, 163, 185}
 
 func (i ErrorCode) String() string {
 	if i < 0 || i >= ErrorCode(len(_ErrorCode_index)-1) {

go/neo/proto/ztestdata_proto_py_test.go

@@ -42,31 +42,32 @@ var pyMsgRegistry = map[uint16]string{
 	38:	"SetNodeState",
 	39:	"AddPendingNodes",
 	40:	"TweakPartitionTable",
-	41:	"SetClusterState",
-	42:	"Repair",
-	43:	"RepairOne",
-	44:	"NotifyClusterState",
-	45:	"AskClusterState",
-	46:	"ObjectUndoSerial",
-	47:	"AskTIDsFrom",
-	48:	"Pack",
-	49:	"CheckReplicas",
-	50:	"CheckPartition",
-	51:	"CheckTIDRange",
-	52:	"CheckSerialRange",
-	53:	"PartitionCorrupted",
-	54:	"NotifyReady",
-	55:	"LastTransaction",
-	56:	"CheckCurrentSerial",
-	57:	"NotifyTransactionFinished",
-	58:	"Replicate",
-	59:	"ReplicationDone",
-	60:	"FetchTransactions",
-	61:	"FetchObjects",
-	62:	"AddTransaction",
-	63:	"AddObject",
-	64:	"Truncate",
-	65:	"FlushLog",
+	41:	"SetNumReplicas",
+	42:	"SetClusterState",
+	43:	"Repair",
+	44:	"RepairOne",
+	45:	"NotifyClusterState",
+	46:	"AskClusterState",
+	47:	"ObjectUndoSerial",
+	48:	"AskTIDsFrom",
+	49:	"Pack",
+	50:	"CheckReplicas",
+	51:	"CheckPartition",
+	52:	"CheckTIDRange",
+	53:	"CheckSerialRange",
+	54:	"PartitionCorrupted",
+	55:	"NotifyReady",
+	56:	"LastTransaction",
+	57:	"CheckCurrentSerial",
+	58:	"NotifyTransactionFinished",
+	59:	"Replicate",
+	60:	"ReplicationDone",
+	61:	"FetchTransactions",
+	62:	"FetchObjects",
+	63:	"AddTransaction",
+	64:	"AddObject",
+	65:	"Truncate",
+	66:	"FlushLog",
 	32768:	"Error",
 	32769:	"AcceptIdentification",
 	32770:	"Pong",
@@ -92,14 +93,15 @@ var pyMsgRegistry = map[uint16]string{
 	32803:	"AnswerObjectHistory",
 	32804:	"AnswerPartitionList",
 	32805:	"AnswerNodeList",
-	32813:	"AnswerClusterState",
-	32814:	"AnswerObjectUndoSerial",
-	32815:	"AnswerTIDsFrom",
-	32816:	"AnswerPack",
-	32819:	"AnswerCheckTIDRange",
-	32820:	"AnswerCheckSerialRange",
-	32823:	"AnswerLastTransaction",
-	32824:	"AnswerCheckCurrentSerial",
-	32828:	"AnswerFetchTransactions",
-	32829:	"AnswerFetchObjects",
+	32808:	"AnswerTweakPartitionTable",
+	32814:	"AnswerClusterState",
+	32815:	"AnswerObjectUndoSerial",
+	32816:	"AnswerTIDsFrom",
+	32817:	"AnswerPack",
+	32820:	"AnswerCheckTIDRange",
+	32821:	"AnswerCheckSerialRange",
+	32824:	"AnswerLastTransaction",
+	32825:	"AnswerCheckCurrentSerial",
+	32829:	"AnswerFetchTransactions",
+	32830:	"AnswerFetchObjects",
 }

go/neo/storage.go

@@ -203,10 +203,13 @@ func (stor *Storage) talkMaster1(ctx context.Context) (err error) {
 
 	// XXX add master UUID -> nodeTab ? or master will notify us with it himself ?
 
+// XXX move -> SetNumReplicas handler
+/*
 	// NumReplicas: neo/py meaning for n(replica) = `n(real-replica) - 1`
 	if !(accept.NumPartitions == 1 && accept.NumReplicas == 0) {
 		return fmt.Errorf("TODO for 1-storage POC: Npt: %v  Nreplica: %v", accept.NumPartitions, accept.NumReplicas)
 	}
+*/
 
 	// XXX -> node.Dial ?
 	if accept.YourUUID != stor.node.MyInfo.UUID {
@@ -284,8 +287,9 @@ func (stor *Storage) m1initialize1(ctx context.Context, req neonet.Request) erro
 	case *proto.AskPartitionTable:
 		// TODO initially read PT from disk
 		err = req.Reply(&proto.AnswerPartitionTable{
-			PTid:	 stor.node.PartTab.PTid,
-			RowList: stor.node.PartTab.Dump()})
+			PTid:	     stor.node.PartTab.PTid,
+			NumReplicas: 0, // FIXME hardcoded; NEO/py uses this as n(replica)-1
+			RowList:     stor.node.PartTab.Dump()})
 
 	case *proto.LockedTransactions:
 		// XXX r/o stub
@@ -304,7 +308,7 @@ func (stor *Storage) m1initialize1(ctx context.Context, req neonet.Request) erro
 
 	case *proto.SendPartitionTable:
 		// TODO M sends us whole PT -> save locally
-		stor.node.UpdatePartTab(ctx, msg)	// XXX lock?
+		stor.node.UpdatePartTab(ctx, msg)	// XXX lock?  XXX handle msg.NumReplicas
 
 	case *proto.NotifyPartitionChanges:
 		// TODO M sends us δPT -> save locally?
@@ -412,8 +416,6 @@ func (stor *Storage) identify(idReq *proto.RequestIdentification) (proto.Msg, bo
 	return &proto.AcceptIdentification{
 		NodeType:	stor.node.MyInfo.Type,
 		MyUUID:		stor.node.MyInfo.UUID,		// XXX lock wrt update
-		NumPartitions:	1,	// XXX
-		NumReplicas:	0,	// XXX
 		YourUUID:	idReq.UUID,
 	}, true
 }

neo/admin/app.py

@@ -21,7 +21,6 @@ from neo.lib.exception import PrimaryFailure
 from .handler import AdminEventHandler, MasterEventHandler, \
     MasterRequestEventHandler
 from neo.lib.bootstrap import BootstrapManager
-from neo.lib.pt import PartitionTable
 from neo.lib.protocol import ClusterStates, Errors, NodeTypes, Packets
 from neo.lib.debug import register as registerLiveDebugger
 
@@ -36,8 +35,8 @@ class Application(BaseApplication):
         cls.addCommonServerOptions('admin', '127.0.0.1:9999')
 
         _ = _.group('admin')
-        _.int('u', 'uuid',
-            help="specify an UUID to use for this process (testing purpose)")
+        _.int('i', 'nid',
+            help="specify an NID to use for this process (testing purpose)")
 
     def __init__(self, config):
         super(Application, self).__init__(
@@ -53,7 +52,7 @@ class Application(BaseApplication):
         # The partition table is initialized after getting the number of
         # partitions.
         self.pt = None
-        self.uuid = config.get('uuid')
+        self.uuid = config.get('nid')
         logging.node(self.name, self.uuid)
         self.request_handler = MasterRequestEventHandler(self)
         self.master_event_handler = MasterEventHandler(self)
@@ -66,7 +65,6 @@ class Application(BaseApplication):
         super(Application, self).close()
 
     def reset(self):
-        self.bootstrapped = False
         self.master_conn = None
         self.master_node = None
 
@@ -117,40 +115,20 @@ class Application(BaseApplication):
         self.cluster_state = None
         # search, find, connect and identify to the primary master
         bootstrap = BootstrapManager(self, NodeTypes.ADMIN, self.server)
-        self.master_node, self.master_conn, num_partitions, num_replicas = \
-            bootstrap.getPrimaryConnection()
-
-        if self.pt is None:
-            self.pt = PartitionTable(num_partitions, num_replicas)
-        elif self.pt.getPartitions() != num_partitions:
-            # XXX: shouldn't we recover instead of raising ?
-            raise RuntimeError('the number of partitions is inconsistent')
-        elif self.pt.getReplicas() != num_replicas:
-            # XXX: shouldn't we recover instead of raising ?
-            raise RuntimeError('the number of replicas is inconsistent')
+        self.master_node, self.master_conn = bootstrap.getPrimaryConnection()
 
         # passive handler
         self.master_conn.setHandler(self.master_event_handler)
         self.master_conn.ask(Packets.AskClusterState())
-        self.master_conn.ask(Packets.AskPartitionTable())
 
     def sendPartitionTable(self, conn, min_offset, max_offset, uuid):
-        # we have a pt
-        self.pt.log()
-        row_list = []
+        pt = self.pt
         if max_offset == 0:
-            max_offset = self.pt.getPartitions()
+            max_offset = pt.getPartitions()
         try:
-            for offset in xrange(min_offset, max_offset):
-                row = []
-                try:
-                    for cell in self.pt.getCellList(offset):
-                        if uuid is None or cell.getUUID() == uuid:
-                            row.append((cell.getUUID(), cell.getState()))
-                except TypeError:
-                    pass
-                row_list.append((offset, row))
+            row_list = map(pt.getRow, xrange(min_offset, max_offset))
         except IndexError:
             conn.send(Errors.ProtocolError('invalid partition table offset'))
         else:
-            conn.answer(Packets.AnswerPartitionList(self.pt.getID(), row_list))
+            conn.answer(Packets.AnswerPartitionList(
+                pt.getID(), pt.getReplicas(), row_list))

neo/admin/handler.py

@@ -17,11 +17,12 @@
 from neo.lib import logging, protocol
 from neo.lib.handler import EventHandler
 from neo.lib.protocol import uuid_str, Packets
+from neo.lib.pt import PartitionTable
 from neo.lib.exception import PrimaryFailure
 
 def check_primary_master(func):
     def wrapper(self, *args, **kw):
-        if self.app.bootstrapped:
+        if self.app.master_conn is not None:
             return func(self, *args, **kw)
         raise protocol.NotReadyError('Not connected to a primary master.')
     return wrapper
@@ -74,6 +75,7 @@ class AdminEventHandler(EventHandler):
     tweakPartitionTable = forward_ask(Packets.TweakPartitionTable)
     setClusterState = forward_ask(Packets.SetClusterState)
     setNodeState = forward_ask(Packets.SetNodeState)
+    setNumReplicas = forward_ask(Packets.SetNumReplicas)
     checkReplicas = forward_ask(Packets.CheckReplicas)
     truncate = forward_ask(Packets.Truncate)
     repair = forward_ask(Packets.Repair)
@@ -112,16 +114,12 @@ class MasterEventHandler(EventHandler):
     def answerClusterState(self, conn, state):
         self.app.cluster_state = state
 
-    def notifyPartitionChanges(self, conn, ptid, cell_list):
-        self.app.pt.update(ptid, cell_list, self.app.nm)
+    def sendPartitionTable(self, conn, ptid, num_replicas, row_list):
+        pt = self.app.pt = object.__new__(PartitionTable)
+        pt.load(ptid, num_replicas, row_list, self.app.nm)
 
-    def answerPartitionTable(self, conn, ptid, row_list):
-        self.app.pt.load(ptid, row_list, self.app.nm)
-        self.app.bootstrapped = True
-
-    def sendPartitionTable(self, conn, ptid, row_list):
-        if self.app.bootstrapped:
-            self.app.pt.load(ptid, row_list, self.app.nm)
+    def notifyPartitionChanges(self, conn, ptid, num_replicas, cell_list):
+        self.app.pt.update(ptid, num_replicas, cell_list, self.app.nm)
 
     def notifyClusterInformation(self, conn, cluster_state):
         self.app.cluster_state = cluster_state

neo/client/app.py

@@ -76,7 +76,7 @@ class Application(ThreadedApplication):
         self.primary_master_node = None
         self.trying_master_node = None
 
-        # no self-assigned UUID, primary master will supply us one
+        # no self-assigned NID, primary master will supply us one
         self._cache = ClientCache() if cache_size is None else \
                       ClientCache(max_size=cache_size)
         self._loading_oid = None
@@ -226,8 +226,8 @@ class Application(ThreadedApplication):
                         self.notifications_handler,
                         node=node,
                         dispatcher=self.dispatcher)
-                p = Packets.RequestIdentification(
-                    NodeTypes.CLIENT, self.uuid, None, self.name, (), None)
+                p = Packets.RequestIdentification(NodeTypes.CLIENT,
+                    self.uuid, None, self.name, None, (), ())
                 try:
                     ask(conn, p, handler=handler)
                 except ConnectionClosed:
@@ -244,7 +244,6 @@ class Application(ThreadedApplication):
                 # operational. Might raise ConnectionClosed so that the new
                 # primary can be looked-up again.
                 logging.info('Initializing from master')
-                ask(conn, Packets.AskPartitionTable(), handler=handler)
                 ask(conn, Packets.AskLastTransaction(), handler=handler)
                 if self.pt.operational():
                     break
@@ -270,7 +269,7 @@ class Application(ThreadedApplication):
         conn = MTClientConnection(self, self.storage_event_handler, node,
                                   dispatcher=self.dispatcher)
         p = Packets.RequestIdentification(NodeTypes.CLIENT,
-            self.uuid, None, self.name, (), self.id_timestamp)
+            self.uuid, None, self.name, self.id_timestamp, (), ())
         try:
             self._ask(conn, p, handler=self.storage_bootstrap_handler)
         except ConnectionClosed:

neo/client/handlers/master.py

@@ -26,10 +26,6 @@ from ..exception import NEOStorageError
 class PrimaryBootstrapHandler(AnswerBaseHandler):
     """ Bootstrap handler used when looking for the primary master """
 
-    def answerPartitionTable(self, conn, ptid, row_list):
-        assert row_list
-        self.app.pt.load(ptid, row_list, self.app.nm)
-
     def answerLastTransaction(*args):
         pass
 
@@ -42,9 +38,6 @@ class PrimaryNotificationsHandler(MTEventHandler):
         except PrimaryElected, e:
             self.app.primary_master_node, = e.args
 
-    def _acceptIdentification(self, node, num_partitions, num_replicas):
-        self.app.pt = PartitionTable(num_partitions, num_replicas)
-
     def answerLastTransaction(self, conn, ltid):
         app = self.app
         app_last_tid = app.__dict__.get('last_tid', '')
@@ -134,9 +127,12 @@ class PrimaryNotificationsHandler(MTEventHandler):
         finally:
             app._cache_lock_release()
 
-    def notifyPartitionChanges(self, conn, ptid, cell_list):
-        if self.app.pt.filled():    # XXX wrong - updating only when already filled ?
-            self.app.pt.update(ptid, cell_list, self.app.nm)
+    def sendPartitionTable(self, conn, ptid, num_replicas, row_list):
+        pt = self.app.pt = object.__new__(PartitionTable)
+        pt.load(ptid, num_replicas, row_list, self.app.nm)
+
+    def notifyPartitionChanges(self, conn, ptid, num_replicas, cell_list):
+        self.app.pt.update(ptid, num_replicas, cell_list, self.app.nm)
 
     def notifyNodeInformation(self, conn, timestamp, node_list):
         super(PrimaryNotificationsHandler, self).notifyNodeInformation(

neo/lib/bootstrap.py

@@ -26,7 +26,7 @@ class BootstrapManager(EventHandler):
     Manage the bootstrap stage, lookup for the primary master then connect to it
     """
 
-    def __init__(self, app, node_type, server=None, devpath=()):
+    def __init__(self, app, node_type, server=None, devpath=(), new_nid=()):
         """
         Manage the bootstrap stage of a non-master node, it lookup for the
         primary master node, connect to it then returns when the master node
@@ -34,9 +34,8 @@ class BootstrapManager(EventHandler):
         """
         self.server = server
         self.devpath = devpath
+        self.new_nid = new_nid
         self.node_type = node_type
-        self.num_replicas = None
-        self.num_partitions = None
         app.nm.reset()
 
     uuid = property(lambda self: self.app.uuid)
@@ -44,7 +43,7 @@ class BootstrapManager(EventHandler):
     def connectionCompleted(self, conn):
         EventHandler.connectionCompleted(self, conn)
         conn.ask(Packets.RequestIdentification(self.node_type, self.uuid,
-            self.server, self.app.name, self.devpath, None))
+            self.server, self.app.name, None, self.devpath, self.new_nid))
 
     def connectionFailed(self, conn):
         EventHandler.connectionFailed(self, conn)
@@ -53,10 +52,8 @@ class BootstrapManager(EventHandler):
     def connectionLost(self, conn, new_state):
         self.current = None
 
-    def _acceptIdentification(self, node, num_partitions, num_replicas):
+    def _acceptIdentification(self, node):
         assert self.current is node, (self.current, node)
-        self.num_partitions = num_partitions
-        self.num_replicas = num_replicas
 
     def getPrimaryConnection(self):
         """
@@ -73,8 +70,7 @@ class BootstrapManager(EventHandler):
             try:
                 while self.current:
                     if self.current.isIdentified():
-                        return (self.current, self.current.getConnection(),
-                            self.num_partitions, self.num_replicas)
+                        return self.current, self.current.getConnection()
                     poll(1)
             except PrimaryElected, e:
                 if self.current:

neo/lib/connection.py

@@ -210,7 +210,7 @@ class BaseConnection(object):
 
     def _getReprInfo(self):
         r = [
-            ('uuid', uuid_str(self.getUUID())),
+            ('nid', uuid_str(self.getUUID())),
             ('address', ('[%s]:%s' if ':' in self.addr[0] else '%s:%s')
                         % self.addr if self.addr else '?'),
             ('handler', self.getHandler()),

neo/lib/handler.py

@@ -26,6 +26,9 @@ from .protocol import (NodeStates, NodeTypes, Packets, uuid_str,
 from .util import cached_property
 
 
+class AnswerDenied(Exception):
+    """Helper exception to stop packet processing and answer a Denied error"""
+
 class DelayEvent(Exception):
     pass
 
@@ -98,6 +101,8 @@ class EventHandler(object):
                 % (m.im_class.__module__, m.im_class.__name__, m.__name__)))
         except NonReadableCell, e:
             conn.answer(Errors.NonReadableCell())
+        except AnswerDenied, e:
+            conn.answer(Errors.Denied(str(e)))
         except AssertionError:
             e = sys.exc_info()
             try:
@@ -160,8 +165,7 @@ class EventHandler(object):
     def _acceptIdentification(*args):
         pass
 
-    def acceptIdentification(self, conn, node_type, uuid,
-                             num_partitions, num_replicas, your_uuid):
+    def acceptIdentification(self, conn, node_type, uuid, your_uuid):
         app = self.app
         node = app.nm.getByAddress(conn.getAddress())
         assert node.getConnection() is conn, (node.getConnection(), conn)
@@ -180,7 +184,7 @@ class EventHandler(object):
             elif node.getUUID() != uuid or app.uuid != your_uuid != None:
                 raise ProtocolError('invalid uuids')
             node.setIdentified()
-            self._acceptIdentification(node, num_partitions, num_replicas)
+            self._acceptIdentification(node)
             return
         conn.close()
 

neo/lib/node.py

@@ -486,7 +486,7 @@ class NodeManager(EventQueue):
             # For the first notification, we receive a full list of nodes from
             # the master. Remove all unknown nodes from a previous connection.
             for node in self._node_set.difference(added_list):
-                if app.pt.dropNode(node):
+                if not node.isStorage() or app.pt.dropNode(node):
                     self.remove(node)
         self.log()
         self.executeQueuedEvents()

neo/lib/protocol.py

@@ -22,7 +22,7 @@ from struct import Struct
 # The protocol version must be increased whenever upgrading a node may require
 # to upgrade other nodes. It is encoded as a 4-bytes big-endian integer and
 # the high order byte 0 is different from TLS Handshake (0x16).
-PROTOCOL_VERSION = 5
+PROTOCOL_VERSION = 6
 ENCODED_VERSION = Struct('!L').pack(PROTOCOL_VERSION)
 
 # Avoid memory errors on corrupted data.
@@ -62,6 +62,7 @@ class Enum(tuple):
 @Enum
 def ErrorCodes():
     ACK
+    DENIED
     NOT_READY
     OID_NOT_FOUND
     TID_NOT_FOUND
@@ -617,10 +618,7 @@ PFCellList = PList('cell_list',
 )
 
 PFRowList = PList('row_list',
-    PStruct('row',
-        PNumber('offset'),
-        PFCellList,
-    ),
+    PFCellList,
 )
 
 PFHistoryList = PList('history_list',
@@ -686,15 +684,15 @@ class RequestIdentification(Packet):
         PUUID('uuid'),
         PAddress('address'),
         PString('name'),
-        PList('devpath', PString('devid')),
         PFloat('id_timestamp'),
+        # storage:
+            PList('devpath', PString('devid')),
+            PList('new_nid', PNumber('offset')),
     )
 
     _answer = PStruct('accept_identification',
         PFNodeType,
         PUUID('my_uuid'),
-        PNumber('num_partitions'),
-        PNumber('num_replicas'),
         PUUID('your_uuid'),
     )
 
@@ -750,23 +748,24 @@ class LastIDs(Packet):
 class PartitionTable(Packet):
     """
     Ask storage node the remaining data needed by master to recover.
-    This is also how the clients get the full partition table on connection.
 
-    :nodes: M -> S; C -> M
+    :nodes: M -> S
     """
     _answer = PStruct('answer_partition_table',
         PPTID('ptid'),
+        PNumber('num_replicas'),
         PFRowList,
     )
 
 class NotifyPartitionTable(Packet):
     """
-    Send the full partition table to admin/storage nodes on connection.
+    Send the full partition table to admin/client/storage nodes on connection.
 
-    :nodes: M -> A, S
+    :nodes: M -> A, C, S
     """
     _fmt = PStruct('send_partition_table',
         PPTID('ptid'),
+        PNumber('num_replicas'),
         PFRowList,
     )
 
@@ -778,6 +777,7 @@ class PartitionChanges(Packet):
     """
     _fmt = PStruct('notify_partition_changes',
         PPTID('ptid'),
+        PNumber('num_replicas'),
         PList('cell_list',
             PStruct('cell',
                 PNumber('offset'),
@@ -1203,6 +1203,7 @@ class PartitionList(Packet):
 
     _answer = PStruct('answer_partition_list',
         PPTID('ptid'),
+        PNumber('num_replicas'),
         PFRowList,
     )
 
@@ -1254,10 +1255,14 @@ class TweakPartitionTable(Packet):
     :nodes: ctl -> A -> M
     """
     _fmt = PStruct('tweak_partition_table',
+        PBoolean('dry_run'),
         PFUUIDList,
     )
 
-    _answer = Error
+    _answer = PStruct('answer_tweak_partition_table',
+        PBoolean('changed'),
+        PFRowList,
+    )
 
 class NotifyNodeInformation(Packet):
     """
@@ -1270,6 +1275,18 @@ class NotifyNodeInformation(Packet):
         PFNodeList,
     )
 
+class SetNumReplicas(Packet):
+    """
+    Set the number of replicas.
+
+    :nodes: ctl -> A -> M
+    """
+    _fmt = PStruct('set_num_replicas',
+        PNumber('num_replicas'),
+    )
+
+    _answer = Error
+
 class SetClusterState(Packet):
     """
     Set the cluster state.
@@ -1763,8 +1780,10 @@ class Packets(dict):
                     SetNodeState, ignore_when_closed=False)
     AddPendingNodes = register(
                     AddPendingNodes, ignore_when_closed=False)
-    TweakPartitionTable = register(
-                    TweakPartitionTable, ignore_when_closed=False)
+    TweakPartitionTable, AnswerTweakPartitionTable = register(
+                    TweakPartitionTable)
+    SetNumReplicas = register(
+                    SetNumReplicas, ignore_when_closed=False)
     SetClusterState = register(
                     SetClusterState, ignore_when_closed=False)
     Repair = register(

neo/lib/pt.py

@@ -86,15 +86,9 @@ class PartitionTable(object):
         'a cell became non-readable whereas all cells were readable'
 
     def __init__(self, num_partitions, num_replicas):
-        self._id = None
         self.np = num_partitions
         self.nr = num_replicas
-        self.num_filled_rows = 0
-        # Note: don't use [[]] * num_partition construct, as it duplicates
-        # instance *references*, so the outer list contains really just one
-        # inner list instance.
-        self.partition_list = [[] for _ in xrange(num_partitions)]
-        self.count_dict = {}
+        self.clear()
 
     def getID(self):
         return self._id
@@ -113,7 +107,16 @@ class PartitionTable(object):
         # instance *references*, so the outer list contains really just one
         # inner list instance.
         self.partition_list = [[] for _ in xrange(self.np)]
-        self.count_dict.clear()
+        self.count_dict = {}
+
+    def addNodeList(self, node_list):
+        """Add nodes"""
+        added_list = []
+        for node in node_list:
+            if node not in self.count_dict:
+                self.count_dict[node] = 0
+                added_list.append(node)
+        return added_list
 
     def getAssignedPartitionList(self, uuid):
         """ Return the partition assigned to the specified UUID """
@@ -203,31 +206,31 @@ class PartitionTable(object):
             del self.count_dict[node]
         return not count
 
-    def load(self, ptid, row_list, nm):
+    def _load(self, ptid, num_replicas, row_list, getByUUID):
+        self.__init__(len(row_list), num_replicas)
+        self._id = ptid
+        for offset, row in enumerate(row_list):
+            for uuid, state in row:
+                node = getByUUID(uuid)
+                self._setCell(offset, node, state)
+
+    def load(self, ptid, num_replicas, row_list, nm):
         """
         Load the partition table with the specified PTID, discard all previous
         content.
         """
-        self.clear()
-        self._id = ptid
-        for offset, row in row_list:
-            if offset >= self.getPartitions():
-                raise IndexError
-            for uuid, state in row:
-                node = nm.getByUUID(uuid)
-                # the node must be known by the node manager
-                assert node is not None
-                self._setCell(offset, node, state)
+        self._load(ptid, num_replicas, row_list, nm.getByUUID)
         logging.debug('partition table loaded (ptid=%s)', ptid)
         self.log()
 
-    def update(self, ptid, cell_list, nm):
+    def update(self, ptid, num_replicas, cell_list, nm):
         """
         Update the partition with the cell list supplied. If a node
         is not known, it is created in the node manager and set as unavailable
         """
         assert self._id < ptid, (self._id, ptid)
         self._id = ptid
+        self.nr = num_replicas
         readable_list = []
         for row in self.partition_list:
             if not all(cell.isReadable() for cell in row):
@@ -310,14 +313,11 @@ class PartitionTable(object):
         return True
 
     def getRow(self, offset):
-        row = self.partition_list[offset]
-        if row is None:
-            return []
-        return [(cell.getUUID(), cell.getState()) for cell in row]
+        return [(cell.getUUID(), cell.getState())
+                for cell in self.partition_list[offset]]
 
     def getRowList(self):
-        getRow = self.getRow
-        return [(x, getRow(x)) for x in xrange(self.np)]
+        return map(self.getRow, xrange(self.np))
 
 class MTPartitionTable(PartitionTable):
     """ Thread-safe aware version of the partition table, override only methods

neo/master/app.py

@@ -16,6 +16,7 @@
 
 import sys
 from collections import defaultdict
+from functools import partial
 from time import time
 
 from neo.lib import logging, util
@@ -76,13 +77,11 @@ class Application(BaseApplication):
 
     @classmethod
     def _buildOptionParser(cls):
-        _ = cls.option_parser
-        _.description = "NEO Master node"
+        parser = cls.option_parser
+        parser.description = "NEO Master node"
         cls.addCommonServerOptions('master', '127.0.0.1:10000', '')
 
-        _ = _.group('master')
-        _.int('r', 'replicas', default=0, help="replicas number")
-        _.int('p', 'partitions', default=100, help="partitions number")
+        _ = parser.group('master')
         _.int('A', 'autostart',
             help="minimum number of pending storage nodes to automatically"
                  " start new cluster (to avoid unwanted recreation of the"
@@ -91,8 +90,12 @@ class Application(BaseApplication):
             help='the name of cluster to backup')
         _('M', 'upstream-masters', parse=util.parseMasterList,
             help='list of master nodes in the cluster to backup')
-        _.int('u', 'uuid',
-            help="specify an UUID to use for this process (testing purpose)")
+        _.int('i', 'nid',
+            help="specify an NID to use for this process (testing purpose)")
+
+        _ = parser.group('database creation')
+        _.int('r', 'replicas', default=0, help="replicas number")
+        _.int('p', 'partitions', default=100, help="partitions number")
 
     def __init__(self, config):
         super(Application, self).__init__(
@@ -108,7 +111,7 @@ class Application(BaseApplication):
         for master_address in config['masters']:
             self.nm.createMaster(address=master_address)
         self._node = self.nm.createMaster(address=self.server,
-                                          uuid=config.get('uuid'))
+                                          uuid=config.get('nid'))
         logging.node(self.name, self.uuid)
 
         logging.debug('IP address is %s, port is %d', *self.server)
@@ -117,14 +120,14 @@ class Application(BaseApplication):
         replicas = config['replicas']
         partitions = config['partitions']
         if replicas < 0:
-            raise RuntimeError, 'replicas must be a positive integer'
+            sys.exit('replicas must be a positive integer')
         if partitions <= 0:
-            raise RuntimeError, 'partitions must be more than zero'
-        self.pt = PartitionTable(partitions, replicas)
+            sys.exit('partitions must be more than zero')
         logging.info('Configuration:')
         logging.info('Partitions: %d', partitions)
         logging.info('Replicas  : %d', replicas)
         logging.info('Name      : %s', self.name)
+        self.newPartitionTable = partial(PartitionTable, partitions, replicas)
 
         self.listening_conn = None
         self.cluster_state = None
@@ -196,7 +199,7 @@ class Application(BaseApplication):
             node_dict[NodeTypes.MASTER].append(node_info)
         return node_dict
 
-    def broadcastNodesInformation(self, node_list, exclude=None):
+    def broadcastNodesInformation(self, node_list):
         """
           Broadcast changes for a set a nodes
           Send only one packet per connection to reduce bandwidth
@@ -209,20 +212,26 @@ class Application(BaseApplication):
             # We don't skip pending storage nodes because we don't send them
             # the full list of nodes when they're added, and it's also quite
             # useful to notify them about new masters.
-            if node_list and node is not exclude:
+            if node_list:
                 node.send(Packets.NotifyNodeInformation(now, node_list))
 
-    def broadcastPartitionChanges(self, cell_list):
+    def broadcastPartitionChanges(self, cell_list, num_replicas=None):
         """Broadcast a Notify Partition Changes packet."""
-        if cell_list:
-            ptid = self.pt.setNextID()
-            self.pt.logUpdated()
-            packet = Packets.NotifyPartitionChanges(ptid, cell_list)
-            for node in self.nm.getIdentifiedList():
-                # As for broadcastNodesInformation, we don't send the full PT
-                # when pending storage nodes are added, so keep them notified.
-                if not node.isMaster():
-                    node.send(packet)
+        pt = self.pt
+        if num_replicas is not None:
+            pt.setReplicas(num_replicas)
+        elif cell_list:
+            num_replicas = pt.getReplicas()
+        else:
+            return
+        packet = Packets.NotifyPartitionChanges(
+            pt.setNextID(), num_replicas, cell_list)
+        pt.logUpdated()
+        for node in self.nm.getIdentifiedList():
+            # As for broadcastNodesInformation, we don't send the full PT
+            # when pending storage nodes are added, so keep them notified.
+            if not node.isMaster():
+                node.send(packet)
 
     def provideService(self):
         """
@@ -437,16 +446,7 @@ class Application(BaseApplication):
                 conn.send(notification_packet)
             elif conn.isServer():
                 continue
-            if node.isClient():
-                if state == ClusterStates.RUNNING:
-                    handler = self.client_service_handler
-                elif state == ClusterStates.BACKINGUP:
-                    handler = self.client_ro_service_handler
-                else:
-                    if state != ClusterStates.STOPPING:
-                        conn.abort()
-                    continue
-            elif node.isMaster():
+            if node.isMaster():
                 if state == ClusterStates.RECOVERING:
                     handler = self.election_handler
                 else:
@@ -454,10 +454,16 @@ class Application(BaseApplication):
             elif node.isStorage() and storage_handler:
                 handler = storage_handler
             else:
-                continue # keep handler             # FIXME handler can be not setup-yet at all
+                # There's a single handler type for admins.
+                # Client can't change handler without being first disconnected.
+                assert state in (
+                    ClusterStates.STOPPING,
+                    ClusterStates.STOPPING_BACKUP,
+                    ) or not node.isClient(), (state, node)
+                continue # keep handler
             if type(handler) is not type(conn.getLastHandler()):
                 conn.setHandler(handler)
-                handler.connectionCompleted(conn, new=False)
+                handler.handlerSwitched(conn, new=False)
         self.cluster_state = state
 
     def getNewUUID(self, uuid, address, node_type):

neo/master/backup_app.py

@@ -112,17 +112,12 @@ class BackupApplication(object):
                     else:
                         break
                     poll(1)
-                node, conn, num_partitions, num_replicas = \
-                    bootstrap.getPrimaryConnection()
+                node, conn = bootstrap.getPrimaryConnection()
                 try:
                     app.changeClusterState(ClusterStates.BACKINGUP)
                     del bootstrap, node
-                    if num_partitions != pt.getPartitions():
-                        raise RuntimeError("inconsistent number of partitions")
                     self.ignore_invalidations = True
-                    self.pt = PartitionTable(num_partitions, num_replicas)
                     conn.setHandler(BackupHandler(self))
-                    conn.ask(Packets.AskPartitionTable())
                     conn.ask(Packets.AskLastTransaction())
                     # debug variable to log how big 'tid_list' can be.
                     self.debug_tid_count = 0

neo/master/handlers/__init__.py

@@ -23,10 +23,6 @@ from neo.lib.protocol import Packets
 class MasterHandler(EventHandler):
     """This class implements a generic part of the event handlers."""
 
-    def connectionCompleted(self, conn, new=None):
-        if new is None:
-            super(MasterHandler, self).connectionCompleted(conn)
-
     def connectionLost(self, conn, new_state=None):
         if self.app.listening_conn: # if running
             self._connectionLost(conn)
@@ -59,17 +55,20 @@ class MasterHandler(EventHandler):
             + app.getNodeInformationDict(node_list)[node.getType()])
         conn.send(Packets.NotifyNodeInformation(monotonic_time(), node_list))
 
-    def askPartitionTable(self, conn):
+    def handlerSwitched(self, conn, new):
         pt = self.app.pt
-        conn.answer(Packets.AnswerPartitionTable(pt.getID(), pt.getRowList()))
+        # Except storages during recovery and secondary masters, all nodes
+        # receives the full partition table as soon as they're identified.
+        # It is also sent in 2 other cases:
+        # - to admins during recovery, whenever a newer PT is loaded;
+        # - to storage when switching from recovery to verification.
+        # After that, non-master nodes only receive incremental updates.
+        conn.send(Packets.SendPartitionTable(
+            pt.getID(), pt.getReplicas(), pt.getRowList()))
 
 
 class BaseServiceHandler(MasterHandler):
-    """This class deals with events for a service phase."""
-
-    def connectionCompleted(self, conn, new):
-        pt = self.app.pt
-        conn.send(Packets.SendPartitionTable(pt.getID(), pt.getRowList()))
+    """Common handler class for storage nodes."""
 
     def connectionLost(self, conn, new_state):
         app = self.app

neo/master/handlers/administration.py

@@ -15,14 +15,16 @@
 # along with this program.  If not, see <http://www.gnu.org/licenses/>.
 
 import random
+from functools import wraps
 
 from . import MasterHandler
 from ..app import monotonic_time, StateChangedException
 from neo.lib import logging
 from neo.lib.exception import StoppedOperation
+from neo.lib.handler import AnswerDenied
 from neo.lib.pt import PartitionTableException
 from neo.lib.protocol import ClusterStates, Errors, \
-    NodeStates, NodeTypes, Packets, ProtocolError, uuid_str
+    NodeStates, NodeTypes, Packets, uuid_str
 from neo.lib.util import dump
 
 CLUSTER_STATE_WORKFLOW = {
@@ -38,9 +40,25 @@ NODE_STATE_WORKFLOW = {
     NodeTypes.STORAGE: (NodeStates.DOWN, NodeStates.UNKNOWN),
 }
 
+def check_state(*states):
+    def decorator(wrapped):
+        def wrapper(self, *args):
+            state = self.app.getClusterState()
+            if state not in states:
+                raise AnswerDenied('%s RPC can not be used in %s state'
+                                   % (wrapped.__name__, state))
+            wrapped(self, *args)
+        return wraps(wrapped)(wrapper)
+    return decorator
+
+
 class AdministrationHandler(MasterHandler):
     """This class deals with messages from the admin node only"""
 
+    def handlerSwitched(self, conn, new):
+        assert new
+        super(AdministrationHandler, self).handlerSwitched(conn, new)
+
     def connectionLost(self, conn, new_state):
         node = self.app.nm.getByUUID(conn.getUUID())
         if node is not None:
@@ -58,30 +76,28 @@ class AdministrationHandler(MasterHandler):
         # check request
         try:
             if app.cluster_state not in CLUSTER_STATE_WORKFLOW[state]:
-                raise ProtocolError('Can not switch to this state')
+                raise AnswerDenied('Can not switch to this state')
         except KeyError:
             if state != ClusterStates.STOPPING:
-                raise ProtocolError('Invalid state requested')
+                raise AnswerDenied('Invalid state requested')
 
         # change state
         if state == ClusterStates.VERIFYING:
             storage_list = app.nm.getStorageList(only_identified=True)
             if not storage_list:
-                raise ProtocolError('Cannot exit recovery without any '
-                    'storage node')
+                raise AnswerDenied(
+                    'Cannot exit recovery without any storage node')
             for node in storage_list:
                 assert node.isPending(), node
                 if node.getConnection().isPending():
-                    # XXX: It's wrong to use ProtocolError here. We must reply
-                    #      less aggressively because the admin has no way to
-                    #      know that there's still pending activity.
-                    raise ProtocolError('Cannot exit recovery now: node %r is '
-                        'entering cluster' % (node, ))
+                    raise AnswerDenied(
+                        'Cannot exit recovery now: node %r is entering cluster'
+                        % node,)
             app._startup_allowed = True
             state = app.cluster_state
         elif state == ClusterStates.STARTING_BACKUP:
             if app.tm.hasPending() or app.nm.getClientList(True):
-                raise ProtocolError("Can not switch to %s state with pending"
+                raise AnswerDenied("Can not switch to %s state with pending"
                     " transactions or connected clients" % state)
 
         conn.answer(Errors.Ack('Cluster state changed'))
@@ -93,21 +109,24 @@ class AdministrationHandler(MasterHandler):
         app = self.app
         node = app.nm.getByUUID(uuid)
         if node is None:
-            raise ProtocolError('unknown node')
+            raise AnswerDenied('unknown node')
         if state not in NODE_STATE_WORKFLOW.get(node.getType(), ()):
-            raise ProtocolError('can not switch node to this state')
+            raise AnswerDenied('can not switch node to %s state' % state)
         if uuid == app.uuid:
-            raise ProtocolError('can not kill primary master node')
+            raise AnswerDenied('can not kill primary master node')
 
         state_changed = state != node.getState()
         message = ('state changed' if state_changed else
                    'node already in %s state' % state)
         if node.isStorage():
             keep = state == NodeStates.DOWN
+            if node.isRunning() and not keep:
+                raise AnswerDenied(
+                    "a running node must be stopped before removal")
             try:
                 cell_list = app.pt.dropNodeList([node], keep)
             except PartitionTableException, e:
-                raise ProtocolError(str(e))
+                raise AnswerDenied(str(e))
             node.setState(state)
             if node.isConnected():
                 # notify itself so it can shutdown
@@ -134,16 +153,17 @@ class AdministrationHandler(MasterHandler):
                     monotonic_time(), [node.asTuple()]))
             app.broadcastNodesInformation([node])
 
+    # XXX: Would it be safe to allow more states ?
+    __change_pt_rpc = check_state(
+        ClusterStates.RUNNING,
+        ClusterStates.STARTING_BACKUP,
+        ClusterStates.BACKINGUP)
+
+    @__change_pt_rpc
     def addPendingNodes(self, conn, uuid_list):
         uuids = ', '.join(map(uuid_str, uuid_list))
         logging.debug('Add nodes %s', uuids)
         app = self.app
-        state = app.getClusterState()
-        # XXX: Would it be safe to allow more states ?
-        if state not in (ClusterStates.RUNNING,
-                         ClusterStates.STARTING_BACKUP,
-                         ClusterStates.BACKINGUP):
-            raise ProtocolError('Can not add nodes in %s state' % state)
         # take all pending nodes
         node_list = list(app.pt.addNodeList(node
             for node in app.nm.getStorageList()
@@ -165,31 +185,50 @@ class AdministrationHandler(MasterHandler):
         for uuid in uuid_list:
             node = getByUUID(uuid)
             if node is None or not (node.isStorage() and node.isIdentified()):
-                raise ProtocolError("invalid storage node %s" % uuid_str(uuid))
+                raise AnswerDenied("invalid storage node %s" % uuid_str(uuid))
             node_list.append(node)
         repair = Packets.NotifyRepair(*args)
         for node in node_list:
             node.send(repair)
         conn.answer(Errors.Ack(''))
 
-    def tweakPartitionTable(self, conn, uuid_list):
-        app = self.app
-        state = app.getClusterState()
-        # XXX: Would it be safe to allow more states ?
-        if state not in (ClusterStates.RUNNING,
-                         ClusterStates.STARTING_BACKUP,
-                         ClusterStates.BACKINGUP):
-            raise ProtocolError('Can not tweak partition table in %s state'
-                                % state)
-        app.broadcastPartitionChanges(app.pt.tweak([node
-            for node in app.nm.getStorageList()
-            if node.getUUID() in uuid_list or not node.isRunning()]))
+    @__change_pt_rpc
+    def setNumReplicas(self, conn, num_replicas):
+        self.app.broadcastPartitionChanges((), num_replicas)
         conn.answer(Errors.Ack(''))
 
-    def truncate(self, conn, tid):
+    @__change_pt_rpc
+    def tweakPartitionTable(self, conn, dry_run, uuid_list):
         app = self.app
-        if app.cluster_state != ClusterStates.RUNNING:
-            raise ProtocolError('Can not truncate in this state')
+        drop_list = []
+        for node in app.nm.getStorageList():
+            if node.getUUID() in uuid_list or node.isPending():
+                drop_list.append(node)
+            elif not node.isRunning():
+                drop_list.append(node)
+                raise AnswerDenied(
+                    'tweak: down nodes must be listed explicitly')
+        if dry_run:
+            pt = object.__new__(app.pt.__class__)
+            new_nodes = pt.load(app.pt.getID(), app.pt.getReplicas(),
+                                app.pt.getRowList(), app.nm)
+            assert not new_nodes
+            pt.addNodeList(node
+                for node, count in app.pt.count_dict.iteritems()
+                if not count)
+        else:
+            pt = app.pt
+        try:
+            changed_list = pt.tweak(drop_list)
+        except PartitionTableException, e:
+            raise AnswerDenied(str(e))
+        if not dry_run:
+            app.broadcastPartitionChanges(changed_list)
+        conn.answer(Packets.AnswerTweakPartitionTable(
+            bool(changed_list), pt.getRowList()))
+
+    @check_state(ClusterStates.RUNNING)
+    def truncate(self, conn, tid):
         conn.answer(Errors.Ack(''))
         raise StoppedOperation(tid)
 
@@ -237,3 +276,5 @@ class AdministrationHandler(MasterHandler):
                 node.send(Packets.CheckPartition(
                     offset, source, min_tid, max_tid))
         conn.answer(Errors.Ack(''))
+
+    del __change_pt_rpc

neo/master/handlers/backup.py

@@ -17,6 +17,7 @@
 from neo.lib.exception import PrimaryFailure
 from neo.lib.handler import EventHandler
 from neo.lib.protocol import ZERO_TID
+from neo.lib.pt import PartitionTable
 
 class BackupHandler(EventHandler):
     """Handler dedicated to upstream master during BACKINGUP state"""
@@ -25,12 +26,15 @@ class BackupHandler(EventHandler):
         if self.app.app.listening_conn: # if running
             raise PrimaryFailure('connection lost')
 
-    def answerPartitionTable(self, conn, ptid, row_list):
-        self.app.pt.load(ptid, row_list, self.app.nm)
+    def sendPartitionTable(self, conn, ptid, num_replicas, row_list):
+        app = self.app
+        pt = app.pt = object.__new__(PartitionTable)
+        pt.load(ptid, num_replicas, row_list, self.app.nm)
+        if pt.getPartitions() != app.app.pt.getPartitions():
+            raise RuntimeError("inconsistent number of partitions")
 
-    def notifyPartitionChanges(self, conn, ptid, cell_list):
-        if self.app.pt.filled():
-            self.app.pt.update(ptid, cell_list, self.app.nm)
+    def notifyPartitionChanges(self, conn, ptid, num_replicas, cell_list):
+        self.app.pt.update(ptid, num_replicas, cell_list, self.app.nm)
 
     # NOTE invalidation from M -> Mb (all partitions)
     def answerLastTransaction(self, conn, tid):

neo/master/handlers/client.py

@@ -22,6 +22,10 @@ from . import MasterHandler
 class ClientServiceHandler(MasterHandler):
     """ Handler dedicated to client during service state """
 
+    def handlerSwitched(self, conn, new):
+        assert new
+        super(ClientServiceHandler, self).handlerSwitched(conn, new)
+
     def _connectionLost(self, conn):
         # cancel its transactions and forgot the node
         app = self.app

neo/master/handlers/identification.py

@@ -17,14 +17,14 @@
 from neo.lib import logging
 from neo.lib.exception import PrimaryElected
 from neo.lib.handler import EventHandler
-from neo.lib.protocol import ClusterStates, NodeStates, NodeTypes, \
-    NotReadyError, Packets, ProtocolError, uuid_str
+from neo.lib.protocol import CellStates, ClusterStates, NodeStates, \
+    NodeTypes, NotReadyError, Packets, ProtocolError, uuid_str
 from ..app import monotonic_time
 
 class IdentificationHandler(EventHandler):
 
     def requestIdentification(self, conn, node_type, uuid,
-                              address, name, devpath, id_timestamp):
+                              address, name, id_timestamp, devpath, new_nid):
         app = self.app
         self.checkClusterName(name)
         if address == app.server:
@@ -77,6 +77,16 @@ class IdentificationHandler(EventHandler):
                 manager = app
             state, handler = manager.identifyStorageNode(
                 uuid is not None and node is not None)
+            if not address:
+                if app.cluster_state == ClusterStates.RECOVERING:
+                    raise NotReadyError
+                if uuid or not new_nid:
+                    raise ProtocolError
+                state = NodeStates.DOWN
+                # We'll let the storage node close the connection. If we
+                # aborted it at the end of the method, BootstrapManager
+                # (which is used by storage nodes) could see the closure
+                # and try to reconnect to a master.
             human_readable_node_type = ' storage (%s) ' % (state, )
         elif node_type == NodeTypes.MASTER:
             if app.election:
@@ -105,24 +115,27 @@ class IdentificationHandler(EventHandler):
             node.devpath = tuple(devpath)
         node.id_timestamp = monotonic_time()
         node.setState(state)
+        app.broadcastNodesInformation([node])
+        if new_nid:
+            changed_list = []
+            for offset in new_nid:
+                changed_list.append((offset, uuid, CellStates.OUT_OF_DATE))
+                app.pt._setCell(offset, node, CellStates.OUT_OF_DATE)
+            app.broadcastPartitionChanges(changed_list)
         conn.setHandler(handler)
         node.setConnection(conn, not node.isIdentified())
-        app.broadcastNodesInformation([node], node)
 
         conn.answer(Packets.AcceptIdentification(
             NodeTypes.MASTER,
             app.uuid,
-            app.pt.getPartitions(),
-            app.pt.getReplicas(),
             uuid))
         handler._notifyNodeInformation(conn)
-        handler.connectionCompleted(conn, True)
-
+        handler.handlerSwitched(conn, True)
 
 class SecondaryIdentificationHandler(EventHandler):
 
     def requestIdentification(self, conn, node_type, uuid,
-                              address, name, devpath, id_timestamp):
+                              address, name, id_timestamp, devpath, new_nid):
         app = self.app
         self.checkClusterName(name)
         if address == app.server:

neo/master/handlers/master.py

@@ -23,6 +23,9 @@ from neo.lib.protocol import ClusterStates, NodeStates, NodeTypes, Packets
 class SecondaryHandler(MasterHandler):
     """Handler used by primary to handle secondary masters"""
 
+    def handlerSwitched(self, conn, new):
+        pass
+
     def _connectionLost(self, conn):
         app = self.app
         node = app.nm.getByUUID(conn.getUUID())
@@ -30,21 +33,20 @@ class SecondaryHandler(MasterHandler):
         app.broadcastNodesInformation([node])
 
 
-class ElectionHandler(MasterHandler):
+class ElectionHandler(SecondaryHandler):
     """Handler used by primary to handle secondary masters during election"""
 
-    def connectionCompleted(self, conn, new=None):
-        if new is None:
-            super(ElectionHandler, self).connectionCompleted(conn)
-            app = self.app
-            conn.ask(Packets.RequestIdentification(NodeTypes.MASTER,
-                app.uuid, app.server, app.name, (), app.election))
+    def connectionCompleted(self, conn):
+        super(ElectionHandler, self).connectionCompleted(conn)
+        app = self.app
+        conn.ask(Packets.RequestIdentification(NodeTypes.MASTER,
+            app.uuid, app.server, app.name, app.election, (), ()))
 
     def connectionFailed(self, conn):
         super(ElectionHandler, self).connectionFailed(conn)
         self.connectionLost(conn)
 
-    def _acceptIdentification(self, node, *args):
+    def _acceptIdentification(self, node):
         raise PrimaryElected(node)
 
     def _connectionLost(self, *args):
@@ -66,7 +68,7 @@ class ElectionHandler(MasterHandler):
 class PrimaryHandler(ElectionHandler):
     """Handler used by secondaries to handle primary master"""
 
-    def _acceptIdentification(self, node, num_partitions, num_replicas):
+    def _acceptIdentification(self, node):
         assert self.app.primary_master is node, (self.app.primary_master, node)
 
     def _connectionLost(self, conn):

neo/master/handlers/storage.py

@@ -26,10 +26,10 @@ from . import BaseServiceHandler
 class StorageServiceHandler(BaseServiceHandler):
     """ Handler dedicated to storages during service state """
 
-    def connectionCompleted(self, conn, new):
+    def handlerSwitched(self, conn, new):
         app = self.app
         if new:
-            super(StorageServiceHandler, self).connectionCompleted(conn, new)
+            super(StorageServiceHandler, self).handlerSwitched(conn, new)
         node = app.nm.getByUUID(conn.getUUID())
         if node.isRunning(): # node may be PENDING
             app.startStorage(node)

neo/master/pt.py

@@ -56,6 +56,10 @@ class PartitionTable(neo.lib.pt.PartitionTable):
         self._id += 1
         return self._id
 
+    def setReplicas(self, num_replicas):
+        assert num_replicas >= 0, num_replicas
+        self.nr = num_replicas
+
     def make(self, node_list):
         """Make a new partition table from scratch."""
         assert self._id is None and node_list, (self._id, node_list)
@@ -108,26 +112,19 @@ class PartitionTable(neo.lib.pt.PartitionTable):
             self.num_filled_rows = len(filter(None, self.partition_list))
         return change_list
 
-    def load(self, ptid, row_list, nm):
+    def load(self, ptid, num_replicas, row_list, nm):
         """
         Load a partition table from a storage node during the recovery.
         Return the new storage nodes registered
         """
-        # check offsets
-        for offset, _row in row_list:
-            if offset >= self.getPartitions():
-                raise IndexError, offset
-        # store the partition table
-        self.clear()
-        self._id = ptid
         new_nodes = []
-        for offset, row in row_list:
-            for uuid, state in row:
-                node = nm.getByUUID(uuid)
-                if node is None:
-                    node = nm.createStorage(uuid=uuid)
-                    new_nodes.append(node.asTuple())
-                self._setCell(offset, node, state)
+        def getByUUID(nid):
+            node = nm.getByUUID(nid)
+            if node is None:
+                node = nm.createStorage(uuid=nid)
+                new_nodes.append(node.asTuple())
+            return node
+        self._load(ptid, num_replicas, row_list, getByUUID)
         return new_nodes
 
     def setUpToDate(self, node, offset):
@@ -166,15 +163,6 @@ class PartitionTable(neo.lib.pt.PartitionTable):
 
         return cell_list
 
-    def addNodeList(self, node_list):
-        """Add nodes"""
-        added_list = []
-        for node in node_list:
-            if node not in self.count_dict:
-                self.count_dict[node] = 0
-                added_list.append(node)
-        return added_list
-
     def tweak(self, drop_list=()):
         """Optimize partition table
 
@@ -183,7 +171,8 @@ class PartitionTable(neo.lib.pt.PartitionTable):
           few readable cells, some cells are instead marked as FEEDING. This is
           a preliminary step to drop these nodes, otherwise the partition table
           could become non-operational.
-        - Other nodes must have the same number of cells, off by 1.
+          In fact, the code touching these cells is disabled (see NOTE below).
+        - Other nodes must have the same number of non-feeding cells, off by 1.
         - When a transaction creates new objects (oids are roughly allocated
           sequentially), we expect better performance by maximizing the number
           of involved nodes (i.e. parallelizing writes).
@@ -232,6 +221,8 @@ class PartitionTable(neo.lib.pt.PartitionTable):
         # Collect some data in a usable form for the rest of the method.
         node_list = {node: {} for node in self.count_dict
                               if node not in drop_list}
+        if not node_list:
+            raise neo.lib.pt.PartitionTableException("Can't remove all nodes.")
         drop_list = defaultdict(list)
         for offset, row in enumerate(self.partition_list):
             for cell in row:
@@ -420,6 +411,22 @@ class PartitionTable(neo.lib.pt.PartitionTable):
                     outdated_list[offset] -= 1
             for offset, cell in cell_dict.iteritems():
                 discard_list[offset].append(cell)
+        # NOTE: The following line disables the next 2 lines, which actually
+        #       causes cells in drop_list to be discarded, now or later;
+        #       drop_list could be renamed into ignore_list.
+        #       1. Deleting data partition per partition is a lot of work, so
+        #          why ask nodes in drop_list to do that when the goal is
+        #          simply to trash the whole underlying database?
+        #       2. By excluding nodes from a tweak, it becomes possible to have
+        #          parts of the partition table that are tweaked differently.
+        #          This may require to temporarily change the number of
+        #          replicas for the part being tweaked. In the future, this
+        #          number may be specified in the 'tweak' command, to avoid
+        #          race conditions with setUpToDate().
+        #       Overall, a common use case is when importing a ZODB to NEO,
+        #       to keep the initial importing node up until the database is
+        #       split and replicated to the final nodes.
+        drop_list = {}
         for offset, drop_list in drop_list.iteritems():
             discard_list[offset] += drop_list
         # We have sorted cells to discard in order to first deallocate nodes

neo/master/recovery.py

@@ -28,7 +28,7 @@ class RecoveryManager(MasterHandler):
 
     def __init__(self, app):
         # The target node's uuid to request next.
-        self.target_ptid = None
+        self.target_ptid = 0
         self.ask_pt = []
         self.backup_tid_dict = {}
         self.truncate_dict = {}
@@ -52,9 +52,8 @@ class RecoveryManager(MasterHandler):
         """
         logging.info('begin the recovery of the status')
         app = self.app
-        pt = app.pt
+        pt = app.pt = app.newPartitionTable()
         app.changeClusterState(ClusterStates.RECOVERING)
-        pt.clear()
 
         self.try_secondary = True
 
@@ -113,7 +112,7 @@ class RecoveryManager(MasterHandler):
                             for node in node_list:
                                 conn = node.getConnection()
                                 conn.send(truncate)
-                                self.connectionCompleted(conn, False)
+                                self.handlerSwitched(conn, False)
                             continue
                     node_list = pt.getConnectedNodeList()
                 break
@@ -140,12 +139,12 @@ class RecoveryManager(MasterHandler):
             logging.info('creating a new partition table')
             pt.make(node_list)
             self._notifyAdmins(Packets.SendPartitionTable(
-                pt.getID(), pt.getRowList()))
+                pt.getID(), pt.getReplicas(), pt.getRowList()))
         else:
             cell_list = pt.outdate()
             if cell_list:
                 self._notifyAdmins(Packets.NotifyPartitionChanges(
-                    pt.setNextID(), cell_list))
+                    pt.setNextID(), pt.getReplicas(), cell_list))
             if app.backup_tid:
                 pt.setBackupTidDict(self.backup_tid_dict)
                 app.backup_tid = pt.getBackupTid()
@@ -175,16 +174,16 @@ class RecoveryManager(MasterHandler):
         if node is None or node.getState() == new_state:
             return
         node.setState(new_state)
-        # broadcast to all so that admin nodes gets informed
         self.app.broadcastNodesInformation([node])
 
-    def connectionCompleted(self, conn, new):
+    def handlerSwitched(self, conn, new):
         # ask the last IDs to perform the recovery
         conn.ask(Packets.AskRecovery())
 
     def answerRecovery(self, conn, ptid, backup_tid, truncate_tid):
         uuid = conn.getUUID()
-        if self.target_ptid <= ptid:
+        # ptid is None if the node has an empty partition table.
+        if ptid and self.target_ptid <= ptid:
             # Maybe a newer partition table.
             if self.target_ptid == ptid and self.ask_pt:
                 # Another node is already asked.
@@ -197,17 +196,14 @@ class RecoveryManager(MasterHandler):
         self.backup_tid_dict[uuid] = backup_tid
         self.truncate_dict[uuid] = truncate_tid
 
-    def answerPartitionTable(self, conn, ptid, row_list):
+    def answerPartitionTable(self, conn, ptid, num_replicas, row_list):
         # If this is not from a target node, ignore it.
         if ptid == self.target_ptid:
             app = self.app
-            try:
-                new_nodes = app.pt.load(ptid, row_list, app.nm)
-            except IndexError:
-                raise ProtocolError('Invalid offset')
+            new_nodes = app.pt.load(ptid, num_replicas, row_list, app.nm)
             self._notifyAdmins(
                 Packets.NotifyNodeInformation(monotonic_time(), new_nodes),
-                Packets.SendPartitionTable(ptid, row_list))
+                Packets.SendPartitionTable(ptid, num_replicas, row_list))
             self.ask_pt = ()
             uuid = conn.getUUID()
             app.backup_tid = self.backup_tid_dict[uuid]

neo/neoctl/app.py

@@ -16,9 +16,11 @@
 
 import sys
 from .neoctl import NeoCTL, NotReadyException
+from neo.lib.node import NodeManager
+from neo.lib.pt import PartitionTable
 from neo.lib.util import p64, u64, tidFromTime, timeStringFromTID
 from neo.lib.protocol import uuid_str, formatNodeList, \
-    ClusterStates, NodeTypes, UUID_NAMESPACES, ZERO_TID
+    ClusterStates, NodeStates, NodeTypes, UUID_NAMESPACES, ZERO_TID
 
 action_dict = {
     'print': {
@@ -30,6 +32,7 @@ action_dict = {
     },
     'set': {
         'cluster': 'setClusterState',
+        'replicas': 'setNumReplicas',
     },
     'check': 'checkReplicas',
     'start': 'startCluster',
@@ -47,6 +50,11 @@ uuid_int = (lambda ns: lambda uuid:
     (ns[uuid[0]] << 24) + int(uuid[1:])
     )({str(k)[0]: v for k, v in UUID_NAMESPACES.iteritems()})
 
+
+class dummy_app:
+    id_timestamp = uuid = 0
+
+
 class TerminalNeoCTL(object):
     def __init__(self, *args, **kw):
         self.neoctl = NeoCTL(*args, **kw)
@@ -68,6 +76,15 @@ class TerminalNeoCTL(object):
 
     asNode = staticmethod(uuid_int)
 
+    def formatPartitionTable(self, row_list):
+        nm = NodeManager()
+        nm.update(dummy_app, 1,
+            self.neoctl.getNodeList(node_type=NodeTypes.STORAGE))
+        pt = object.__new__(PartitionTable)
+        pt._load(None, None, row_list, nm.getByUUID)
+        pt.addNodeList(nm.getByStateList(NodeStates.RUNNING))
+        return '\n'.join(line[4:] for line in pt._format())
+
     def formatRowList(self, row_list):
         return '\n'.join('%03d |%s' % (offset,
             ''.join(' %s - %s |' % (uuid_str(uuid), state)
@@ -106,10 +123,12 @@ class TerminalNeoCTL(object):
         max_offset = int(max_offset)
         if node is not None:
             node = self.asNode(node)
-        ptid, row_list = self.neoctl.getPartitionRowList(
+        ptid, num_replicas, row_list = self.neoctl.getPartitionRowList(
                 min_offset=min_offset, max_offset=max_offset, node=node)
-        # TODO: return ptid
-        return self.formatRowList(row_list)
+        return '# ptid: %s, replicas: %s\n%s' % (ptid, num_replicas,
+            self.formatRowList(enumerate(row_list, min_offset))
+            if min_offset or max_offset else
+            self.formatPartitionTable(row_list))
 
     def getNodeList(self, params):
         """
@@ -141,6 +160,18 @@ class TerminalNeoCTL(object):
         assert len(params) == 1
         return self.neoctl.setClusterState(self.asClusterState(params[0]))
 
+    def setNumReplicas(self, params):
+        """
+          Set number of replicas.
+          Parameters: nr
+            nr: positive number (0 means no redundancy)
+        """
+        assert len(params) == 1
+        nr = int(params[0])
+        if nr < 0:
+            sys.exit('invalid number of replicas')
+        return self.neoctl.setNumReplicas(nr)
+
     def startCluster(self, params):
         """
           Starts cluster operation after a startup.
@@ -168,10 +199,18 @@ class TerminalNeoCTL(object):
     def tweakPartitionTable(self, params):
         """
           Optimize partition table.
-          No partition will be assigned to specified storage nodes.
-          Parameters: [node [...]]
+          No change is done to the specified/down storage nodes and they don't
+          count as replicas. The purpose of listing nodes is usually to drop
+          them once the data is replicated to other nodes.
+          Parameters: [-n] [node [...]]
+            -n: dry run
         """
-        return self.neoctl.tweakPartitionTable(map(self.asNode, params))
+        dry_run = params[0] == '-n'
+        changed, row_list = self.neoctl.tweakPartitionTable(
+            map(self.asNode, params[dry_run:]), dry_run)
+        if changed:
+            return self.formatPartitionTable(row_list)
+        return 'No change done.'
 
     def killNode(self, params):
         """

neo/neoctl/handler.py

@@ -14,6 +14,7 @@
 # You should have received a copy of the GNU General Public License
 # along with this program.  If not, see <http://www.gnu.org/licenses/>.
 
+import sys
 from neo.lib.handler import EventHandler
 from neo.lib.protocol import ErrorCodes, Packets
 
@@ -44,8 +45,8 @@ class CommandEventHandler(EventHandler):
     def ack(self, conn, msg):
         self.__respond((Packets.Error, ErrorCodes.ACK, msg))
 
-    def protocolError(self, conn, msg):
-        self.__respond((Packets.Error, ErrorCodes.PROTOCOL_ERROR, msg))
+    def denied(self, conn, msg):
+        sys.exit(msg)
 
     def notReady(self, conn, msg):
         self.__respond((Packets.Error, ErrorCodes.NOT_READY, msg))
@@ -62,3 +63,4 @@ class CommandEventHandler(EventHandler):
     answerLastIDs = __answer(Packets.AnswerLastIDs)
     answerLastTransaction = __answer(Packets.AnswerLastTransaction)
     answerRecovery = __answer(Packets.AnswerRecovery)
+    answerTweakPartitionTable = __answer(Packets.AnswerTweakPartitionTable)

neo/neoctl/neoctl.py

@@ -91,8 +91,14 @@ class NeoCTL(BaseApplication):
             raise RuntimeError(response)
         return response[2]
 
-    def tweakPartitionTable(self, uuid_list=()):
-        response = self.__ask(Packets.TweakPartitionTable(uuid_list))
+    def tweakPartitionTable(self, uuid_list=(), dry_run=False):
+        response = self.__ask(Packets.TweakPartitionTable(dry_run, uuid_list))
+        if response[0] != Packets.AnswerTweakPartitionTable:
+            raise RuntimeError(response)
+        return response[1:]
+
+    def setNumReplicas(self, nr):
+        response = self.__ask(Packets.SetNumReplicas(nr))
         if response[0] != Packets.Error or response[1] != ErrorCodes.ACK:
             raise RuntimeError(response)
         return response[2]
@@ -163,7 +169,7 @@ class NeoCTL(BaseApplication):
         response = self.__ask(packet)
         if response[0] != Packets.AnswerPartitionList:
             raise RuntimeError(response)
-        return response[1:3] # ptid, row_list
+        return response[1:]
 
     def startCluster(self):
         """

neo/scripts/runner.py

@@ -51,13 +51,11 @@ UNIT_TEST_MODULES = [
     'neo.tests.master.testClientHandler',
     'neo.tests.master.testMasterApp',
     'neo.tests.master.testMasterPT',
-    'neo.tests.master.testRecovery',
     'neo.tests.master.testStorageHandler',
     'neo.tests.master.testTransactions',
     # storage application
     'neo.tests.storage.testClientHandler',
     'neo.tests.storage.testMasterHandler',
-    'neo.tests.storage.testStorageApp',
     'neo.tests.storage.testStorage' + os.getenv('NEO_TESTS_ADAPTER', 'SQLite'),
     'neo.tests.storage.testTransactions',
     # client application

neo/storage/app.py

@@ -63,11 +63,16 @@ class Application(BaseApplication):
             help="do not delete data of discarded cells, which is useful for"
                  " big databases because the current implementation is"
                  " inefficient (this option should disappear in the future)")
+        _.bool('new-nid',
+            help="request a new NID from a cluster that is already"
+                 " operational, update the database with the new NID and exit,"
+                 " which makes easier to quickly set up a replica by copying"
+                 " the database of another node while it was stopped")
 
         _ = parser.group('database creation')
-        _.int('u', 'uuid',
-            help="specify an UUID to use for this process. Previously"
-                 " assigned UUID takes precedence (i.e. you should"
+        _.int('i', 'nid',
+            help="specify an NID to use for this process. Previously"
+                 " assigned NID takes precedence (i.e. you should"
                  " always use reset with this switch)")
         _('e', 'engine', help="database engine (MySQL only)")
         _.bool('dedup',
@@ -118,10 +123,16 @@ class Application(BaseApplication):
         self.loadConfiguration()
         self.devpath = self.dm.getTopologyPath()
 
-        # force node uuid from command line argument, for testing purpose only
-        if 'uuid' in config:
-            self.uuid = config['uuid']
-            logging.node(self.name, self.uuid)
+        if config.get('new_nid'):
+            self.new_nid = [x[0] for x in self.dm.iterAssignedCells()]
+            if not self.new_nid:
+                sys.exit('database is empty')
+            self.uuid = None
+        else:
+            self.new_nid = ()
+            if 'nid' in config: # for testing purpose only
+                self.uuid = config['nid']
+                logging.node(self.name, self.uuid)
 
         registerLiveDebugger(on_log=self.log)
 
@@ -158,36 +169,27 @@ class Application(BaseApplication):
         # load configuration
         self.uuid = dm.getUUID()
         logging.node(self.name, self.uuid)
-        num_partitions = dm.getNumPartitions()
-        num_replicas = dm.getNumReplicas()
-        ptid = dm.getPTID()
-
-        # check partition table configuration
-        if num_partitions is not None and num_replicas is not None:
-            if num_partitions <= 0:
-                raise RuntimeError, 'partitions must be more than zero'
-            # create a partition table
-            self.pt = PartitionTable(num_partitions, num_replicas)
 
         logging.info('Configuration loaded:')
-        logging.info('PTID      : %s', dump(ptid))
+        logging.info('PTID      : %s', dump(dm.getPTID()))
         logging.info('Name      : %s', self.name)
-        logging.info('Partitions: %s', num_partitions)
-        logging.info('Replicas  : %s', num_replicas)
 
     def loadPartitionTable(self):
         """Load a partition table from the database."""
-        self.pt.clear()
         ptid = self.dm.getPTID()
         if ptid is None:
+            self.pt = PartitionTable(0, 0)
             return
-        cell_list = []
+        row_list = []
         for offset, uuid, state in self.dm.getPartitionTable():
+            while len(row_list) <= offset:
+                row_list.append([])
             # register unknown nodes
             if self.nm.getByUUID(uuid) is None:
                 self.nm.createStorage(uuid=uuid)
-            cell_list.append((offset, uuid, CellStates[state]))
-        self.pt.update(ptid, cell_list, self.nm)
+            row_list[offset].append((uuid, CellStates[state]))
+        self.pt = object.__new__(PartitionTable)
+        self.pt.load(ptid, self.dm.getNumReplicas(), row_list, self.nm)
 
     def run(self):
         try:
@@ -247,29 +249,16 @@ class Application(BaseApplication):
 
         Note that I do not accept any connection from non-master nodes
         at this stage."""
-        pt = self.pt
-
         # search, find, connect and identify to the primary master
-        bootstrap = BootstrapManager(self, NodeTypes.STORAGE, self.server,
-                                     self.devpath)
-        self.master_node, self.master_conn, num_partitions, num_replicas = \
-            bootstrap.getPrimaryConnection()
+        bootstrap = BootstrapManager(self, NodeTypes.STORAGE,
+                                     None if self.new_nid else self.server,
+                                     self.devpath, self.new_nid)
+        self.master_node, self.master_conn = bootstrap.getPrimaryConnection()
         self.dm.setUUID(self.uuid)
 
-        # Reload a partition table from the database. This is necessary
-        # when a previous primary master died while sending a partition
-        # table, because the table might be incomplete.
-        if pt is not None:
-            self.loadPartitionTable()
-            if num_partitions != pt.getPartitions():
-                raise RuntimeError('the number of partitions is inconsistent')
-
-        if pt is None or pt.getReplicas() != num_replicas:
-            # changing number of replicas is not an issue
-            self.dm.setNumPartitions(num_partitions)
-            self.dm.setNumReplicas(num_replicas)
-            self.pt = PartitionTable(num_partitions, num_replicas)
-            self.loadPartitionTable()
+        # Reload a partition table from the database,
+        # in case that we're in RECOVERING phase.
+        self.loadPartitionTable()
 
     def initialize(self):
         logging.debug('initializing...')

neo/storage/checker.py

@@ -51,7 +51,7 @@ class Checker(object):
             else:
                 conn = ClientConnection(app, StorageOperationHandler(app), node)
                 conn.ask(Packets.RequestIdentification(NodeTypes.STORAGE,
-                    uuid, app.server, name, (), app.id_timestamp))
+                    uuid, app.server, name, app.id_timestamp, (), ()))
             self.conn_dict[conn] = node.isIdentified()
         conn_set = set(self.conn_dict)
         conn_set.discard(None)

neo/storage/database/importer.py

@@ -216,7 +216,7 @@ class ZODB(object):
         self._connect = _connect
         config = section.config
         if 'read_only' in config.getSectionAttributes():
-            has_next_oid = config.read_only = hasattr(self, 'next_oid')
+            has_next_oid = config.read_only = 'next_oid' in self.__dict__
             if not has_next_oid:
                 import gc
                 # This will reopen read-only as soon as we know the last oid.
@@ -378,8 +378,8 @@ class ImporterDatabaseManager(DatabaseManager):
         conf = self._conf
         db = self.db = buildDatabaseManager(conf['adapter'],
             (conf['database'], conf.get('engine'), conf['wait']))
-        for x in """getConfiguration _setConfiguration setNumPartitions
-                    query erase getPartitionTable _iterAssignedCells
+        for x in """getConfiguration _setConfiguration _getMaxPartition
+                    query erase getPartitionTable iterAssignedCells
                     updateCellTID getUnfinishedTIDDict dropUnfinishedData
                     abortTransaction storeTransaction lockTransaction
                     loadData storeData getOrphanList _pruneData deferCommit
@@ -396,9 +396,16 @@ class ImporterDatabaseManager(DatabaseManager):
                 self._writeback.committed()
         self.commit = db.commit = commit
 
-    def _updateReadable(self):
+    def _updateReadable(*_):
         raise AssertionError
 
+    def setUUID(self, nid):
+        old_nid = self.getUUID()
+        if old_nid:
+            assert old_nid == nid, (old_nid, nid)
+        else:
+            self.setConfiguration('nid', str(nid))
+
     def changePartitionTable(self, *args, **kw):
         self.db.changePartitionTable(*args, **kw)
         if self._writeback:
@@ -413,7 +420,7 @@ class ImporterDatabaseManager(DatabaseManager):
         if self._writeback:
             self._writeback.close()
         self.db.close()
-        if isinstance(self.zodb, list): # _setup called
+        if isinstance(self.zodb, tuple): # _setup called
             for zodb in self.zodb:
                 zodb.close()
 
@@ -436,9 +443,13 @@ class ImporterDatabaseManager(DatabaseManager):
         self.zodb_ltid = max(x.ltid for x in self.zodb)
         zodb = self.zodb[-1]
         self.zodb_loid = zodb.shift_oid + zodb.next_oid - 1
-        self.zodb_tid = self.db.getLastTID(self.zodb_ltid) or 0
-        if callable(self._import):
-            self._import = self._import()
+        self.zodb_tid = self._getMaxPartition() is not None and \
+            self.db.getLastTID(self.zodb_ltid) or 0
+        if callable(self._import): # XXX: why ?
+            if self.zodb_tid == self.zodb_ltid:
+                self._finished()
+            else:
+                self._import = self._import()
 
     def doOperation(self, app):
         if self._import:
@@ -498,12 +509,19 @@ class ImporterDatabaseManager(DatabaseManager):
         if process:
             process.join()
         self.commit()
+        self._finished()
+
+    def _finished(self):
         logging.warning("All data are imported. You should change"
             " your configuration to use the native backend and restart.")
         self._import = None
         for x in """getObject getReplicationTIDList getReplicationObjectList
+                    _fetchObject
                  """.split():
             setattr(self, x, getattr(self.db, x))
+        for zodb in self.zodb:
+            zodb.close()
+        self.zodb = None
 
     def _iter_zodb(self, zodb_list):
         util.setproctitle('neostorage: import')
@@ -667,6 +685,9 @@ class ImporterDatabaseManager(DatabaseManager):
                                                    length, partition)
         return r
 
+    def _fetchObject(*_):
+        raise AssertionError
+
     def getObjectHistory(self, *args, **kw):
         raise BackendNotImplemented(self.getObjectHistory)
 
@@ -678,6 +699,7 @@ class WriteBack(object):
 
     _changed = False
     _process = None
+    chunk_size = 100
 
     def __init__(self, db, storage):
         self._db = db
@@ -705,7 +727,7 @@ class WriteBack(object):
                 self._event = Event()
                 self._idle = Event()
                 self._stop = Event()
-                self._np = self._db.getNumPartitions()
+                self._np = 1 + self._db._getMaxPartition()
                 self._db = cPickle.dumps(self._db, 2)
                 self._process = Process(target=self._run)
                 self._process.daemon = True
@@ -737,7 +759,6 @@ class WriteBack(object):
     def iterator(self):
         db = self._db
         np = self._np
-        chunk_size = max(2, 1000 // np)
         offset_list = xrange(np)
         while 1:
             with db:
@@ -748,23 +769,26 @@ class WriteBack(object):
                 if np == len(db._readable_set):
                     while 1:
                         tid_list = []
-                        loop = False
+                        max_tid = MAX_TID
                         for offset in offset_list:
                             x = db.getReplicationTIDList(
-                                self.min_tid, MAX_TID, chunk_size, offset)
+                                self.min_tid, max_tid, self.chunk_size, offset)
                             tid_list += x
-                            if len(x) == chunk_size:
-                                loop = True
-                        if tid_list:
-                            tid_list.sort()
-                            for tid in tid_list:
-                                if self._stop.is_set():
-                                    return
-                                yield TransactionRecord(db, tid)
+                            if len(x) == self.chunk_size:
+                                max_tid = x[-1]
+                        if not tid_list:
+                            break
+                        tid_list.sort()
+                        for tid in tid_list:
+                            if self._stop.is_set():
+                                return
+                            yield TransactionRecord(db, tid)
+                            if tid == max_tid:
+                                break
+                        else:
                             self.min_tid = util.add64(tid, 1)
-                            if loop:
-                                continue
-                        break
+                            break
+                        self.min_tid = util.add64(tid, 1)
             if not self._event.is_set():
                 self._idle.set()
                 self._event.wait()

neo/storage/database/manager.py

@@ -102,25 +102,24 @@ class DatabaseManager(object):
         finally:
             db.close()
 
+    _cached_attr_list = (
+        '_readable_set', '_getPartition', '_getReadablePartition')
+
     def __getattr__(self, attr):
-        if attr in ('_readable_set', '_getPartition', '_getReadablePartition'):
+        if attr in self._cached_attr_list:
             self._updateReadable()
         return self.__getattribute__(attr)
 
-    def _partitionTableChanged(self):
-        try:
-            del (self._readable_set,
-                 self._getPartition,
-                 self._getReadablePartition)
-        except AttributeError:
-            pass
-
     def __enter__(self):
         assert not self.LOCK, "not a secondary connection"
         # XXX: All config caching should be done in this class,
         #      rather than in backend classes.
         self._config.clear()
-        self._partitionTableChanged()
+        try:
+            for attr in self._cached_attr_list:
+                delattr(self, attr)
+        except AttributeError:
+            pass
 
     def __exit__(self, t, v, tb):
         if v is None:
@@ -180,6 +179,10 @@ class DatabaseManager(object):
     def erase(self):
         """"""
 
+    def restore(self, dump): # for tests
+        self.erase()
+        self._restore(dump)
+
     def _setup(self, dedup=False):
         """To be overridden by the backend to set up a database
 
@@ -271,6 +274,18 @@ class DatabaseManager(object):
     def _setConfiguration(self, key, value):
         """"""
 
+    def _changePartitionTable(self, cell_list, reset=False):
+        """Change a part of a partition table. The list of cells is
+        a tuple of tuples, each of which consists of an offset (row ID),
+        the NID of a storage node, and a cell state. If reset is True,
+        existing data is first thrown away.
+        """
+
+    def _getPartitionTable(self):
+        """Return a whole partition table as a sequence of rows. Each row
+        is again a tuple of an offset (row ID), the NID of a storage
+        node, and a cell state."""
+
     def getUUID(self):
         """
             Load a NID from a database.
@@ -279,26 +294,19 @@ class DatabaseManager(object):
         if nid is not None:
             return int(nid)
 
+    @requires(_changePartitionTable, _getPartitionTable)
     def setUUID(self, nid):
         """
             Store a NID into a database.
         """
-        self.setConfiguration('nid', str(nid))
-
-    def getNumPartitions(self):
-        """
-            Load the number of partitions from a database.
-        """
-        n = self.getConfiguration('partitions')
-        if n is not None:
-            return int(n)
-
-    def setNumPartitions(self, num_partitions):
-        """
-            Store the number of partitions into a database.
-        """
-        self.setConfiguration('partitions', num_partitions)
-        self._partitionTableChanged()
+        old_nid = self.getUUID()
+        if nid != old_nid:
+            if old_nid:
+                self._changePartitionTable((offset, x, tid)
+                    for offset, x, tid in self._getPartitionTable()
+                    if x == old_nid
+                    for x, tid in ((x, None), (nid, tid)))
+            self.setConfiguration('nid', str(nid))
 
     def getNumReplicas(self):
         """
@@ -308,12 +316,6 @@ class DatabaseManager(object):
         if n is not None:
             return int(n)
 
-    def setNumReplicas(self, num_replicas):
-        """
-            Store the number of replicas into a database.
-        """
-        self.setConfiguration('replicas', num_replicas)
-
     def getName(self):
         """
             Load a name from a database.
@@ -374,8 +376,9 @@ class DatabaseManager(object):
 
         tids are in unpacked format.
         """
-        if self.getNumPartitions():
-            return max(map(self._getLastTID, self._readable_set))
+        x = self._readable_set
+        if x:
+            return max(self._getLastTID(x, max_tid) for x in x)
 
     def _getLastIDs(self, partition):
         """Return max(tid) & max(oid) for objects of given partition
@@ -395,7 +398,7 @@ class DatabaseManager(object):
         x = self._readable_set
         if x:
             tid, oid = zip(*map(self._getLastIDs, x))
-            tid = max(self.getLastTID(None), max(tid))
+            tid = max(self.getLastTID(), max(tid))
             oid = max(oid)
             return (None if tid is None else util.p64(tid),
                     None if oid is None else util.p64(oid))
@@ -511,13 +514,8 @@ class DatabaseManager(object):
             return (util.p64(serial), compression, checksum, data,
                 None if data_serial is None else util.p64(data_serial))
 
-    def _getPartitionTable(self):
-        """Return a whole partition table as a sequence of rows. Each row
-        is again a tuple of an offset (row ID), the NID of a storage
-        node, and a cell state."""
-
     @requires(_getPartitionTable)
-    def _iterAssignedCells(self):
+    def iterAssignedCells(self):
         my_nid = self.getUUID()
         return ((offset, tid) for offset, nid, tid in self._getPartitionTable()
                               if my_nid == nid)
@@ -537,24 +535,19 @@ class DatabaseManager(object):
         finally:
             readable_set.remove(offset)
 
-    def _changePartitionTable(self, cell_list, reset=False):
-        """Change a part of a partition table. The list of cells is
-        a tuple of tuples, each of which consists of an offset (row ID),
-        the NID of a storage node, and a cell state. If reset is True,
-        existing data is first thrown away.
+    def _getDataLastId(self, partition):
+        """
         """
 
-    def _getDataLastId(self, partition):
+    def _getMaxPartition(self):
         """
         """
 
-    @requires(_getDataLastId)
-    def _updateReadable(self):
-        try:
-            readable_set = self.__dict__['_readable_set']
-        except KeyError:
+    @requires(_getDataLastId, _getMaxPartition)
+    def _updateReadable(self, reset=True):
+        if reset:
             readable_set = self._readable_set = set()
-            np = self.getNumPartitions()
+            np = 1 + self._getMaxPartition()
             def _getPartition(x, np=np):
                 return x % np
             def _getReadablePartition(x, np=np, r=readable_set):
@@ -569,14 +562,15 @@ class DatabaseManager(object):
                 i = self._getDataLastId(p)
                 d.append(p << 48 if i is None else i + 1)
         else:
+            readable_set = self._readable_set
             readable_set.clear()
-        readable_set.update(x[0] for x in self._iterAssignedCells()
+        readable_set.update(x[0] for x in self.iterAssignedCells()
                                  if -x[1] in READABLE)
 
     @requires(_changePartitionTable, _getLastIDs, _getLastTID)
-    def changePartitionTable(self, ptid, cell_list, reset=False):
+    def changePartitionTable(self, ptid, num_replicas, cell_list, reset=False):
         my_nid = self.getUUID()
-        pt = dict(self._iterAssignedCells())
+        pt = dict(self.iterAssignedCells())
         # In backup mode, the last transactions of a readable cell may be
         # incomplete.
         backup_tid = self.getBackupTID()
@@ -595,13 +589,14 @@ class DatabaseManager(object):
                 outofdate_tid(offset)))
             for offset, nid, state in cell_list]
         self._changePartitionTable(cell_list, reset)
-        self._updateReadable()
+        self._updateReadable(reset)
         assert isinstance(ptid, (int, long)), ptid
         self._setConfiguration('ptid', str(ptid))
+        self._setConfiguration('replicas', str(num_replicas))
 
     @requires(_changePartitionTable)
     def updateCellTID(self, partition, tid):
-        t, = (t for p, t in self._iterAssignedCells() if p == partition)
+        t, = (t for p, t in self.iterAssignedCells() if p == partition)
         if t < 0:
             return
         tid = util.u64(tid)
@@ -623,7 +618,7 @@ class DatabaseManager(object):
             next_tid = util.u64(backup_tid)
             if next_tid:
                 next_tid += 1
-        for offset, tid in self._iterAssignedCells():
+        for offset, tid in self.iterAssignedCells():
             if tid >= 0: # OUT_OF_DATE
                 yield offset, p64(tid and tid + 1)
             elif -tid in READABLE:
@@ -865,7 +860,7 @@ class DatabaseManager(object):
             assert tid, tid
             cell_list = []
             my_nid = self.getUUID()
-            for partition, state in self._iterAssignedCells():
+            for partition, state in self.iterAssignedCells():
                 if state > tid:
                     cell_list.append((partition, my_nid, tid))
                 self._deleteRange(partition, tid)

neo/storage/database/mysqldb.py

@@ -117,9 +117,11 @@ class MySQLDatabaseManager(DatabaseManager):
         return super(MySQLDatabaseManager, self).__getattr__(attr)
 
     def _tryConnect(self):
-        kwd = {'db' : self.db, 'user' : self.user}
-        if self.passwd is not None:
-            kwd['passwd'] = self.passwd
+        kwd = {'db' : self.db}
+        if self.user:
+            kwd['user'] = self.user
+            if self.passwd is not None:
+                kwd['passwd'] = self.passwd
         if self.socket:
             kwd['unix_socket'] = os.path.expanduser(self.socket)
         logging.info('connecting to MySQL on the database %s with user %s',
@@ -198,6 +200,7 @@ class MySQLDatabaseManager(DatabaseManager):
             self._connect()
 
     def _commit(self):
+        # XXX: Should we translate OperationalError into MysqlError ?
         self.conn.commit()
         self._active = 0
 
@@ -270,6 +273,12 @@ class MySQLDatabaseManager(DatabaseManager):
             " ELSE 1-state"
             " END as tid")
 
+    # Let's wait for a more important change to clean up,
+    # so that users can still downgrade.
+    if 0:
+      def _migrate4(self, schema_dict):
+        self._setConfiguration('partitions', None)
+
     def _setup(self, dedup=False):
         self._config.clear()
         q = self.query
@@ -295,6 +304,12 @@ class MySQLDatabaseManager(DatabaseManager):
             p += """ PARTITION BY LIST (`partition`) (
                 PARTITION dummy VALUES IN (NULL))"""
 
+        if engine == "RocksDB":
+            cf = lambda name, rev=False: " COMMENT '%scf_neo_%s'" % (
+                'rev:' if rev else '', name)
+        else:
+            cf = lambda *_: ''
+
         # The table "trans" stores information on committed transactions.
         schema_dict['trans'] =  """CREATE TABLE %s (
                  `partition` SMALLINT UNSIGNED NOT NULL,
@@ -305,8 +320,8 @@ class MySQLDatabaseManager(DatabaseManager):
                  description BLOB NOT NULL,
                  ext BLOB NOT NULL,
                  ttid BIGINT UNSIGNED NOT NULL,
-                 PRIMARY KEY (`partition`, tid)
-             ) ENGINE=""" + p
+                 PRIMARY KEY (`partition`, tid){}
+             ) ENGINE={}""".format(cf('append_meta'), p)
 
         # The table "obj" stores committed object metadata.
         schema_dict['obj'] = """CREATE TABLE %s (
@@ -315,10 +330,11 @@ class MySQLDatabaseManager(DatabaseManager):
                  tid BIGINT UNSIGNED NOT NULL,
                  data_id BIGINT UNSIGNED NULL,
                  value_tid BIGINT UNSIGNED NULL,
-                 PRIMARY KEY (`partition`, oid, tid),
-                 KEY tid (`partition`, tid, oid),
-                 KEY (data_id)
-             ) ENGINE=""" + p
+                 PRIMARY KEY (`partition`, oid, tid){},
+                 KEY tid (`partition`, tid, oid){},
+                 KEY (data_id){}
+             ) ENGINE={}""".format(cf('obj_pk', True),
+                 cf('append_meta'), cf('append_meta'), p)
 
         if engine == "TokuDB":
             engine += " compression='tokudb_uncompressed'"
@@ -326,18 +342,21 @@ class MySQLDatabaseManager(DatabaseManager):
         # The table "data" stores object data.
         # We'd like to have partial index on 'hash' column (e.g. hash(4))
         # but 'UNIQUE' constraint would not work as expected.
-        schema_dict['data'] = """CREATE TABLE %%s (
-                 id BIGINT UNSIGNED NOT NULL PRIMARY KEY,
+        schema_dict['data'] = """CREATE TABLE %s (
+                 id BIGINT UNSIGNED NOT NULL,
                  hash BINARY(20) NOT NULL,
                  compression TINYINT UNSIGNED NULL,
-                 value MEDIUMBLOB NOT NULL%s
-             ) ENGINE=%s""" % (""",
-                 UNIQUE (hash, compression)""" if dedup else "", engine)
+                 value MEDIUMBLOB NOT NULL,
+                 PRIMARY KEY (id){}{}
+             ) ENGINE={}""".format(cf('append'), """,
+                 UNIQUE (hash, compression)""" + cf('no_comp') if dedup else "",
+                 engine)
 
         schema_dict['bigdata'] = """CREATE TABLE %s (
-                 id INT UNSIGNED NOT NULL AUTO_INCREMENT PRIMARY KEY,
-                 value MEDIUMBLOB NOT NULL
-             ) ENGINE=""" + engine
+                 id INT UNSIGNED NOT NULL AUTO_INCREMENT,
+                 value MEDIUMBLOB NOT NULL,
+                 PRIMARY KEY (id){}
+             ) ENGINE={}""".format(cf('append'), p)
 
         # The table "ttrans" stores information on uncommitted transactions.
         schema_dict['ttrans'] = """CREATE TABLE %s (
@@ -348,8 +367,9 @@ class MySQLDatabaseManager(DatabaseManager):
                  user BLOB NOT NULL,
                  description BLOB NOT NULL,
                  ext BLOB NOT NULL,
-                 ttid BIGINT UNSIGNED NOT NULL
-             ) ENGINE=""" + engine
+                 ttid BIGINT UNSIGNED NOT NULL,
+                 PRIMARY KEY (ttid){}
+             ) ENGINE={}""".format(cf('no_comp'), p)
 
         # The table "tobj" stores uncommitted object metadata.
         schema_dict['tobj'] = """CREATE TABLE %s (
@@ -358,8 +378,8 @@ class MySQLDatabaseManager(DatabaseManager):
                  tid BIGINT UNSIGNED NOT NULL,
                  data_id BIGINT UNSIGNED NULL,
                  value_tid BIGINT UNSIGNED NULL,
-                 PRIMARY KEY (tid, oid)
-             ) ENGINE=""" + engine
+                 PRIMARY KEY (tid, oid){}
+             ) ENGINE={}""".format(cf('no_comp'), p)
 
         if self.nonempty('config') is None:
             q(schema_dict.pop('config') % 'config')
@@ -407,6 +427,9 @@ class MySQLDatabaseManager(DatabaseManager):
             q("ALTER TABLE config MODIFY value VARBINARY(%s) NULL" % len(value))
             q(sql)
 
+    def _getMaxPartition(self):
+        return self.query("SELECT MAX(`partition`) FROM pt")[0][0]
+
     def _getPartitionTable(self):
         return self.query("SELECT * FROM pt")
 
@@ -965,7 +988,7 @@ class MySQLDatabaseManager(DatabaseManager):
         cmd += self._cmdline()
         return subprocess.check_output(cmd)
 
-    def restore(self, sql):
+    def _restore(self, sql):
         import subprocess
         cmd = ['mysql']
         cmd += self._cmdline()

neo/storage/database/sqlite.py

@@ -79,6 +79,7 @@ class SQLiteDatabaseManager(DatabaseManager):
     def _connect(self):
         logging.info('connecting to SQLite database %r', self.db)
         self.conn = sqlite3.connect(self.db, check_same_thread=False)
+        self.conn.text_factory = str
         self.lock(self.db)
         if self.UNSAFE:
             q = self.query
@@ -144,6 +145,12 @@ class SQLiteDatabaseManager(DatabaseManager):
             " WHEN 2 THEN -2"  # FEEDING
             " ELSE 1-state END")
 
+    # Let's wait for a more important change to clean up,
+    # so that users can still downgrade.
+    if 0:
+      def _migrate4(self, schema_dict, index_dict):
+        self._setConfiguration('partitions', None)
+
     def _setup(self, dedup=False):
         # BBB: SQLite has transactional DDL but before Python 3.6,
         #      the binding automatically commits between such statements.
@@ -265,6 +272,9 @@ class SQLiteDatabaseManager(DatabaseManager):
         else:
             q("REPLACE INTO config VALUES (?,?)", (key, str(value)))
 
+    def _getMaxPartition(self):
+        return self.query("SELECT MAX(`partition`) FROM pt").next()[0]
+
     def _getPartitionTable(self):
         return self.query("SELECT * FROM pt")
 
@@ -712,5 +722,5 @@ class SQLiteDatabaseManager(DatabaseManager):
         main[-1:-1] = data
         return '\n'.join(main) + '\n'
 
-    def restore(self, sql):
+    def _restore(self, sql):
         self.conn.executescript(sql)

neo/storage/handlers/__init__.py

@@ -65,14 +65,14 @@ class BaseMasterHandler(BaseHandler):
                 # See comment in ClientOperationHandler.connectionClosed
                 self.app.tm.abortFor(uuid, even_if_voted=True)
 
-    def notifyPartitionChanges(self, conn, ptid, cell_list):
+    def notifyPartitionChanges(self, conn, ptid, num_replicas, cell_list):
         """This is very similar to Send Partition Table, except that
        the information is only about changes from the previous."""
         app = self.app
         if ptid != 1 + app.pt.getID():
             raise ProtocolError('wrong partition table id')
-        app.pt.update(ptid, cell_list, app.nm)
-        app.dm.changePartitionTable(ptid, cell_list)
+        app.pt.update(ptid, num_replicas, cell_list, app.nm)
+        app.dm.changePartitionTable(ptid, num_replicas, cell_list)
         if app.operational:
             app.replicator.notifyPartitionChanges(cell_list)
         app.dm.commit()

neo/storage/handlers/identification.py

@@ -32,7 +32,7 @@ class IdentificationHandler(EventHandler):
         return self.app.nm
 
     def requestIdentification(self, conn, node_type, uuid, address, name,
-                              devpath, id_timestamp):
+                              id_timestamp, devpath, new_nid):
         self.checkClusterName(name)
         app = self.app
         # reject any incoming connections if not ready
@@ -65,6 +65,6 @@ class IdentificationHandler(EventHandler):
             conn.setHandler(handler)
             node.setConnection(conn, force)
         # accept the identification and trigger an event
-        conn.answer(Packets.AcceptIdentification(NodeTypes.STORAGE, uuid and
-            app.uuid, app.pt.getPartitions(), app.pt.getReplicas(), uuid))
+        conn.answer(Packets.AcceptIdentification(
+            NodeTypes.STORAGE, uuid and app.uuid, uuid))
         handler.connectionCompleted(conn)

neo/storage/handlers/initialization.py

@@ -20,10 +20,10 @@ from neo.lib.protocol import Packets, ProtocolError, ZERO_TID
 
 class InitializationHandler(BaseMasterHandler):
 
-    def sendPartitionTable(self, conn, ptid, row_list):
+    def sendPartitionTable(self, conn, ptid, num_replicas, row_list):
         app = self.app
         pt = app.pt
-        pt.load(ptid, row_list, app.nm)
+        pt.load(ptid, num_replicas, row_list, app.nm)
         if not pt.filled():
             raise ProtocolError('Partial partition table received')
         # Install the partition table into the database for persistence.
@@ -44,7 +44,7 @@ class InitializationHandler(BaseMasterHandler):
             logging.debug('drop data for partitions %r', unassigned)
             dm.dropPartitions(unassigned)
 
-        dm.changePartitionTable(ptid, cell_list, reset=True)
+        dm.changePartitionTable(ptid, num_replicas, cell_list, reset=True)
         dm.commit()
 
     def truncate(self, conn, tid):
@@ -68,7 +68,8 @@ class InitializationHandler(BaseMasterHandler):
 
     def askPartitionTable(self, conn):
         pt = self.app.pt
-        conn.answer(Packets.AnswerPartitionTable(pt.getID(), pt.getRowList()))
+        conn.answer(Packets.AnswerPartitionTable(
+            pt.getID(), pt.getReplicas(), pt.getRowList()))
 
     def askLockedTransactions(self, conn):
         conn.answer(Packets.AnswerLockedTransactions(

neo/storage/replicator.py

@@ -350,7 +350,7 @@ class Replicator(object):
             try:
                 conn.ask(Packets.RequestIdentification(NodeTypes.STORAGE,
                     None if name else app.uuid, app.server, name or app.name,
-                    (), app.id_timestamp))
+                    app.id_timestamp, (), ()))
             except ConnectionClosed:
                 if previous_node is self.current_node:
                     return

neo/storage/transactions.py

@@ -98,9 +98,12 @@ class TransactionManager(EventQueue):
         self._load_lock_dict = {}
         self._replicated = {}
         self._replicating = set()
+
+    def getPartition(self, oid):
         from neo.lib.util import u64
-        np = app.pt.getPartitions()
+        np = self._app.pt.getPartitions()
         self.getPartition = lambda oid: u64(oid) % np
+        return self.getPartition(oid)
 
     def discarded(self, offset_list):
         self._replicating.difference_update(offset_list)

neo/tests/__init__.py

@@ -21,6 +21,7 @@ import gc
 import os
 import random
 import socket
+import subprocess
 import sys
 import tempfile
 import unittest
@@ -41,7 +42,7 @@ from .mock import Mock
 from neo.lib import debug, logging, protocol
 from neo.lib.protocol import NodeTypes, Packets, UUID_NAMESPACES
 from neo.lib.util import cached_property
-from time import time
+from time import time, sleep
 from struct import pack, unpack
 from unittest.case import _ExpectedFailure, _UnexpectedSuccess
 try:
@@ -72,6 +73,9 @@ DB_ADMIN = os.getenv('NEO_DB_ADMIN', 'root')
 DB_PASSWD = os.getenv('NEO_DB_PASSWD', '')
 DB_USER = os.getenv('NEO_DB_USER', 'test')
 DB_SOCKET = os.getenv('NEO_DB_SOCKET', '')
+DB_INSTALL = os.getenv('NEO_DB_INSTALL', 'mysql_install_db')
+DB_MYSQLD = os.getenv('NEO_DB_MYSQLD', '/usr/sbin/mysqld')
+DB_MYCNF = os.getenv('NEO_DB_MYCNF')
 
 IP_VERSION_FORMAT_DICT = {
     socket.AF_INET:  '127.0.0.1',
@@ -134,8 +138,12 @@ def getTempDirectory():
         print 'Using temp directory %r.' % temp_dir
     return temp_dir
 
-def setupMySQLdb(db_list, user=DB_USER, password='', clear_databases=True):
+def setupMySQLdb(db_list, clear_databases=True):
+    if mysql_pool:
+        return mysql_pool.setup(db_list, clear_databases)
     from MySQLdb.constants.ER import BAD_DB_ERROR
+    user = DB_USER
+    password = ''
     kw = {'unix_socket': os.path.expanduser(DB_SOCKET)} if DB_SOCKET else {}
     conn = MySQLdb.connect(user=DB_ADMIN, passwd=DB_PASSWD, **kw)
     cursor = conn.cursor()
@@ -154,6 +162,88 @@ def setupMySQLdb(db_list, user=DB_USER, password='', clear_databases=True):
     cursor.close()
     conn.commit()
     conn.close()
+    return '{}:{}@%s{}'.format(user, password, DB_SOCKET).__mod__
+
+class MySQLPool(object):
+
+    def __init__(self, pool_dir=None):
+        self._args = {}
+        self._mysqld_dict = {}
+        if not pool_dir:
+            pool_dir = getTempDirectory()
+        self._base = pool_dir + os.sep
+        self._sock_template = os.path.join(pool_dir, '%s', 'mysql.sock')
+
+    def __del__(self):
+        self.kill(*self._mysqld_dict)
+
+    def setup(self, db_list, clear_databases):
+        start_list = set(db_list).difference(self._mysqld_dict)
+        if start_list:
+            start_list = sorted(start_list)
+            x = []
+            with open(os.devnull, 'wb') as f:
+                for db in start_list:
+                    base = self._base + db
+                    datadir = os.path.join(base, 'datadir')
+                    sock = self._sock_template % db
+                    tmpdir = os.path.join(base, 'tmp')
+                    args = [DB_INSTALL,
+                        '--defaults-file=' + DB_MYCNF,
+                        '--datadir=' + datadir,
+                        '--socket=' + sock,
+                        '--tmpdir=' + tmpdir,
+                        '--log_error=' + os.path.join(base, 'error.log')]
+                    if os.path.exists(datadir):
+                        try:
+                            os.remove(sock)
+                        except OSError, e:
+                            if e.errno != errno.ENOENT:
+                                raise
+                    else:
+                        os.makedirs(tmpdir)
+                        x.append(subprocess.Popen(args,
+                            stdout=f, stderr=subprocess.STDOUT))
+                    args[0] = DB_MYSQLD
+                    self._args[db] = args
+            for x in x:
+                x = x.wait()
+                if x:
+                    raise subprocess.CalledProcessError(x, DB_INSTALL)
+            self.start(*start_list)
+            for db in start_list:
+                sock = self._sock_template % db
+                p = self._mysqld_dict[db]
+                while not os.path.exists(sock):
+                    sleep(1)
+                    x = p.poll()
+                    if x is not None:
+                        raise subprocess.CalledProcessError(x, DB_MYSQLD)
+        for db in db_list:
+            db = MySQLdb.connect(unix_socket=self._sock_template % db,
+                                 user='root')
+            if clear_databases:
+                db.query('DROP DATABASE IF EXISTS neo')
+            db.query('CREATE DATABASE IF NOT EXISTS neo')
+            db.close()
+        return ('root@neo' + self._sock_template).__mod__
+
+    def start(self, *db, **kw):
+        assert set(db).isdisjoint(self._mysqld_dict)
+        for db in db:
+            self._mysqld_dict[db] = subprocess.Popen(self._args[db], **kw)
+
+    def kill(self, *db):
+        processes = []
+        for db in db:
+            p = self._mysqld_dict.pop(db)
+            processes.append(p)
+            p.kill()
+        for p in processes:
+            p.wait()
+
+mysql_pool = MySQLPool() if DB_MYCNF else None
+
 
 def ImporterConfigParser(adapter, zodb, **kw):
     cfg = SafeConfigParser()
@@ -244,13 +334,15 @@ class NeoUnitTestBase(NeoTestBase):
         """ create empty databases """
         adapter = os.getenv('NEO_TESTS_ADAPTER', 'MySQL')
         if adapter == 'MySQL':
-            setupMySQLdb([prefix + str(i) for i in xrange(number)])
+            db_template = setupMySQLdb(
+                [prefix + str(i) for i in xrange(number)])
+            self.db_template = lambda i: db_template(prefix + str(i))
         elif adapter == 'SQLite':
-            temp_dir = getTempDirectory()
+            self.db_template = os.path.join(getTempDirectory(),
+                                       prefix + '%s.sqlite').__mod__
             for i in xrange(number):
                 try:
-                    os.remove(os.path.join(temp_dir,
-                        '%s%s.sqlite' % (prefix, i)))
+                    os.remove(self.db_template(i))
                 except OSError, e:
                     if e.errno != errno.ENOENT:
                         raise
@@ -274,21 +366,14 @@ class NeoUnitTestBase(NeoTestBase):
     def getStorageConfiguration(self, cluster='main', master_number=2,
             index=0, prefix=DB_PREFIX, uuid=None):
         assert master_number >= 1 and master_number <= 10
-        assert index >= 0 and index <= 9
         masters = [(buildUrlFromString(self.local_ip),
                      10010 + i) for i in xrange(master_number)]
         adapter = os.getenv('NEO_TESTS_ADAPTER', 'MySQL')
-        if adapter == 'MySQL':
-            db = '%s@%s%s%s' % (DB_USER, prefix, index, DB_SOCKET)
-        elif adapter == 'SQLite':
-            db = os.path.join(getTempDirectory(), 'test_neo%s.sqlite' % index)
-        else:
-            assert False, adapter
         return {
                 'cluster': cluster,
                 'bind': (masters[0], 10020 + index),
                 'masters': masters,
-                'database': db,
+                'database': self.db_template(index),
                 'uuid': uuid,
                 'adapter': adapter,
                 'wait': 0,

neo/tests/functional/__init__.py

@@ -36,7 +36,7 @@ from neo.lib import logging
 from neo.lib.protocol import ClusterStates, NodeTypes, CellStates, NodeStates, \
     UUID_NAMESPACES
 from neo.lib.util import dump, setproctitle
-from .. import (ADDRESS_TYPE, DB_SOCKET, DB_USER, IP_VERSION_FORMAT_DICT, SSL,
+from .. import (ADDRESS_TYPE, IP_VERSION_FORMAT_DICT, SSL,
     buildUrlFromString, cluster, getTempDirectory, setupMySQLdb,
     ImporterConfigParser, NeoTestBase, Patch)
 from neo.client.Storage import Storage
@@ -282,7 +282,7 @@ class NEOProcess(Process):
     def _args(self):
         args = super(NEOProcess, self)._args()
         if self.uuid:
-            args[:0] = '--uuid', str(self.uuid)
+            args[:0] = '--nid', str(self.uuid)
         return args
 
     def run(self):
@@ -306,11 +306,11 @@ class NEOCluster(object):
     SSL = None
 
     def __init__(self, db_list, master_count=1, partitions=1, replicas=0,
-                 db_user=DB_USER, db_password='', name=None,
+                 name=None,
                  cleanup_on_delete=False, temp_dir=None, clear_databases=True,
                  adapter=os.getenv('NEO_TESTS_ADAPTER'),
                  address_type=ADDRESS_TYPE, bind_ip=None, logger=True,
-                 importer=None, upstream_masters=None, upstream_cluster=None):
+                 importer=None, upstream_masters=None, upstream_cluster=None, storage_kw={}):
         if not adapter:
             adapter = 'MySQL'
         self.adapter = adapter
@@ -322,20 +322,28 @@ class NEOCluster(object):
             temp_dir = tempfile.mkdtemp(prefix='neo_')
             print 'Using temp directory ' + temp_dir
         if adapter == 'MySQL':
-            self.db_user = db_user
-            self.db_password = db_password
-            self.db_template = ('%s:%s@%%s%s' % (db_user, db_password,
-                                                 DB_SOCKET)).__mod__
+            self.db_template = setupMySQLdb(db_list, clear_databases)
         elif adapter == 'SQLite':
             self.db_template = (lambda t: lambda db:
                 ':memory:' if db is None else db if os.sep in db else t % db
                 )(os.path.join(temp_dir, '%s.sqlite'))
+            if clear_databases:
+                for db in self.db_list:
+                    if db is None:
+                        continue
+                    db = self.db_template(db)
+                    try:
+                        os.remove(db)
+                    except OSError, e:
+                        if e.errno != errno.ENOENT:
+                            raise
+                    else:
+                        logging.debug('%r deleted', db)
         else:
             assert False, adapter
         self.address_type = address_type
         self.local_ip = local_ip = bind_ip or \
             IP_VERSION_FORMAT_DICT[self.address_type]
-        self.setupDB(clear_databases)
         if importer:
             cfg = ImporterConfigParser(adapter, **importer)
             cfg.set("neo", "database", self.db_template(*db_list))
@@ -372,7 +380,8 @@ class NEOCluster(object):
         # create storage nodes
         for i, db in enumerate(db_list):
             self._newProcess(NodeTypes.STORAGE, logger and 'storage_%u' % i,
-                             0, adapter=adapter, database=self.db_template(db))
+                             0, adapter=adapter, database=self.db_template(db),
+                             **storage_kw)
         # create neoctl
         self.neoctl = NeoCTL((self.local_ip, admin_port), ssl=self.SSL)
 
@@ -390,23 +399,10 @@ class NEOCluster(object):
         self.process_dict.setdefault(node_type, []).append(
             NEOProcess(command_dict[node_type], uuid=uuid, **kw))
 
-    def setupDB(self, clear_databases=True):
-        if self.adapter == 'MySQL':
-            setupMySQLdb(self.db_list, self.db_user, self.db_password,
-                         clear_databases)
-        elif self.adapter == 'SQLite':
-            if clear_databases:
-                for db in self.db_list:
-                    if db is None:
-                        continue
-                    db = self.db_template(db)
-                    try:
-                        os.remove(db)
-                    except OSError, e:
-                        if e.errno != errno.ENOENT:
-                            raise
-                    else:
-                        logging.debug('%r deleted', db)
+    def resetDB(self):
+        for db in self.db_list:
+            dm = buildDatabaseManager(self.adapter, (self.db_template(db),))
+            dm.setup(True)
 
     def run(self, except_storages=()):
         """ Start cluster processes except some storage nodes """
@@ -445,7 +441,7 @@ class NEOCluster(object):
                         pending_count += 1
                     if pending_count == target[0]:
                         neoctl.startCluster()
-            except (NotReadyException, RuntimeError):
+            except (NotReadyException, SystemExit):
                 pass
         if not pdb.wait(test, MAX_START_TIME):
             raise AssertionError('Timeout when starting cluster')
@@ -457,7 +453,7 @@ class NEOCluster(object):
         def start(last_try):
             try:
                 self.neoctl.startCluster()
-            except (NotReadyException, RuntimeError), e:
+            except (NotReadyException, SystemExit), e:
                 return False, e
             return True, None
         self.expectCondition(start)
@@ -661,10 +657,10 @@ class NEOCluster(object):
 
     def expectOudatedCells(self, number, *args, **kw):
         def callback(last_try):
-            row_list = self.neoctl.getPartitionRowList()[1]
+            row_list = self.neoctl.getPartitionRowList()[2]
             number_of_outdated = 0
             for row in row_list:
-                for cell in row[1]:
+                for cell in row:
                     if cell[1] == CellStates.OUT_OF_DATE:
                         number_of_outdated += 1
             return number_of_outdated == number, number_of_outdated
@@ -672,10 +668,10 @@ class NEOCluster(object):
 
     def expectAssignedCells(self, process, number, *args, **kw):
         def callback(last_try):
-            row_list = self.neoctl.getPartitionRowList()[1]
+            row_list = self.neoctl.getPartitionRowList()[2]
             assigned_cells_number = 0
             for row in row_list:
-                for cell in row[1]:
+                for cell in row:
                     if cell[0] == process.getUUID():
                         assigned_cells_number += 1
             return assigned_cells_number == number, assigned_cells_number

neo/tests/functional/testClient.py

@@ -62,8 +62,6 @@ class ClientTests(NEOFunctionalTest):
         NEOFunctionalTest._tearDown(self, success)
 
     def __setup(self):
-        # start cluster
-        self.neo.setupDB()
         self.neo.start()
         self.neo.expectClusterRunning()
         self.db = ZODB.DB(self.neo.getZODBStorage())

neo/tests/functional/testCluster.py

@@ -71,7 +71,6 @@ class ClusterTests(NEOFunctionalTest):
     def testClusterBreaks(self):
         self.neo = NEOCluster(['test_neo1'],
                 master_count=1, temp_dir=self.getTempDirectory())
-        self.neo.setupDB()
         self.neo.start()
         self.neo.expectClusterRunning()
         self.neo.expectOudatedCells(number=0)
@@ -82,7 +81,6 @@ class ClusterTests(NEOFunctionalTest):
         self.neo = NEOCluster(['test_neo1', 'test_neo2'],
                  partitions=2, master_count=1, replicas=0,
                  temp_dir=self.getTempDirectory())
-        self.neo.setupDB()
         self.neo.start()
         self.neo.expectClusterRunning()
         self.neo.expectOudatedCells(number=0)
@@ -93,7 +91,6 @@ class ClusterTests(NEOFunctionalTest):
         self.neo = NEOCluster(['test_neo1', 'test_neo2'],
                          partitions=2, replicas=1, master_count=1,
                          temp_dir=self.getTempDirectory())
-        self.neo.setupDB()
         self.neo.start()
         self.neo.expectClusterRunning()
         self.neo.expectOudatedCells(number=0)

neo/tests/functional/testMaster.py

@@ -47,7 +47,7 @@ class MasterTests(NEOFunctionalTest):
                 break
         neoctl.killNode(uuid)
         self.neo.expectDead(master)
-        self.assertRaises(RuntimeError, neoctl.killNode, primary_uuid)
+        self.assertRaises(SystemExit, neoctl.killNode, primary_uuid)
 
     def testStoppingPrimaryWithTwoSecondaries(self):
         # Wait for masters to stabilize

neo/tests/functional/testStorage.py

@@ -172,7 +172,7 @@ class StorageTests(NEOFunctionalTest):
         self.neo.expectOudatedCells(2)
         self.neo.expectClusterRunning()
 
-        self.assertRaises(RuntimeError, self.neo.neoctl.killNode,
+        self.assertRaises(SystemExit, self.neo.neoctl.killNode,
             started[1].getUUID())
         started[1].stop()
         # Cluster not operational anymore. Only cells of second storage that
@@ -323,7 +323,7 @@ class StorageTests(NEOFunctionalTest):
         self.neo.expectStorageUnknown(started[0])
         self.neo.expectAssignedCells(started[0], 0)
         self.neo.expectAssignedCells(started[1], 10)
-        self.assertRaises(RuntimeError, self.neo.neoctl.dropNode,
+        self.assertRaises(SystemExit, self.neo.neoctl.dropNode,
                           started[1].getUUID())
         self.neo.expectClusterRunning()
 

neo/tests/master/testClientHandler.py

@@ -30,8 +30,6 @@ class MasterClientHandlerTests(NeoUnitTestBase):
         config = self.getMasterConfiguration(master_number=1, replicas=1)
         self.app = Application(config)
         self.app.em.close()
-        self.app.pt.clear()
-        self.app.pt.setID(1)
         self.app.em = Mock()
         self.app.loid = '\0' * 8
         self.app.tm.setLastTID('\0' * 8)

neo/tests/master/testMasterApp.py

@@ -26,7 +26,6 @@ class MasterAppTests(NeoUnitTestBase):
         # create an application object
         config = self.getMasterConfiguration()
         self.app = Application(config)
-        self.app.pt.clear()
 
     def _tearDown(self, success):
         self.app.close()

neo/tests/master/testMasterPT.py

@@ -289,7 +289,9 @@ class MasterPartitionTableTests(NeoUnitTestBase):
         pt.addNodeList(sn[1:3])
         self.assertPartitionTable(pt, 'U..|U..|U..|U..|U..|U..|U..')
         self.update(pt, self.tweak(pt, sn[:1]))
-        self.assertPartitionTable(pt, '.U.|..U|.U.|..U|.U.|..U|.U.')
+        # See note in PartitionTable.tweak() about drop_list.
+        #self.assertPartitionTable(pt,'.U.|..U|.U.|..U|.U.|..U|.U.')
+        self.assertPartitionTable(pt, 'UU.|U.U|UU.|U.U|UU.|U.U|UU.')
 
     def test_18_tweakBigPT(self):
         seed = repr(time.time())

neo/tests/master/testRecovery.py

-#
-# Copyright (C) 2009-2019  Nexedi SA
-#
-# This program is free software; you can redistribute it and/or
-# modify it under the terms of the GNU General Public License
-# as published by the Free Software Foundation; either version 2
-# of the License, or (at your option) any later version.
-#
-# This program is distributed in the hope that it will be useful,
-# but WITHOUT ANY WARRANTY; without even the implied warranty of
-# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-# GNU General Public License for more details.
-#
-# You should have received a copy of the GNU General Public License
-# along with this program.  If not, see <http://www.gnu.org/licenses/>.
-
-import unittest
-from .. import NeoUnitTestBase
-from neo.lib.protocol import NodeTypes, NodeStates, CellStates
-from neo.master.recovery import RecoveryManager
-from neo.master.app import Application
-
-class MasterRecoveryTests(NeoUnitTestBase):
-
-    def setUp(self):
-        NeoUnitTestBase.setUp(self)
-        # create an application object
-        config = self.getMasterConfiguration()
-        self.app = Application(config)
-        self.app.pt.clear()
-        self.recovery = RecoveryManager(self.app)
-        self.app.unconnected_master_node_set = set()
-        self.app.negotiating_master_node_set = set()
-        for node in self.app.nm.getMasterList():
-            self.app.unconnected_master_node_set.add(node.getAddress())
-            node.setState(NodeStates.RUNNING)
-
-        # define some variable to simulate client and storage node
-        self.storage_port = 10021
-        self.master_port = 10011
-
-    def _tearDown(self, success):
-        self.app.close()
-        NeoUnitTestBase._tearDown(self, success)
-
-    # Common methods
-    def identifyToMasterNode(self, node_type=NodeTypes.STORAGE, ip="127.0.0.1",
-                             port=10021):
-        """Do first step of identification to MN
-        """
-        address = (ip, port)
-        uuid = self.getNewUUID(node_type)
-        self.app.nm.createFromNodeType(node_type, address=address, uuid=uuid,
-            state=NodeStates.RUNNING)
-        return uuid
-
-    # Tests
-    def test_10_answerPartitionTable(self):
-        # XXX: This test does much less that it seems, because all 'for' loops
-        #      iterate over empty lists. Currently, only testRecovery covers
-        #      some paths in NodeManager._createNode: apart from that, we could
-        #      delete it entirely.
-        recovery = self.recovery
-        uuid = self.identifyToMasterNode(NodeTypes.MASTER, port=self.master_port)
-        # not from target node, ignore
-        uuid = self.identifyToMasterNode(NodeTypes.STORAGE, port=self.storage_port)
-        conn = self.getFakeConnection(uuid, self.storage_port)
-        node = self.app.nm.getByUUID(conn.getUUID())
-        offset = 1
-        cell_list = [(offset, uuid, CellStates.UP_TO_DATE)]
-        cells = self.app.pt.getRow(offset)
-        for cell, state in cells:
-            self.assertEqual(state, CellStates.OUT_OF_DATE)
-        recovery.target_ptid = 2
-        node.setPending()
-        recovery.answerPartitionTable(conn, 1, cell_list)
-        cells = self.app.pt.getRow(offset)
-        for cell, state in cells:
-            self.assertEqual(state, CellStates.OUT_OF_DATE)
-        # from target node, taken into account
-        conn = self.getFakeConnection(uuid, self.storage_port)
-        offset = 1
-        cell_list = [(offset, ((uuid, CellStates.UP_TO_DATE,),),)]
-        cells = self.app.pt.getRow(offset)
-        for cell, state in cells:
-            self.assertEqual(state, CellStates.OUT_OF_DATE)
-        node.setPending()
-        recovery.answerPartitionTable(conn, None, cell_list)
-        cells = self.app.pt.getRow(offset)
-        for cell, state in cells:
-            self.assertEqual(state, CellStates.UP_TO_DATE)
-        # give a bad offset, must send error
-        self.recovery.target_uuid = uuid
-        conn = self.getFakeConnection(uuid, self.storage_port)
-        offset = 1000000
-        self.assertFalse(self.app.pt.hasOffset(offset))
-        cell_list = [(offset, ((uuid, NodeStates.UNKNOWN,),),)]
-        node.setPending()
-        self.checkProtocolErrorRaised(recovery.answerPartitionTable, conn,
-            2, cell_list)
-
-
-if __name__ == '__main__':
-    unittest.main()
-

neo/tests/master/testStorageHandler.py

@@ -18,8 +18,8 @@ import unittest
 from ..mock import Mock
 from .. import NeoUnitTestBase
 from neo.lib.protocol import NodeTypes, Packets
-from neo.master.handlers.storage import StorageServiceHandler
 from neo.master.app import Application
+from neo.master.handlers.storage import StorageServiceHandler
 
 class MasterStorageHandlerTests(NeoUnitTestBase):
 
@@ -29,7 +29,6 @@ class MasterStorageHandlerTests(NeoUnitTestBase):
         config = self.getMasterConfiguration(master_number=1, replicas=1)
         self.app = Application(config)
         self.app.em.close()
-        self.app.pt.clear()
         self.app.em = Mock()
         self.service = StorageServiceHandler(self.app)
 

neo/tests/storage/testMasterHandler.py

@@ -56,7 +56,7 @@ class StorageMasterHandlerTests(NeoUnitTestBase):
         self.app.pt = Mock({'getID': 1})
         count = len(self.app.nm.getList())
         self.assertRaises(ProtocolError, self.operation.notifyPartitionChanges,
-                          conn, 0, ())
+                          conn, 0, 0, ())
         self.assertEqual(self.app.pt.getID(), 1)
         self.assertEqual(len(self.app.nm.getList()), count)
         calls = self.app.replicator.mockGetNamedCalls('removePartition')
@@ -84,13 +84,13 @@ class StorageMasterHandlerTests(NeoUnitTestBase):
         ptid = 2
         app.dm = Mock({ })
         app.replicator = Mock({})
-        self.operation.notifyPartitionChanges(conn, ptid, cells)
+        self.operation.notifyPartitionChanges(conn, ptid, 1, cells)
         # ptid set
         self.assertEqual(app.pt.getID(), ptid)
         # dm call
         calls = self.app.dm.mockGetNamedCalls('changePartitionTable')
         self.assertEqual(len(calls), 1)
-        calls[0].checkArgs(ptid, cells)
+        calls[0].checkArgs(ptid, 1, cells)
 
 if __name__ == "__main__":
     unittest.main()

neo/tests/storage/testStorageApp.py

-#
-# Copyright (C) 2009-2019  Nexedi SA
-#
-# This program is free software; you can redistribute it and/or
-# modify it under the terms of the GNU General Public License
-# as published by the Free Software Foundation; either version 2
-# of the License, or (at your option) any later version.
-#
-# This program is distributed in the hope that it will be useful,
-# but WITHOUT ANY WARRANTY; without even the implied warranty of
-# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-# GNU General Public License for more details.
-#
-# You should have received a copy of the GNU General Public License
-# along with this program.  If not, see <http://www.gnu.org/licenses/>.
-
-import unittest
-from ..mock import Mock
-from .. import NeoUnitTestBase
-from neo.storage.app import Application
-from neo.lib.protocol import CellStates
-from neo.lib.pt import PartitionTable
-
-class StorageAppTests(NeoUnitTestBase):
-
-    def setUp(self):
-        NeoUnitTestBase.setUp(self)
-        self.prepareDatabase(number=1)
-        # create an application object
-        config = self.getStorageConfiguration(master_number=1)
-        self.app = Application(config)
-
-    def _tearDown(self, success):
-        self.app.close()
-        del self.app
-        super(StorageAppTests, self)._tearDown(success)
-
-    def test_01_loadPartitionTable(self):
-        self.app.dm = Mock({
-            'getPartitionTable': [],
-        })
-        self.assertEqual(self.app.pt, None)
-        num_partitions = 3
-        num_replicas = 2
-        self.app.pt = PartitionTable(num_partitions, num_replicas)
-        self.assertFalse(self.app.pt.getNodeSet())
-        self.assertFalse(self.app.pt.filled())
-        for x in xrange(num_partitions):
-            self.assertFalse(self.app.pt.hasOffset(x))
-
-        # load an empty table
-        self.app.loadPartitionTable()
-        self.assertFalse(self.app.pt.getNodeSet())
-        self.assertFalse(self.app.pt.filled())
-        for x in xrange(num_partitions):
-            self.assertFalse(self.app.pt.hasOffset(x))
-
-        # add some node, will be remove when loading table
-        master_uuid = self.getMasterUUID()
-        master = self.app.nm.createMaster(uuid=master_uuid)
-        storage_uuid = self.getStorageUUID()
-        storage = self.app.nm.createStorage(uuid=storage_uuid)
-        client_uuid = self.getClientUUID()
-
-        self.app.pt._setCell(0, master, CellStates.UP_TO_DATE)
-        self.app.pt._setCell(0, storage, CellStates.UP_TO_DATE)
-        self.assertEqual(len(self.app.pt.getNodeSet()), 2)
-        self.assertFalse(self.app.pt.filled())
-        for x in xrange(num_partitions):
-            if x == 0:
-                self.assertTrue(self.app.pt.hasOffset(x))
-            else:
-                self.assertFalse(self.app.pt.hasOffset(x))
-        # load an empty table, everything removed
-        self.app.loadPartitionTable()
-        self.assertFalse(self.app.pt.getNodeSet())
-        self.assertFalse(self.app.pt.filled())
-        for x in xrange(num_partitions):
-            self.assertFalse(self.app.pt.hasOffset(x))
-
-        # add some node
-        self.app.pt._setCell(0, master, CellStates.UP_TO_DATE)
-        self.app.pt._setCell(0, storage, CellStates.UP_TO_DATE)
-        self.assertEqual(len(self.app.pt.getNodeSet()), 2)
-        self.assertFalse(self.app.pt.filled())
-        for x in xrange(num_partitions):
-            if x == 0:
-                self.assertTrue(self.app.pt.hasOffset(x))
-            else:
-                self.assertFalse(self.app.pt.hasOffset(x))
-        # fill partition table
-        self.app.dm = Mock({
-            'getPartitionTable': [
-                (0, client_uuid, CellStates.UP_TO_DATE),
-                (1, client_uuid, CellStates.UP_TO_DATE),
-                (1, storage_uuid, CellStates.UP_TO_DATE),
-                (2, storage_uuid, CellStates.UP_TO_DATE),
-                (2, master_uuid, CellStates.UP_TO_DATE),
-            ],
-            'getPTID': 1,
-        })
-        self.app.pt.clear()
-        self.app.loadPartitionTable()
-        self.assertTrue(self.app.pt.filled())
-        for x in xrange(num_partitions):
-            self.assertTrue(self.app.pt.hasOffset(x))
-        # check each row
-        cell_list = self.app.pt.getCellList(0)
-        self.assertEqual(len(cell_list), 1)
-        self.assertEqual(cell_list[0].getUUID(), client_uuid)
-        cell_list = self.app.pt.getCellList(1)
-        self.assertEqual(len(cell_list), 2)
-        self.assertTrue(cell_list[0].getUUID() in (client_uuid, storage_uuid))
-        self.assertTrue(cell_list[1].getUUID() in (client_uuid, storage_uuid))
-        cell_list = self.app.pt.getCellList(2)
-        self.assertEqual(len(cell_list), 2)
-        self.assertTrue(cell_list[0].getUUID() in (master_uuid, storage_uuid))
-        self.assertTrue(cell_list[1].getUUID() in (master_uuid, storage_uuid))
-
-if __name__ == '__main__':
-    unittest.main()
-

neo/tests/storage/testStorageDBTests.py

@@ -48,30 +48,15 @@ class StorageDBTests(NeoUnitTestBase):
         raise NotImplementedError
 
     def setNumPartitions(self, num_partitions, reset=0):
-        try:
-            db = self._db
-        except AttributeError:
-            self._db = db = self.getDB(reset)
-        else:
-            if reset:
-                db.setup(reset)
-            else:
-                try:
-                    n = db.getNumPartitions()
-                except KeyError:
-                    n = 0
-                if num_partitions == n:
-                    return
-                if num_partitions < n:
-                    db.dropPartitions(n)
-        db.setNumPartitions(num_partitions)
-        self.assertEqual(num_partitions, db.getNumPartitions())
+        assert not hasattr(self, '_db')
+        self._db = db = self.getDB(reset)
         uuid = self.getStorageUUID()
         db.setUUID(uuid)
         self.assertEqual(uuid, db.getUUID())
-        db.changePartitionTable(1,
+        db.changePartitionTable(1, 0,
             [(i, uuid, CellStates.UP_TO_DATE) for i in xrange(num_partitions)],
             reset=True)
+        self.assertEqual(num_partitions, 1 + db._getMaxPartition())
         db.commit()
 
     def checkConfigEntry(self, get_call, set_call, value):
@@ -102,16 +87,6 @@ class StorageDBTests(NeoUnitTestBase):
         db = self.getDB()
         self.checkConfigEntry(db.getName, db.setName, 'TEST_NAME')
 
-    def test_getPartitionTable(self):
-        db = self.getDB()
-        db.setNumPartitions(3)
-        uuid1, uuid2 = self.getStorageUUID(), self.getStorageUUID()
-        cell1 = (0, uuid1, CellStates.OUT_OF_DATE)
-        cell2 = (1, uuid1, CellStates.UP_TO_DATE)
-        db.changePartitionTable(1, [cell1, cell2], 1)
-        result = db.getPartitionTable()
-        self.assertEqual(set(result), {cell1, cell2})
-
     def getOIDs(self, count):
         return map(p64, xrange(count))
 
@@ -202,52 +177,6 @@ class StorageDBTests(NeoUnitTestBase):
         self.assertEqual(self.db.getObject(oid1, before_tid=tid2),
             OBJECT_T1_NEXT)
 
-    def test_setPartitionTable(self):
-        db = self.getDB()
-        db.setNumPartitions(3)
-        ptid = 1
-        uuid = self.getStorageUUID()
-        cell1 = 0, uuid, CellStates.OUT_OF_DATE
-        cell2 = 1, uuid, CellStates.UP_TO_DATE
-        cell3 = 1, uuid, CellStates.DISCARDED
-        # no partition table
-        self.assertEqual(list(db.getPartitionTable()), [])
-        # set one
-        db.changePartitionTable(ptid, [cell1], 1)
-        result = db.getPartitionTable()
-        self.assertEqual(list(result), [cell1])
-        # then another
-        db.changePartitionTable(ptid, [cell2], 1)
-        result = db.getPartitionTable()
-        self.assertEqual(list(result), [cell2])
-        # drop discarded cells
-        db.changePartitionTable(ptid, [cell2, cell3], 1)
-        result = db.getPartitionTable()
-        self.assertEqual(list(result), [])
-
-    def test_changePartitionTable(self):
-        db = self.getDB()
-        db.setNumPartitions(3)
-        ptid = 1
-        uuid = self.getStorageUUID()
-        cell1 = 0, uuid, CellStates.OUT_OF_DATE
-        cell2 = 1, uuid, CellStates.UP_TO_DATE
-        cell3 = 1, uuid, CellStates.DISCARDED
-        # no partition table
-        self.assertEqual(list(db.getPartitionTable()), [])
-        # set one
-        db.changePartitionTable(ptid, [cell1])
-        result = db.getPartitionTable()
-        self.assertEqual(list(result), [cell1])
-        # add more entries
-        db.changePartitionTable(ptid, [cell2])
-        result = db.getPartitionTable()
-        self.assertEqual(set(result), {cell1, cell2})
-        # drop discarded cells
-        db.changePartitionTable(ptid, [cell2, cell3])
-        result = db.getPartitionTable()
-        self.assertEqual(list(result), [cell1])
-
     def test_commitTransaction(self):
         oid1, oid2 = self.getOIDs(2)
         tid1, tid2 = self.getTIDs(2)

neo/tests/storage/testStorageMySQL.py

@@ -22,7 +22,7 @@ from MySQLdb.constants.ER import UNKNOWN_STORAGE_ENGINE
 from ..mock import Mock
 from neo.lib.protocol import ZERO_OID
 from neo.lib.util import p64
-from .. import DB_PREFIX, DB_SOCKET, DB_USER, Patch
+from .. import DB_PREFIX, DB_USER, Patch, setupMySQLdb
 from .testStorageDBTests import StorageDBTests
 from neo.storage.database import DatabaseFailure
 from neo.storage.database.mysqldb import MySQLDatabaseManager
@@ -46,8 +46,8 @@ class StorageMySQLdbTests(StorageDBTests):
     engine = None
 
     def _test_lockDatabase_open(self):
-        self.prepareDatabase(number=1, prefix=DB_PREFIX)
-        database = '%s@%s0%s' % (DB_USER, DB_PREFIX, DB_SOCKET)
+        self.prepareDatabase(1)
+        database = self.db_template(0)
         return MySQLDatabaseManager(database, self.engine)
 
     def getDB(self, reset=0):

neo/tests/stress.py

@@ -19,12 +19,9 @@ class Handler(MasterEventHandler):
         super(Handler, self).answerClusterState(conn, state)
         self.app.refresh('state')
 
-    def answerPartitionTable(self, *args):
-        super(Handler, self).answerPartitionTable(*args)
-        self.app.refresh('pt')
-
     def sendPartitionTable(self, *args):
-        raise AssertionError
+        super(Handler, self).sendPartitionTable(*args)
+        self.app.refresh('pt')
 
     def notifyPartitionChanges(self, *args):
         super(Handler, self).notifyPartitionChanges(*args)
@@ -50,6 +47,7 @@ class StressApplication(AdminApplication):
 
     cluster_state = server = uuid = None
     listening_conn = True
+    fault_probability = 1
     restart_ratio = float('inf') # no firewall support
     _stress = False
 
@@ -191,7 +189,7 @@ class StressApplication(AdminApplication):
         self.loid = loid
         self.ltid = ltid
         self.em.setTimeout(int(time.time() + 1), self.askLastIDs)
-        if self._stress:
+        if self._stress and random.random() < self.fault_probability:
             node_list = self.nm.getStorageList()
             random.shuffle(node_list)
             fw = []

neo/tests/threaded/__init__.py

@@ -40,7 +40,7 @@ from neo.lib.util import cached_property, parseMasterList, p64
 from neo.master.recovery import  RecoveryManager
 from .. import (getTempDirectory, setupMySQLdb,
     ImporterConfigParser, NeoTestBase, Patch,
-    ADDRESS_TYPE, IP_VERSION_FORMAT_DICT, DB_PREFIX, DB_SOCKET, DB_USER)
+    ADDRESS_TYPE, IP_VERSION_FORMAT_DICT, DB_PREFIX)
 
 BIND = IP_VERSION_FORMAT_DICT[ADDRESS_TYPE], 0
 LOCAL_IP = socket.inet_pton(ADDRESS_TYPE, IP_VERSION_FORMAT_DICT[ADDRESS_TYPE])
@@ -321,7 +321,13 @@ class TestSerialized(Serialized):   # NOTE used only in .NeoCTL
 class Node(object):
 
     def getConnectionList(self, *peers):
-        addr = lambda c: c and (c.addr if c.is_server else c.getAddress())
+        def addr(c):
+            # Do not identify only by source address because 2 TCP connections
+            # can have same source host:port to different destinations.
+            if c:
+                a = c.addr
+                b = c.getAddress()
+                return (b, a) if c.is_server else (ServerNode.resolv(a), b)
         addr_set = {addr(c.connector) for peer in peers
             for c in peer.em.connection_dict.itervalues()
             if isinstance(c, Connection)}
@@ -395,7 +401,10 @@ class ServerNode(Node):
         assert not self.is_alive()
         init_args = self._init_args
         init_args['reset'] = False
-        assert set(kw).issubset(init_args), (kw, init_args)
+        if __debug__:
+            x = set(kw).difference(init_args)
+            assert not x or x.issubset(self.option_parser.getOptionDict()), (
+                kw, init_args)
         init_args.update(kw)
         self.close()
         self.__init__(**init_args)
@@ -730,7 +739,7 @@ class NEOCluster(object):
     def __init__(self, master_count=1, partitions=1, replicas=0, upstream=None,
                        adapter=os.getenv('NEO_TESTS_ADAPTER', 'SQLite'),
                        storage_count=None, db_list=None, clear_databases=True,
-                       db_user=DB_USER, db_password='', compress=True,
+                       compress=True,
                        importer=None, autostart=None, dedup=False, name=None):
         self.name = name or 'neo_%s' % self._allocate('name',
             lambda: random.randint(0, 100))
@@ -757,21 +766,20 @@ class NEOCluster(object):
             db_list = ['%s%u' % (DB_PREFIX, self._allocate('db', index))
                        for _ in xrange(storage_count)]
         if adapter == 'MySQL':
-            setupMySQLdb(db_list, db_user, db_password, clear_databases)
-            db = '%s:%s@%%s%s' % (db_user, db_password, DB_SOCKET)
+            db = setupMySQLdb(db_list, clear_databases)
         elif adapter == 'SQLite':
-            db = os.path.join(getTempDirectory(), '%s.sqlite')
+            db = os.path.join(getTempDirectory(), '%s.sqlite').__mod__
         else:
             assert False, adapter
         if importer:
             cfg = ImporterConfigParser(adapter, **importer)
-            cfg.set("neo", "database", db % tuple(db_list))
-            db = os.path.join(getTempDirectory(), '%s.conf')
-            with open(db % tuple(db_list), "w") as f:
+            cfg.set("neo", "database", db(*db_list))
+            db = os.path.join(getTempDirectory(), '%s.conf').__mod__
+            with open(db(*db_list), "w") as f:
                 cfg.write(f)
             kw["adapter"] = "Importer"
         kw['wait'] = 0
-        self.storage_list = [StorageApplication(database=db % x, **kw)
+        self.storage_list = [StorageApplication(database=db(x), **kw)
                              for x in db_list]
         self.admin_list = [AdminApplication(**kw)]
 
@@ -841,7 +849,7 @@ class NEOCluster(object):
             master_list = self.master_list
         if storage_list is None:
             storage_list = self.storage_list
-        def answerPartitionTable(release, orig, *args):
+        def sendPartitionTable(release, orig, *args):
             orig(*args)
             release()
         def dispatch(release, orig, handler, *args):
@@ -857,7 +865,7 @@ class NEOCluster(object):
             if state in expected_state:
                 release()
         with Serialized.until(MasterEventHandler,
-                answerPartitionTable=answerPartitionTable) as tic1, \
+                sendPartitionTable=sendPartitionTable) as tic1, \
              Serialized.until(RecoveryManager, dispatch=dispatch) as tic2, \
              Serialized.until(MasterEventHandler,
                 notifyClusterInformation=notifyClusterInformation) as tic3:
@@ -882,9 +890,13 @@ class NEOCluster(object):
         expected_state = (NodeStates.PENDING
             if state == ClusterStates.RECOVERING
             else NodeStates.RUNNING)
-        for node in self.storage_list if storage_list is None else storage_list:
+        for node, expected_state in (
+                storage_list if isinstance(storage_list, dict) else
+                dict.fromkeys(self.storage_list if storage_list is None else
+                              storage_list, expected_state)
+                ).iteritems():
             state = self.getNodeState(node)
-            assert state == expected_state, (repr(node), state)
+            assert state == expected_state, (repr(node), state, expected_state)
 
     def stop(self, clear_database=False, __print_exc=traceback.print_exc, **kw):
         if self.started:
@@ -958,7 +970,7 @@ class NEOCluster(object):
     def startCluster(self):
         try:
             self.neoctl.startCluster()
-        except RuntimeError:
+        except SystemExit:
             Serialized.tic()
             if self.neoctl.getClusterState() not in (
                       ClusterStates.BACKINGUP,
@@ -1037,18 +1049,18 @@ class NEOCluster(object):
         """Sort storages so that storage_list[i] has partition i for all i"""
         pt = [{x.getUUID() for x in x}
             for x in self.primary_master.pt.partition_list]
+        n = len(self.storage_list)
         r = []
         x = [iter(pt[0])]
-        try:
-            while 1:
-                try:
-                    r.append(next(x[-1]))
-                except StopIteration:
-                    del r[-1], x[-1]
-                else:
-                    x.append(iter(pt[len(r)].difference(r)))
-        except IndexError:
-            assert len(r) == len(self.storage_list)
+        while 1:
+            try:
+                r.append(next(x[-1]))
+            except StopIteration:
+                del r[-1], x[-1]
+            else:
+                if len(r) == n:
+                    break
+                x.append(iter(pt[len(r)].difference(r)))
         x = {x.uuid: x for x in self.storage_list}
         self.storage_list[:] = (x[r] for r in r)
         return self.storage_list

neo/tests/threaded/test.py

@@ -42,6 +42,7 @@ from neo.lib.util import add64, makeChecksum, p64, u64
 from neo.client.exception import NEOPrimaryMasterLost, NEOStorageError
 from neo.client.transactions import Transaction
 from neo.master.handlers.client import ClientServiceHandler
+from neo.master.pt import PartitionTable
 from neo.storage.database import DatabaseFailure
 from neo.storage.handlers.client import ClientOperationHandler
 from neo.storage.handlers.identification import IdentificationHandler
@@ -471,6 +472,7 @@ class Test(NEOThreadedTest):
             self.assertFalse(conn.isClosed())
             getCellSortKey = cluster.client.getCellSortKey
             self.assertEqual(getCellSortKey(s0, good), 0)
+            cluster.neoctl.killNode(s0.getUUID())
             cluster.neoctl.dropNode(s0.getUUID())
             self.assertEqual([s1], cluster.client.nm.getStorageList())
             self.assertTrue(conn.isClosed())
@@ -776,6 +778,7 @@ class Test(NEOThreadedTest):
             checkNodeState(NodeStates.RUNNING)
             self.assertEqual([], cluster.getOutdatedCells())
             # drop one
+            cluster.neoctl.killNode(s1.uuid)
             cluster.neoctl.dropNode(s1.uuid)
             checkNodeState(None)
             self.tic() # Let node state update reach remaining storage
@@ -1055,6 +1058,10 @@ class Test(NEOThreadedTest):
                 # Check that the storage hasn't answered to the store,
                 # which means that a lock is still taken for r['x'] by t2.
                 self.tic()
+                try:
+                    txn = txn.data(c1)
+                except (AttributeError, KeyError): # BBB: ZODB < 5
+                    pass
                 txn_context = cluster.client._txn_container.get(txn)
                 empty = txn_context.queue.empty()
                 ll()
@@ -1303,7 +1310,7 @@ class Test(NEOThreadedTest):
             del conn._queue[:] # XXX
             conn.close()
         if 1:
-            with Patch(cluster.master.pt, make=make), \
+            with Patch(PartitionTable, make=make), \
                  Patch(InitializationHandler,
                        askPartitionTable=askPartitionTable) as p:
                 cluster.start()
@@ -1834,18 +1841,7 @@ class Test(NEOThreadedTest):
                     x.value += 1
                     c2.root()['x'].value += 2
                     TransactionalResource(t1, 1, tpc_begin=begin1)
-                    # BUG: Very rarely, getConnectionList returns more that 1
-                    #      connection ("too many values to unpack"), which is
-                    #       a mystery and impossible to reproduce:
-                    #      - 1st time: v1.8.1 on a test machine (no SSL)
-                    #      - last: current revision on my laptop (SSL),
-                    #              at the first iteration of this loop
-                    _sm = list(s1.getConnectionList(cluster.master))
-                    try:
-                        s1m, = _sm
-                    except ValueError:
-                        self.fail((_sm, list(
-                            s1.getConnectionList(cluster.master))))
+                    s1m, = s1.getConnectionList(cluster.master)
                     try:
                         s1.em.removeReader(s1m)
                         with ConnectionFilter() as f, \
@@ -2293,6 +2289,10 @@ class Test(NEOThreadedTest):
                     # Check that the storage hasn't answered to the store,
                     # which means that a lock is still taken for r[''] by t1.
                     self.tic()
+                    try:
+                        txn = txn.data(c3)
+                    except (AttributeError, KeyError): # BBB: ZODB < 5
+                        pass
                     txn_context = db.storage.app._txn_container.get(txn)
                     raise Abort(txn_context.queue.empty())
                 TransactionalResource(t3, 1, commit=t3_commit)
@@ -2339,8 +2339,8 @@ class Test(NEOThreadedTest):
         for x in 'ab':
             r[x] = PCounterWithResolution()
             t1.commit()
-        cluster.stop(replicas=1)
-        cluster.start()
+        cluster.neoctl.setNumReplicas(1)
+        self.tic()
         s0, s1 = cluster.sortStorageList()
         t1, c1 = cluster.getTransaction()
         r = c1.root()
@@ -2524,8 +2524,8 @@ class Test(NEOThreadedTest):
         for x in 'ab':
             r[x] = PCounterWithResolution()
             t1.commit()
-        cluster.stop(replicas=1)
-        cluster.start()
+        cluster.neoctl.setNumReplicas(1)
+        self.tic()
         s0, s1 = cluster.sortStorageList()
         t1, c1 = cluster.getTransaction()
         r = c1.root()
@@ -2826,9 +2826,9 @@ class Test(NEOThreadedTest):
             dm = s.dm
             dm.commit()
             dump_dict[s.uuid] = dm.dump()
-            dm.erase()
             with open(path % (s.getAdapter(), s.uuid)) as f:
                 dm.restore(f.read())
+            dm.setConfiguration('partitions', None) # XXX: see dm._migrate4
         with NEOCluster(storage_count=3, partitions=3, replicas=1,
                         name=self._testMethodName) as cluster:
             s1, s2, s3 = cluster.storage_list

neo/tests/zodb/testRecovery.py

@@ -33,8 +33,6 @@ class RecoveryTests(ZODBTestCase, StorageTestBase, RecoveryStorage):
             os.makedirs(dst_temp_dir)
         self.neo_dst = NEOCluster(['test_neo1-dst'], partitions=1, replicas=0,
                 master_count=1, temp_dir=dst_temp_dir)
-        self.neo_dst.stop()
-        self.neo_dst.setupDB()
         self.neo_dst.start()
         self._dst = self.neo.getZODBStorage()
         self._dst_db = ZODB.DB(self._dst)

setup.py

@@ -78,7 +78,7 @@ else:
 
 setup(
     name = 'neoppod',
-    version = '1.11',
+    version = '1.12.0',
     description = __doc__.strip(),
     author = 'Nexedi SA',
     author_email = 'neo-dev@erp5.org',

tools/matrix

@@ -129,7 +129,7 @@ class MatrixImportBenchmark(BenchmarkRunner):
             finally:
                 zodb.stop()
             # Clear DB if no error happened.
-            zodb.setupDB()
+            zodb.resetDB()
             return end - start
         except:
             traceback.print_exc()

tools/perfs

@@ -53,7 +53,7 @@ class ImportBenchmark(BenchmarkRunner):
             finally:
                 neo.stop()
             # Clear DB if no error happened.
-            neo.setupDB()
+            neo.resetDB()
             return result
         except:
             return 'Perf: import failed', ''.join(traceback.format_exc())