Skip to content

N
neoppod
Commit d4be13da authored by Kirill Smelkov's avatar Kirill Smelkov
Browse files
Options

.

parent b978935d
Hide whitespace changes
Inline Side-by-side
Showing with 868 additions and 852 deletions
t/neo/connection.go
View file @ d4be13da
...@@ -10,7 +10,7 @@ ...@@ -10,7 +10,7 @@
// //
// See COPYING file for full licensing terms. // See COPYING file for full licensing terms.
// NEO | Connection management // NEO. Connection management
package neo package neo
...@@ -20,8 +20,6 @@ import ( ...@@ -20,8 +20,6 @@ import (
"net" "net"
"sync" "sync"
"unsafe" "unsafe"
//"fmt"
) )
// NodeLink is a node-node link in NEO // NodeLink is a node-node link in NEO
...@@ -48,8 +46,8 @@ import ( ...@@ -48,8 +46,8 @@ import (
type NodeLink struct { type NodeLink struct {
peerLink net.Conn // raw conn to peer peerLink net.Conn // raw conn to peer
connMu sync.Mutex // TODO -> RW ? connMu sync.Mutex // TODO -> RW ?
connTab map[uint32]*Conn // connId -> Conn associated with connId connTab map[uint32]*Conn // connId -> Conn associated with connId
nextConnId uint32 // next connId to use for Conn initiated by us nextConnId uint32 // next connId to use for Conn initiated by us
serveWg sync.WaitGroup // for serve{Send,Recv} serveWg sync.WaitGroup // for serve{Send,Recv}
...@@ -72,43 +70,28 @@ type Conn struct { ...@@ -72,43 +70,28 @@ type Conn struct {
rxq chan *PktBuf // received packets for this Conn go here rxq chan *PktBuf // received packets for this Conn go here
txerr chan error // transmit errors for this Conn go back here txerr chan error // transmit errors for this Conn go back here
// Conn has to be explicitly closed by user; it can also be closed by NodeLink.Close // once because: Conn has to be explicitly closed by user; it can also
// be closed by NodeLink.Close .
closeOnce sync.Once closeOnce sync.Once
closed chan struct{} closed chan struct{}
} }
// Buffer with packet data // A role our end of NodeLink is intended to play
// XXX move me out of here type LinkRole int
type PktBuf struct {
Data []byte // whole packet data including all headers XXX -> Buf ?
}
// Get pointer to packet header
func (pkt *PktBuf) Header() *PktHead {
// XXX check len(Data) < PktHead ? -> no, Data has to be allocated with cap >= PktHeadLen
return (*PktHead)(unsafe.Pointer(&pkt.Data[0]))
}
// Get packet payload
func (pkt *PktBuf) Payload() []byte {
return pkt.Data[PktHeadLen:]
}
type ConnRole int
const ( const (
ConnServer ConnRole = iota // connection created as server LinkServer ConnRole = iota // link created as server
ConnClient // connection created as client LinkClient // link created as client
// for testing: // for testing:
connNoRecvSend ConnRole = 1<<16 // do not spawn serveRecv & serveSend linkNoRecvSend ConnRole = 1<<16 // do not spawn serveRecv & serveSend
connFlagsMask ConnRole = (1<<32 - 1) << 16 linkFlagsMask ConnRole = (1<<32 - 1) << 16
) )
// Make a new NodeLink from already established net.Conn // Make a new NodeLink from already established net.Conn
// //
// role specifies how to treat conn - either as server or client one. // role specifies how to treat conn - either as server or client one.
// The difference in between client and server roles are in connid % 2 XXX text // The difference in between client and server roles is only in how connection
// ids are allocated for connections initiated at our side: there is no overlap in identifiers if one side always allocates them as even and its peer as odd. in connId % 2 XXX text
// //
// Usually server role should be used for connections created via // Usually server role should be used for connections created via
// net.Listen/net.Accept and client role for connections created via net.Dial. // net.Listen/net.Accept and client role for connections created via net.Dial.
......
t/neo/connection_test.go
View file @ d4be13da
...@@ -15,12 +15,9 @@ ...@@ -15,12 +15,9 @@
package neo package neo
import ( import (
//"fmt"
"bytes" "bytes"
"context" "context"
"io" "io"
//"fmt"
"net" "net"
"testing" "testing"
"time" "time"
...@@ -28,7 +25,6 @@ import ( ...@@ -28,7 +25,6 @@ import (
"golang.org/x/sync/errgroup" "golang.org/x/sync/errgroup"
"lab.nexedi.com/kirr/go123/exc" "lab.nexedi.com/kirr/go123/exc"
//"lab.nexedi.com/kirr/go123/myname"
"lab.nexedi.com/kirr/go123/xerr" "lab.nexedi.com/kirr/go123/xerr"
) )
...@@ -118,7 +114,7 @@ func xverifyPkt(pkt *PktBuf, connid uint32, msgcode uint16, payload []byte) { ...@@ -118,7 +114,7 @@ func xverifyPkt(pkt *PktBuf, connid uint32, msgcode uint16, payload []byte) {
errv.Appendf("header: unexpected length %v (want %v)", ntoh32(h.Len), PktHeadLen + len(payload)) errv.Appendf("header: unexpected length %v (want %v)", ntoh32(h.Len), PktHeadLen + len(payload))
} }
if !bytes.Equal(pkt.Payload(), payload) { if !bytes.Equal(pkt.Payload(), payload) {
errv.Appendf("payload differ") // XXX also print payload ? errv.Appendf("payload differ")
} }
exc.Raiseif( errv.Err() ) exc.Raiseif( errv.Err() )
...@@ -131,13 +127,14 @@ func tdelay() { ...@@ -131,13 +127,14 @@ func tdelay() {
time.Sleep(1*time.Millisecond) time.Sleep(1*time.Millisecond)
} }
// create NodeLinks connected via net.Pipe
func _nodeLinkPipe(flags1, flags2 ConnRole) (nl1, nl2 *NodeLink) { func _nodeLinkPipe(flags1, flags2 ConnRole) (nl1, nl2 *NodeLink) {
node1, node2 := net.Pipe() node1, node2 := net.Pipe()
nl1 = NewNodeLink(node1, ConnClient | flags1) nl1 = NewNodeLink(node1, ConnClient | flags1)
nl2 = NewNodeLink(node2, ConnServer | flags2) nl2 = NewNodeLink(node2, ConnServer | flags2)
return nl1, nl2 return nl1, nl2
} }
// create NodeLinks connected via net.Pipe
func nodeLinkPipe() (nl1, nl2 *NodeLink) { func nodeLinkPipe() (nl1, nl2 *NodeLink) {
return _nodeLinkPipe(0, 0) return _nodeLinkPipe(0, 0)
} }
...@@ -174,7 +171,7 @@ func TestNodeLink(t *testing.T) { ...@@ -174,7 +171,7 @@ func TestNodeLink(t *testing.T) {
xwait(wg) xwait(wg)
xclose(nl2) xclose(nl2)
// check raw exchange works // raw exchange
nl1, nl2 = _nodeLinkPipe(connNoRecvSend, connNoRecvSend) nl1, nl2 = _nodeLinkPipe(connNoRecvSend, connNoRecvSend)
wg, ctx := WorkGroupCtx(context.Background()) wg, ctx := WorkGroupCtx(context.Background())
...@@ -286,10 +283,10 @@ func TestNodeLink(t *testing.T) { ...@@ -286,10 +283,10 @@ func TestNodeLink(t *testing.T) {
xsend(c, mkpkt(35, []byte("ping2"))) xsend(c, mkpkt(35, []byte("ping2")))
pkt = xrecv(c) pkt = xrecv(c)
xverifyPkt(pkt, c.connId, 36, []byte("pong2")) xverifyPkt(pkt, c.connId, 36, []byte("pong2"))
nl2.Wait()
xclose(c) xclose(c)
xclose(nl1) xclose(nl1)
nl2.Wait()
xclose(nl2) xclose(nl2)
// test 2 channels with replies comming in reversed time order // test 2 channels with replies comming in reversed time order
...@@ -304,6 +301,7 @@ func TestNodeLink(t *testing.T) { ...@@ -304,6 +301,7 @@ func TestNodeLink(t *testing.T) {
close(replyOrder[2].start) close(replyOrder[2].start)
nl2.HandleNewConn(func(c *Conn) { nl2.HandleNewConn(func(c *Conn) {
// TODO raised err -> errch
pkt := xrecv(c) pkt := xrecv(c)
n := ntoh16(pkt.Header().MsgCode) n := ntoh16(pkt.Header().MsgCode)
x := replyOrder[n] x := replyOrder[n]
...@@ -332,10 +330,10 @@ func TestNodeLink(t *testing.T) { ...@@ -332,10 +330,10 @@ func TestNodeLink(t *testing.T) {
} }
xechoWait(c2, 2) xechoWait(c2, 2)
xechoWait(c1, 1) xechoWait(c1, 1)
nl2.Wait()
xclose(c1) xclose(c1)
xclose(c2) xclose(c2)
xclose(nl1) xclose(nl1)
nl2.Wait()
xclose(nl2) xclose(nl2)
} }
t/neo/pkt.go
View file @ d4be13da
package neo // Copyright (C) 2016 Nexedi SA and Contributors.
//package proto // Kirill Smelkov <kirr@nexedi.com>
//
/* // This program is free software: you can Use, Study, Modify and Redistribute
import ( // it under the terms of the GNU General Public License version 2, or (at your
. "../" // option) any later version, as published by the Free Software Foundation.
) //
*/ // This program is distributed WITHOUT ANY WARRANTY; without even the implied
// warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
const ( //
PROTOCOL_VERSION = 8 // See COPYING file for full licensing terms.
MIN_PACKET_SIZE = 10 // XXX link this to len(pkthead) ? // NEO. Packets and packet buffers management
PktHeadLen = MIN_PACKET_SIZE // TODO link this to PktHead.Encode/Decode size
MAX_PACKET_SIZE = 0x4000000
RESPONSE_MASK = 0x800
)
type ErrorCode int
const (
ACK ErrorCode = iota
NOT_READY
OID_NOT_FOUND
TID_NOT_FOUND
OID_DOES_NOT_EXIST
PROTOCOL_ERROR
BROKEN_NODE
ALREADY_PENDING
REPLICATION_ERROR
CHECKING_ERROR
BACKEND_NOT_IMPLEMENTED
READ_ONLY_ACCESS
)
type ClusterState int
const (
// NOTE cluster states descriptions is in protocol.py
RECOVERING ClusterState = iota
VERIFYING
CLUSTER_RUNNING // XXX conflict with NodeState.RUNNING
STOPPING
STARTING_BACKUP
BACKINGUP
STOPPING_BACKUP
)
type NodeType int
const (
MASTER NodeType = iota
STORAGE
CLIENT
ADMIN
)
type NodeState int
const (
RUNNING NodeState = iota //short: R // XXX tag prefix name ?
TEMPORARILY_DOWN //short: T
DOWN //short: D
BROKEN //short: B
HIDDEN //short: H
PENDING //short: P
UNKNOWN //short: U
)
type CellState int
const (
// NOTE cell states description is in protocol.py
UP_TO_DATE CellState = iota //short: U // XXX tag prefix name ?
OUT_OF_DATE //short: O
FEEDING //short: F
DISCARDED //short: D
CORRUPTED //short: C
)
type LockState int
const (
NOT_LOCKED LockState = iota
GRANTED
GRANTED_TO_OTHER
)
// An UUID (node identifier, 4-bytes signed integer)
type UUID int32
// TODO UUID_NAMESPACES
type Address struct {
Host string
Port uint16 // TODO if Host == 0 -> Port not added to wire (see py.PAddress)
}
// A SHA1 hash
type Checksum [20]byte
// Partition Table identifier
// Zero value means "invalid id" (<-> None in py.PPTID)
type PTid uint64 // XXX move to common place ?
// TODO None encodes as '\xff' * 8 (XXX use nan for None ?)
type Float64 float64
// NOTE original NodeList = []NodeInfo
type NodeInfo struct {
NodeType
Address
UUID
NodeState
IdTimestamp Float64
}
//type CellList []struct {
type CellInfo struct {
UUID UUID // XXX maybe simply 'UUID' ?
CellState CellState // ----///----
}
//type RowList []struct {
type RowInfo struct {
Offset uint32 // PNumber
CellList []CellInfo
}
// XXX link request <-> answer ?
// TODO ensure len(encoded packet header) == 10
type PktHead struct {
ConnId be32 // NOTE is .msgid in py
MsgCode be16
Len be32 // whole packet length (including header)
}
// TODO generate .Encode() / .Decode()
// General purpose notification (remote logging)
type Notify struct {
PktHead
Message string
}
// Error is a special type of message, because this can be sent against
// any other message, even if such a message does not expect a reply
// usually. Any -> Any.
type Error struct {
PktHead
Code uint32 // PNumber
Message string
}
// Check if a peer is still alive. Any -> Any.
type Ping struct {
PktHead
// TODO _answer = PFEmpty
}
// Tell peer it can close the connection if it has finished with us. Any -> Any
type CloseClient struct {
PktHead
}
// Request a node identification. This must be the first packet for any
// connection. Any -> Any.
type RequestIdentification struct {
PktHead
ProtocolVersion uint32 // TODO py.PProtocol upon decoding checks for != PROTOCOL_VERSION
NodeType NodeType // XXX name
UUID UUID
Address
Name string
IdTimestamp Float64
}
// XXX -> ReplyIdentification? RequestIdentification.Answer somehow ?
type AcceptIdentification struct {
PktHead
NodeType NodeType // XXX name
MyUUID UUID
NumPartitions uint32 // PNumber
NumReplicas uint32 // PNumber
YourUUID UUID
Primary Address // TODO
KnownMasterList []struct {
Address
UUID UUID
}
}
// Ask current primary master's uuid. CTL -> A.
type PrimaryMaster struct {
PktHead
}
type AnswerPrimary struct {
PktHead
PrimaryUUID UUID
}
// Announce a primary master node election. PM -> SM.
type AnnouncePrimary struct {
PktHead
}
// Force a re-election of a primary master node. M -> M.
type ReelectPrimary struct {
PktHead
}
// Ask all data needed by master to recover. PM -> S, S -> PM.
type Recovery struct {
PktHead
}
type AnswerRecovery struct {
PktHead
PTid
BackupTID Tid
TruncateTID Tid
}
// Ask the last OID/TID so that a master can initialize its TransactionManager.
// PM -> S, S -> PM.
type LastIDs struct {
PktHead
}
type AnswerLastIDs struct {
PktHead
LastOID Oid
LastTID Tid
}
// Ask the full partition table. PM -> S.
// Answer rows in a partition table. S -> PM.
type PartitionTable struct {
PktHead
}
type AnswerPartitionTable struct {
PktHead
PTid
RowList []RowInfo
}
// Send rows in a partition table to update other nodes. PM -> S, C.
type NotifyPartitionTable struct {
PktHead
PTid
RowList []RowInfo
}
// Notify a subset of a partition table. This is used to notify changes.
// PM -> S, C.
type PartitionChanges struct {
PktHead
PTid
CellList []struct {
// XXX does below correlate with Cell inside top-level CellList ?
Offset uint32 // PNumber
UUID UUID
CellState CellState
}
}
// Tell a storage nodes to start an operation. Until a storage node receives
// this message, it must not serve client nodes. PM -> S.
type StartOperation struct {
PktHead
// XXX: Is this boolean needed ? Maybe this
// can be deduced from cluster state.
Backup bool
}
// Tell a storage node to stop an operation. Once a storage node receives
// this message, it must not serve client nodes. PM -> S.
type StopOperation struct {
PktHead
}
// Ask unfinished transactions S -> PM.
// Answer unfinished transactions PM -> S.
type UnfinishedTransactions struct {
PktHead
}
type AnswerUnfinishedTransactions struct {
PktHead
MaxTID Tid
TidList []struct{
UnfinishedTID Tid
}
}
// Ask locked transactions PM -> S.
// Answer locked transactions S -> PM.
type LockedTransactions struct {
PktHead
}
type AnswerLockedTransactions struct {
PktHead
TidDict map[Tid]Tid // ttid -> tid
}
// Return final tid if ttid has been committed. * -> S. C -> PM.
type FinalTID struct {
PktHead
TTID Tid
}
type AnswerFinalTID struct {
PktHead
Tid Tid
}
// Commit a transaction. PM -> S.
type ValidateTransaction struct {
PktHead
TTID Tid
Tid Tid
}
// Ask to begin a new transaction. C -> PM.
// Answer when a transaction begin, give a TID if necessary. PM -> C.
type BeginTransaction struct {
PktHead
Tid Tid
}
type AnswerBeginTransaction struct {
PktHead
Tid Tid
}
// Finish a transaction. C -> PM.
// Answer when a transaction is finished. PM -> C.
type FinishTransaction struct {
PktHead
Tid Tid
OIDList []Oid
CheckedList []Oid
}
type AnswerFinishTransaction struct {
PktHead
TTID Tid
Tid Tid
}
// Notify that a transaction blocking a replication is now finished
// M -> S
type NotifyTransactionFinished struct {
PktHead
TTID Tid
MaxTID Tid
}
// Lock information on a transaction. PM -> S.
// Notify information on a transaction locked. S -> PM.
type LockInformation struct {
PktHead
Ttid Tid
Tid Tid
}
// XXX AnswerInformationLocked ?
type AnswerLockInformation struct {
Ttid Tid
}
// Invalidate objects. PM -> C.
// XXX ask_finish_transaction ?
type InvalidateObjects struct {
PktHead
Tid Tid
OidList []Oid
}
// Unlock information on a transaction. PM -> S.
type UnlockInformation struct {
PktHead
TTID Tid
}
// Ask new object IDs. C -> PM.
// Answer new object IDs. PM -> C.
type GenerateOIDs struct {
PktHead
NumOIDs uint32 // PNumber
}
// XXX answer_new_oids ?
type AnswerGenerateOIDs struct {
PktHead
OidList []Oid
}
// Ask to store an object. Send an OID, an original serial, a current
// transaction ID, and data. C -> S.
// Answer if an object has been stored. If an object is in conflict,
// a serial of the conflicting transaction is returned. In this case,
// if this serial is newer than the current transaction ID, a client
// node must not try to resolve the conflict. S -> C.
type StoreObject struct {
PktHead
Oid Oid
Serial Tid
Compression bool
Checksum Checksum
Data []byte // XXX or string ?
DataSerial Tid
Tid Tid
Unlock bool
}
type AnswerStoreObject struct {
PktHead
Conflicting bool
Oid Oid
Serial Tid
}
// Abort a transaction. C -> S, PM.
type AbortTransaction struct {
PktHead
Tid Tid
}
// Ask to store a transaction. C -> S.
// Answer if transaction has been stored. S -> C.
type StoreTransaction struct {
PktHead
Tid Tid
User string
Description string
Extension string
OidList []Oid
// TODO _answer = PFEmpty
}
// Ask to store a transaction. C -> S.
// Answer if transaction has been stored. S -> C.
type VoteTransaction struct {
PktHead
Tid Tid
// TODO _answer = PFEmpty
}
// Ask a stored object by its OID and a serial or a TID if given. If a serial
// is specified, the specified revision of an object will be returned. If
// a TID is specified, an object right before the TID will be returned. C -> S.
// Answer the requested object. S -> C.
type GetObject struct {
PktHead
Oid Oid
Serial Tid
Tid Tid
}
// XXX answer_object ?
type AnswerGetObject struct {
PktHead
Oid Oid
SerialStart Tid
SerialEnd Tid
Compression bool
Checksum Checksum
Data []byte // XXX or string ?
DataSerial Tid
}
// Ask for TIDs between a range of offsets. The order of TIDs is descending,
// and the range is [first, last). C -> S.
// Answer the requested TIDs. S -> C.
type TIDList struct {
PktHead
First uint64 // PIndex XXX this is TID actually ? -> no it is offset in list
Last uint64 // PIndex ----//----
Partition uint32 // PNumber
}
// XXX answer_tids ?
type AnswerTIDList struct {
PktHead
TIDList []Tid
}
// Ask for length TIDs starting at min_tid. The order of TIDs is ascending.
// C -> S.
// Answer the requested TIDs. S -> C
type TIDListFrom struct {
PktHead
MinTID Tid
MaxTID Tid
Length uint32 // PNumber
Partition uint32 // PNumber
}
// XXX answer_tids ?
type AnswerTIDListFrom struct {
PktHead
TidList []Tid
}
// Ask information about a transaction. Any -> S.
// Answer information (user, description) about a transaction. S -> Any.
type TransactionInformation struct {
PktHead
Tid Tid
}
type AnswerTransactionInformation struct {
PktHead
Tid Tid
User string
Description string
Extension string
Packed bool
OidList []Oid
}
// Ask history information for a given object. The order of serials is
// descending, and the range is [first, last]. C -> S.
// Answer history information (serial, size) for an object. S -> C.
type ObjectHistory struct {
PktHead
Oid Oid
First uint64 // PIndex XXX this is actually TID
Last uint64 // PIndex ----//----
}
type AnswerObjectHistory struct {
PktHead
Oid Oid
HistoryList []struct {
Serial Tid
Size uint32 // PNumber
}
}
// All the following messages are for neoctl to admin node
// Ask information about partition
// Answer information about partition
type PartitionList struct {
PktHead
MinOffset uint32 // PNumber
MaxOffset uint32 // PNumber
UUID UUID
}
type AnswerPartitionList struct {
PktHead
PTid
RowList []RowInfo
}
// Ask information about nodes
// Answer information about nodes
type X_NodeList struct {
PktHead
NodeType
}
type AnswerNodeList struct {
PktHead
NodeList []NodeInfo
}
// Set the node state
type SetNodeState struct {
PktHead
UUID
NodeState
// XXX _answer = Error ?
}
// Ask the primary to include some pending node in the partition table
type AddPendingNodes struct {
PktHead
UUIDList []UUID
// XXX _answer = Error
}
// Ask the primary to optimize the partition table. A -> PM.
type TweakPartitionTable struct {
PktHead
UUIDList []UUID
// XXX _answer = Error
}
// Notify information about one or more nodes. PM -> Any.
type NotifyNodeInformation struct {
PktHead
NodeList []NodeInfo
}
// Ask node information
type NodeInformation struct {
PktHead
// XXX _answer = PFEmpty
}
// Set the cluster state
type SetClusterState struct {
PktHead
State ClusterState
// XXX _answer = Error
}
// Notify information about the cluster
type ClusterInformation struct {
PktHead
State ClusterState
}
// Ask state of the cluster
// Answer state of the cluster
type X_ClusterState struct { // XXX conflicts with ClusterState enum
PktHead
State ClusterState
}
// Ask storage the serial where object data is when undoing given transaction,
// for a list of OIDs.
// C -> S
// Answer serials at which object data is when undoing a given transaction.
// object_tid_dict has the following format:
// key: oid
// value: 3-tuple
// current_serial (TID)
// The latest serial visible to the undoing transaction.
// undo_serial (TID)
// Where undone data is (tid at which data is before given undo).
// is_current (bool)
// If current_serial's data is current on storage.
// S -> C
type ObjectUndoSerial struct {
PktHead
Tid Tid
LTID Tid
UndoneTID Tid
OidList []Oid
}
// XXX answer_undo_transaction ?
type AnswerObjectUndoSerial struct {
PktHead
ObjectTIDDict map[Oid]struct {
CurrentSerial Tid
UndoSerial Tid
IsCurrent bool
}
}
// Ask a storage is oid is locked by another transaction.
// C -> S
// Answer whether a transaction holds the write lock for requested object.
type HasLock struct {
PktHead
Tid Tid
Oid Oid
}
type AnswerHasLock struct {
PktHead
Oid Oid
LockState LockState
}
// Verifies if given serial is current for object oid in the database, and
// take a write lock on it (so that this state is not altered until
// transaction ends).
// Answer to AskCheckCurrentSerial.
// Same structure as AnswerStoreObject, to handle the same way, except there
// is nothing to invalidate in any client's cache.
type CheckCurrentSerial struct {
PktHead
Tid Tid
Serial Tid
Oid Oid
}
// XXX answer_store_object ?
type AnswerCheckCurrentSerial struct {
Conflicting bool
Oid Oid
Serial Tid
}
// Request a pack at given TID.
// C -> M
// M -> S
// Inform that packing it over.
// S -> M
// M -> C
type Pack struct {
PktHead
Tid Tid
}
type AnswerPack struct {
PktHead
Status bool
}
// ctl -> A
// A -> M
type CheckReplicas struct {
PktHead
PartitionDict map[uint32/*PNumber*/]UUID // partition -> source
MinTID Tid
MaxTID Tid
// XXX _answer = Error
}
// M -> S
type CheckPartition struct {
PktHead
Partition uint32 // PNumber
Source struct {
UpstreamName string
Address Address
}
MinTID Tid
MaxTID Tid
}
// Ask some stats about a range of transactions.
// Used to know if there are differences between a replicating node and
// reference node.
// S -> S
// Stats about a range of transactions.
// Used to know if there are differences between a replicating node and
// reference node.
// S -> S
type CheckTIDRange struct {
PktHead
Partition uint32 // PNumber
Length uint32 // PNumber
MinTID Tid
MaxTID Tid
}
type AnswerCheckTIDRange struct {
PktHead
Count uint32 // PNumber
Checksum Checksum
MaxTID Tid
}
// Ask some stats about a range of object history.
// Used to know if there are differences between a replicating node and
// reference node.
// S -> S
// Stats about a range of object history.
// Used to know if there are differences between a replicating node and
// reference node.
// S -> S
type CheckSerialRange struct {
PktHead
Partition uint32 // PNumber
Length uint32 // PNumber
MinTID Tid
MaxTID Tid
MinOID Oid
}
type AnswerCheckSerialRange struct {
Count uint32 // PNumber
TidChecksum Checksum
MaxTID Tid
OidChecksum Checksum
MaxOID Oid
}
// S -> M
type PartitionCorrupted struct {
PktHead
Partition uint32 // PNumber
CellList []UUID
}
package neo
// Ask last committed TID. // Buffer with packet data
// C -> M type PktBuf struct {
// Answer last committed TID. Data []byte // whole packet data including all headers XXX -> Buf ?
// M -> C
type LastTransaction struct {
PktHead
} }
type AnswerLastTransaction struct { // Get pointer to packet header
PktHead func (pkt *PktBuf) Header() *PktHead {
Tid Tid // XXX check len(Data) < PktHead ? -> no, Data has to be allocated with cap >= PktHeadLen
return (*PktHead)(unsafe.Pointer(&pkt.Data[0]))
} }
// Get packet payload
// Notify that node is ready to serve requests. func (pkt *PktBuf) Payload() []byte {
// S -> M return pkt.Data[PktHeadLen:]
type NotifyReady struct {
PktHead
} }
// replication
// TODO
t/neo/proto.go 0 → 100644
View file @ d4be13da
// NEO. Protocol description
package neo
//package proto
/*
import (
. "../"
)
*/
const (
PROTOCOL_VERSION = 8
MIN_PACKET_SIZE = 10 // XXX link this to len(pkthead) ?
PktHeadLen = MIN_PACKET_SIZE // TODO link this to PktHead.Encode/Decode size
MAX_PACKET_SIZE = 0x4000000
RESPONSE_MASK = 0x800
)
type ErrorCode int
const (
ACK ErrorCode = iota
NOT_READY
OID_NOT_FOUND
TID_NOT_FOUND
OID_DOES_NOT_EXIST
PROTOCOL_ERROR
BROKEN_NODE
ALREADY_PENDING
REPLICATION_ERROR
CHECKING_ERROR
BACKEND_NOT_IMPLEMENTED
READ_ONLY_ACCESS
)
type ClusterState int
const (
// NOTE cluster states descriptions is in protocol.py
RECOVERING ClusterState = iota
VERIFYING
CLUSTER_RUNNING // XXX conflict with NodeState.RUNNING
STOPPING
STARTING_BACKUP
BACKINGUP
STOPPING_BACKUP
)
type NodeType int
const (
MASTER NodeType = iota
STORAGE
CLIENT
ADMIN
)
type NodeState int
const (
RUNNING NodeState = iota //short: R // XXX tag prefix name ?
TEMPORARILY_DOWN //short: T
DOWN //short: D
BROKEN //short: B
HIDDEN //short: H
PENDING //short: P
UNKNOWN //short: U
)
type CellState int
const (
// NOTE cell states description is in protocol.py
UP_TO_DATE CellState = iota //short: U // XXX tag prefix name ?
OUT_OF_DATE //short: O
FEEDING //short: F
DISCARDED //short: D
CORRUPTED //short: C
)
type LockState int
const (
NOT_LOCKED LockState = iota
GRANTED
GRANTED_TO_OTHER
)
// An UUID (node identifier, 4-bytes signed integer)
type UUID int32
// TODO UUID_NAMESPACES
type Address struct {
Host string
Port uint16 // TODO if Host == 0 -> Port not added to wire (see py.PAddress)
}
// A SHA1 hash
type Checksum [20]byte
// Partition Table identifier
// Zero value means "invalid id" (<-> None in py.PPTID)
type PTid uint64 // XXX move to common place ?
// TODO None encodes as '\xff' * 8 (XXX use nan for None ?)
type Float64 float64
// NOTE original NodeList = []NodeInfo
type NodeInfo struct {
NodeType
Address
UUID
NodeState
IdTimestamp Float64
}
//type CellList []struct {
type CellInfo struct {
UUID UUID // XXX maybe simply 'UUID' ?
CellState CellState // ----///----
}
//type RowList []struct {
type RowInfo struct {
Offset uint32 // PNumber
CellList []CellInfo
}
// XXX link request <-> answer ?
// TODO ensure len(encoded packet header) == 10
type PktHead struct {
ConnId be32 // NOTE is .msgid in py
MsgCode be16
Len be32 // whole packet length (including header)
}
// TODO generate .Encode() / .Decode()
// General purpose notification (remote logging)
type Notify struct {
PktHead
Message string
}
// Error is a special type of message, because this can be sent against
// any other message, even if such a message does not expect a reply
// usually. Any -> Any.
type Error struct {
PktHead
Code uint32 // PNumber
Message string
}
// Check if a peer is still alive. Any -> Any.
type Ping struct {
PktHead
// TODO _answer = PFEmpty
}
// Tell peer it can close the connection if it has finished with us. Any -> Any
type CloseClient struct {
PktHead
}
// Request a node identification. This must be the first packet for any
// connection. Any -> Any.
type RequestIdentification struct {
PktHead
ProtocolVersion uint32 // TODO py.PProtocol upon decoding checks for != PROTOCOL_VERSION
NodeType NodeType // XXX name
UUID UUID
Address
Name string
IdTimestamp Float64
}
// XXX -> ReplyIdentification? RequestIdentification.Answer somehow ?
type AcceptIdentification struct {
PktHead
NodeType NodeType // XXX name
MyUUID UUID
NumPartitions uint32 // PNumber
NumReplicas uint32 // PNumber
YourUUID UUID
Primary Address // TODO
KnownMasterList []struct {
Address
UUID UUID
}
}
// Ask current primary master's uuid. CTL -> A.
type PrimaryMaster struct {
PktHead
}
type AnswerPrimary struct {
PktHead
PrimaryUUID UUID
}
// Announce a primary master node election. PM -> SM.
type AnnouncePrimary struct {
PktHead
}
// Force a re-election of a primary master node. M -> M.
type ReelectPrimary struct {
PktHead
}
// Ask all data needed by master to recover. PM -> S, S -> PM.
type Recovery struct {
PktHead
}
type AnswerRecovery struct {
PktHead
PTid
BackupTID Tid
TruncateTID Tid
}
// Ask the last OID/TID so that a master can initialize its TransactionManager.
// PM -> S, S -> PM.
type LastIDs struct {
PktHead
}
type AnswerLastIDs struct {
PktHead
LastOID Oid
LastTID Tid
}
// Ask the full partition table. PM -> S.
// Answer rows in a partition table. S -> PM.
type PartitionTable struct {
PktHead
}
type AnswerPartitionTable struct {
PktHead
PTid
RowList []RowInfo
}
// Send rows in a partition table to update other nodes. PM -> S, C.
type NotifyPartitionTable struct {
PktHead
PTid
RowList []RowInfo
}
// Notify a subset of a partition table. This is used to notify changes.
// PM -> S, C.
type PartitionChanges struct {
PktHead
PTid
CellList []struct {
// XXX does below correlate with Cell inside top-level CellList ?
Offset uint32 // PNumber
UUID UUID
CellState CellState
}
}
// Tell a storage nodes to start an operation. Until a storage node receives
// this message, it must not serve client nodes. PM -> S.
type StartOperation struct {
PktHead
// XXX: Is this boolean needed ? Maybe this
// can be deduced from cluster state.
Backup bool
}
// Tell a storage node to stop an operation. Once a storage node receives
// this message, it must not serve client nodes. PM -> S.
type StopOperation struct {
PktHead
}
// Ask unfinished transactions S -> PM.
// Answer unfinished transactions PM -> S.
type UnfinishedTransactions struct {
PktHead
}
type AnswerUnfinishedTransactions struct {
PktHead
MaxTID Tid
TidList []struct{
UnfinishedTID Tid
}
}
// Ask locked transactions PM -> S.
// Answer locked transactions S -> PM.
type LockedTransactions struct {
PktHead
}
type AnswerLockedTransactions struct {
PktHead
TidDict map[Tid]Tid // ttid -> tid
}
// Return final tid if ttid has been committed. * -> S. C -> PM.
type FinalTID struct {
PktHead
TTID Tid
}
type AnswerFinalTID struct {
PktHead
Tid Tid
}
// Commit a transaction. PM -> S.
type ValidateTransaction struct {
PktHead
TTID Tid
Tid Tid
}
// Ask to begin a new transaction. C -> PM.
// Answer when a transaction begin, give a TID if necessary. PM -> C.
type BeginTransaction struct {
PktHead
Tid Tid
}
type AnswerBeginTransaction struct {
PktHead
Tid Tid
}
// Finish a transaction. C -> PM.
// Answer when a transaction is finished. PM -> C.
type FinishTransaction struct {
PktHead
Tid Tid
OIDList []Oid
CheckedList []Oid
}
type AnswerFinishTransaction struct {
PktHead
TTID Tid
Tid Tid
}
// Notify that a transaction blocking a replication is now finished
// M -> S
type NotifyTransactionFinished struct {
PktHead
TTID Tid
MaxTID Tid
}
// Lock information on a transaction. PM -> S.
// Notify information on a transaction locked. S -> PM.
type LockInformation struct {
PktHead
Ttid Tid
Tid Tid
}
// XXX AnswerInformationLocked ?
type AnswerLockInformation struct {
Ttid Tid
}
// Invalidate objects. PM -> C.
// XXX ask_finish_transaction ?
type InvalidateObjects struct {
PktHead
Tid Tid
OidList []Oid
}
// Unlock information on a transaction. PM -> S.
type UnlockInformation struct {
PktHead
TTID Tid
}
// Ask new object IDs. C -> PM.
// Answer new object IDs. PM -> C.
type GenerateOIDs struct {
PktHead
NumOIDs uint32 // PNumber
}
// XXX answer_new_oids ?
type AnswerGenerateOIDs struct {
PktHead
OidList []Oid
}
// Ask to store an object. Send an OID, an original serial, a current
// transaction ID, and data. C -> S.
// Answer if an object has been stored. If an object is in conflict,
// a serial of the conflicting transaction is returned. In this case,
// if this serial is newer than the current transaction ID, a client
// node must not try to resolve the conflict. S -> C.
type StoreObject struct {
PktHead
Oid Oid
Serial Tid
Compression bool
Checksum Checksum
Data []byte // XXX or string ?
DataSerial Tid
Tid Tid
Unlock bool
}
type AnswerStoreObject struct {
PktHead
Conflicting bool
Oid Oid
Serial Tid
}
// Abort a transaction. C -> S, PM.
type AbortTransaction struct {
PktHead
Tid Tid
}
// Ask to store a transaction. C -> S.
// Answer if transaction has been stored. S -> C.
type StoreTransaction struct {
PktHead
Tid Tid
User string
Description string
Extension string
OidList []Oid
// TODO _answer = PFEmpty
}
// Ask to store a transaction. C -> S.
// Answer if transaction has been stored. S -> C.
type VoteTransaction struct {
PktHead
Tid Tid
// TODO _answer = PFEmpty
}
// Ask a stored object by its OID and a serial or a TID if given. If a serial
// is specified, the specified revision of an object will be returned. If
// a TID is specified, an object right before the TID will be returned. C -> S.
// Answer the requested object. S -> C.
type GetObject struct {
PktHead
Oid Oid
Serial Tid
Tid Tid
}
// XXX answer_object ?
type AnswerGetObject struct {
PktHead
Oid Oid
SerialStart Tid
SerialEnd Tid
Compression bool
Checksum Checksum
Data []byte // XXX or string ?
DataSerial Tid
}
// Ask for TIDs between a range of offsets. The order of TIDs is descending,
// and the range is [first, last). C -> S.
// Answer the requested TIDs. S -> C.
type TIDList struct {
PktHead
First uint64 // PIndex XXX this is TID actually ? -> no it is offset in list
Last uint64 // PIndex ----//----
Partition uint32 // PNumber
}
// XXX answer_tids ?
type AnswerTIDList struct {
PktHead
TIDList []Tid
}
// Ask for length TIDs starting at min_tid. The order of TIDs is ascending.
// C -> S.
// Answer the requested TIDs. S -> C
type TIDListFrom struct {
PktHead
MinTID Tid
MaxTID Tid
Length uint32 // PNumber
Partition uint32 // PNumber
}
// XXX answer_tids ?
type AnswerTIDListFrom struct {
PktHead
TidList []Tid
}
// Ask information about a transaction. Any -> S.
// Answer information (user, description) about a transaction. S -> Any.
type TransactionInformation struct {
PktHead
Tid Tid
}
type AnswerTransactionInformation struct {
PktHead
Tid Tid
User string
Description string
Extension string
Packed bool
OidList []Oid
}
// Ask history information for a given object. The order of serials is
// descending, and the range is [first, last]. C -> S.
// Answer history information (serial, size) for an object. S -> C.
type ObjectHistory struct {
PktHead
Oid Oid
First uint64 // PIndex XXX this is actually TID
Last uint64 // PIndex ----//----
}
type AnswerObjectHistory struct {
PktHead
Oid Oid
HistoryList []struct {
Serial Tid
Size uint32 // PNumber
}
}
// All the following messages are for neoctl to admin node
// Ask information about partition
// Answer information about partition
type PartitionList struct {
PktHead
MinOffset uint32 // PNumber
MaxOffset uint32 // PNumber
UUID UUID
}
type AnswerPartitionList struct {
PktHead
PTid
RowList []RowInfo
}
// Ask information about nodes
// Answer information about nodes
type X_NodeList struct {
PktHead
NodeType
}
type AnswerNodeList struct {
PktHead
NodeList []NodeInfo
}
// Set the node state
type SetNodeState struct {
PktHead
UUID
NodeState
// XXX _answer = Error ?
}
// Ask the primary to include some pending node in the partition table
type AddPendingNodes struct {
PktHead
UUIDList []UUID
// XXX _answer = Error
}
// Ask the primary to optimize the partition table. A -> PM.
type TweakPartitionTable struct {
PktHead
UUIDList []UUID
// XXX _answer = Error
}
// Notify information about one or more nodes. PM -> Any.
type NotifyNodeInformation struct {
PktHead
NodeList []NodeInfo
}
// Ask node information
type NodeInformation struct {
PktHead
// XXX _answer = PFEmpty
}
// Set the cluster state
type SetClusterState struct {
PktHead
State ClusterState
// XXX _answer = Error
}
// Notify information about the cluster
type ClusterInformation struct {
PktHead
State ClusterState
}
// Ask state of the cluster
// Answer state of the cluster
type X_ClusterState struct { // XXX conflicts with ClusterState enum
PktHead
State ClusterState
}
// Ask storage the serial where object data is when undoing given transaction,
// for a list of OIDs.
// C -> S
// Answer serials at which object data is when undoing a given transaction.
// object_tid_dict has the following format:
// key: oid
// value: 3-tuple
// current_serial (TID)
// The latest serial visible to the undoing transaction.
// undo_serial (TID)
// Where undone data is (tid at which data is before given undo).
// is_current (bool)
// If current_serial's data is current on storage.
// S -> C
type ObjectUndoSerial struct {
PktHead
Tid Tid
LTID Tid
UndoneTID Tid
OidList []Oid
}
// XXX answer_undo_transaction ?
type AnswerObjectUndoSerial struct {
PktHead
ObjectTIDDict map[Oid]struct {
CurrentSerial Tid
UndoSerial Tid
IsCurrent bool
}
}
// Ask a storage is oid is locked by another transaction.
// C -> S
// Answer whether a transaction holds the write lock for requested object.
type HasLock struct {
PktHead
Tid Tid
Oid Oid
}
type AnswerHasLock struct {
PktHead
Oid Oid
LockState LockState
}
// Verifies if given serial is current for object oid in the database, and
// take a write lock on it (so that this state is not altered until
// transaction ends).
// Answer to AskCheckCurrentSerial.
// Same structure as AnswerStoreObject, to handle the same way, except there
// is nothing to invalidate in any client's cache.
type CheckCurrentSerial struct {
PktHead
Tid Tid
Serial Tid
Oid Oid
}
// XXX answer_store_object ?
type AnswerCheckCurrentSerial struct {
Conflicting bool
Oid Oid
Serial Tid
}
// Request a pack at given TID.
// C -> M
// M -> S
// Inform that packing it over.
// S -> M
// M -> C
type Pack struct {
PktHead
Tid Tid
}
type AnswerPack struct {
PktHead
Status bool
}
// ctl -> A
// A -> M
type CheckReplicas struct {
PktHead
PartitionDict map[uint32/*PNumber*/]UUID // partition -> source
MinTID Tid
MaxTID Tid
// XXX _answer = Error
}
// M -> S
type CheckPartition struct {
PktHead
Partition uint32 // PNumber
Source struct {
UpstreamName string
Address Address
}
MinTID Tid
MaxTID Tid
}
// Ask some stats about a range of transactions.
// Used to know if there are differences between a replicating node and
// reference node.
// S -> S
// Stats about a range of transactions.
// Used to know if there are differences between a replicating node and
// reference node.
// S -> S
type CheckTIDRange struct {
PktHead
Partition uint32 // PNumber
Length uint32 // PNumber
MinTID Tid
MaxTID Tid
}
type AnswerCheckTIDRange struct {
PktHead
Count uint32 // PNumber
Checksum Checksum
MaxTID Tid
}
// Ask some stats about a range of object history.
// Used to know if there are differences between a replicating node and
// reference node.
// S -> S
// Stats about a range of object history.
// Used to know if there are differences between a replicating node and
// reference node.
// S -> S
type CheckSerialRange struct {
PktHead
Partition uint32 // PNumber
Length uint32 // PNumber
MinTID Tid
MaxTID Tid
MinOID Oid
}
type AnswerCheckSerialRange struct {
Count uint32 // PNumber
TidChecksum Checksum
MaxTID Tid
OidChecksum Checksum
MaxOID Oid
}
// S -> M
type PartitionCorrupted struct {
PktHead
Partition uint32 // PNumber
CellList []UUID
}
// Ask last committed TID.
// C -> M
// Answer last committed TID.
// M -> C
type LastTransaction struct {
PktHead
}
type AnswerLastTransaction struct {
PktHead
Tid Tid
}
// Notify that node is ready to serve requests.
// S -> M
type NotifyReady struct {
PktHead
}
// replication
// TODO
t/neo/pktgen.go → t/neo/protogen.go
View file @ d4be13da
// TODO text what it does (generates code for pkt.go) // TODO text what it does (generates code for proto.go)
// +build ignore // +build ignore
...@@ -17,7 +17,7 @@ func main() { ...@@ -17,7 +17,7 @@ func main() {
fset := token.NewFileSet() fset := token.NewFileSet()
var mode parser.Mode = 0 // parser.Trace var mode parser.Mode = 0 // parser.Trace
f, err := parser.ParseFile(fset, "pkt.go", nil, mode) f, err := parser.ParseFile(fset, "proto.go", nil, mode)
if err != nil { if err != nil {
panic(err) // XXX log panic(err) // XXX log
} }
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7