Skip to content

M
MariaDB
Commit 850211eb authored by pekka@mysql.com's avatar pekka@mysql.com
Browse files
Options

Merge

parents b75789e3 6778b029
Show whitespace changes
Inline Side-by-side
Showing with 234 additions and 77 deletions
ndb/src/kernel/blocks/dbtup/Dbtup.hpp
View file @ 850211eb
......@@ -1779,6 +1779,10 @@ private:
Operationrec* const regOperPtr,
Tablerec* const regTabPtr);
int addTuxEntries(Signal* signal,
Operationrec* regOperPtr,
Tablerec* regTabPtr);
// these crash the node on error
void executeTuxCommitTriggers(Signal* signal,
......@@ -1789,6 +1793,10 @@ private:
Operationrec* regOperPtr,
Tablerec* const regTabPtr);
void removeTuxEntries(Signal* signal,
Operationrec* regOperPtr,
Tablerec* regTabPtr);
// *****************************************************************
// Error Handling routines.
// *****************************************************************
......
ndb/src/kernel/blocks/dbtup/DbtupTrigger.cpp
View file @ 850211eb
......@@ -973,25 +973,7 @@ Dbtup::executeTuxInsertTriggers(Signal* signal,
req->pageOffset = regOperPtr->pageOffset;
req->tupVersion = tupVersion;
req->opInfo = TuxMaintReq::OpAdd;
// loop over index list
const ArrayList<TupTriggerData>& triggerList = regTabPtr->tuxCustomTriggers;
TriggerPtr triggerPtr;
triggerList.first(triggerPtr);
while (triggerPtr.i != RNIL) {
ljam();
req->indexId = triggerPtr.p->indexId;
req->errorCode = RNIL;
EXECUTE_DIRECT(DBTUX, GSN_TUX_MAINT_REQ,
signal, TuxMaintReq::SignalLength);
ljamEntry();
if (req->errorCode != 0) {
ljam();
terrorCode = req->errorCode;
return -1;
}
triggerList.next(triggerPtr);
}
return 0;
return addTuxEntries(signal, regOperPtr, regTabPtr);
}
int
......@@ -1012,9 +994,18 @@ Dbtup::executeTuxUpdateTriggers(Signal* signal,
req->pageOffset = regOperPtr->pageOffset;
req->tupVersion = tupVersion;
req->opInfo = TuxMaintReq::OpAdd;
// loop over index list
return addTuxEntries(signal, regOperPtr, regTabPtr);
}
int
Dbtup::addTuxEntries(Signal* signal,
Operationrec* regOperPtr,
Tablerec* regTabPtr)
{
TuxMaintReq* const req = (TuxMaintReq*)signal->getDataPtrSend();
const ArrayList<TupTriggerData>& triggerList = regTabPtr->tuxCustomTriggers;
TriggerPtr triggerPtr;
Uint32 failPtrI;
triggerList.first(triggerPtr);
while (triggerPtr.i != RNIL) {
ljam();
......@@ -1026,11 +1017,29 @@ Dbtup::executeTuxUpdateTriggers(Signal* signal,
if (req->errorCode != 0) {
ljam();
terrorCode = req->errorCode;
return -1;
failPtrI = triggerPtr.i;
goto fail;
}
triggerList.next(triggerPtr);
}
return 0;
fail:
req->opInfo = TuxMaintReq::OpRemove;
triggerList.first(triggerPtr);
while (triggerPtr.i != failPtrI) {
ljam();
req->indexId = triggerPtr.p->indexId;
req->errorCode = RNIL;
EXECUTE_DIRECT(DBTUX, GSN_TUX_MAINT_REQ,
signal, TuxMaintReq::SignalLength);
ljamEntry();
ndbrequire(req->errorCode == 0);
triggerList.next(triggerPtr);
}
#ifdef VM_TRACE
ndbout << "aborted partial tux update: op " << hex << regOperPtr << endl;
#endif
return -1;
}
int
......@@ -1049,7 +1058,6 @@ Dbtup::executeTuxCommitTriggers(Signal* signal,
{
TuxMaintReq* const req = (TuxMaintReq*)signal->getDataPtrSend();
// get version
// XXX could add prevTupVersion to Operationrec
Uint32 tupVersion;
if (regOperPtr->optype == ZINSERT) {
if (! regOperPtr->deleteInsertFlag)
......@@ -1087,21 +1095,7 @@ Dbtup::executeTuxCommitTriggers(Signal* signal,
req->pageOffset = regOperPtr->pageOffset;
req->tupVersion = tupVersion;
req->opInfo = TuxMaintReq::OpRemove;
// loop over index list
const ArrayList<TupTriggerData>& triggerList = regTabPtr->tuxCustomTriggers;
TriggerPtr triggerPtr;
triggerList.first(triggerPtr);
while (triggerPtr.i != RNIL) {
ljam();
req->indexId = triggerPtr.p->indexId;
req->errorCode = RNIL;
EXECUTE_DIRECT(DBTUX, GSN_TUX_MAINT_REQ,
signal, TuxMaintReq::SignalLength);
ljamEntry();
// commit must succeed
ndbrequire(req->errorCode == 0);
triggerList.next(triggerPtr);
}
removeTuxEntries(signal, regOperPtr, regTabPtr);
}
void
......@@ -1132,7 +1126,15 @@ Dbtup::executeTuxAbortTriggers(Signal* signal,
req->pageOffset = regOperPtr->pageOffset;
req->tupVersion = tupVersion;
req->opInfo = TuxMaintReq::OpRemove;
// loop over index list
removeTuxEntries(signal, regOperPtr, regTabPtr);
}
void
Dbtup::removeTuxEntries(Signal* signal,
Operationrec* regOperPtr,
Tablerec* regTabPtr)
{
TuxMaintReq* const req = (TuxMaintReq*)signal->getDataPtrSend();
const ArrayList<TupTriggerData>& triggerList = regTabPtr->tuxCustomTriggers;
TriggerPtr triggerPtr;
triggerList.first(triggerPtr);
......@@ -1143,7 +1145,7 @@ Dbtup::executeTuxAbortTriggers(Signal* signal,
EXECUTE_DIRECT(DBTUX, GSN_TUX_MAINT_REQ,
signal, TuxMaintReq::SignalLength);
ljamEntry();
// abort must succeed
// must succeed
ndbrequire(req->errorCode == 0);
triggerList.next(triggerPtr);
}
......
ndb/src/kernel/blocks/dbtup/Notes.txt
View file @ 850211eb
......@@ -135,6 +135,24 @@ abort DELETE none -
1) alternatively, store prevTupVersion in operation record.
Abort from ordered index error
------------------------------
Obviously, index update failure causes operation failure.
The operation is then aborted later by TC.
The problem here is with multiple indexes. Some may have been
updated successfully before the one that failed. Therefore
the trigger code aborts the successful ones already in
the prepare phase.
In other words, multiple indexes are treated as one.
Abort from any cause
--------------------
[ hairy stuff ]
Read attributes, query status
-----------------------------
......@@ -170,14 +188,11 @@ used to decide if the scan can see the tuple.
This signal may also be called during any phase since commit/abort
of all operations is not done in one time-slice.
Commit and abort
----------------
[ hairy stuff ]
Problems
--------
Current abort code can destroy a tuple version too early. This
happens in test case "ticuur" (insert-commit-update-update-rollback),
if abort of first update arrives before abort of second update.
vim: set textwidth=68:
ndb/src/kernel/blocks/dbtux/DbtuxNode.cpp
View file @ 850211eb
......@@ -23,6 +23,11 @@
int
Dbtux::allocNode(Signal* signal, NodeHandle& node)
{
if (ERROR_INSERTED(12007)) {
jam();
CLEAR_ERROR_INSERT_VALUE;
return TuxMaintReq::NoMemError;
}
Frag& frag = node.m_frag;
Uint32 pageId = NullTupLoc.getPageId();
Uint32 pageOffset = NullTupLoc.getPageOffset();
......@@ -34,6 +39,12 @@ Dbtux::allocNode(Signal* signal, NodeHandle& node)
node.m_loc = TupLoc(pageId, pageOffset);
node.m_node = reinterpret_cast<TreeNode*>(node32);
ndbrequire(node.m_loc != NullTupLoc && node.m_node != 0);
} else {
switch (errorCode) {
case 827:
errorCode = TuxMaintReq::NoMemError;
break;
}
}
return errorCode;
}
......
ndb/src/ndbapi/ndberror.c
View file @ 850211eb
......@@ -179,10 +179,11 @@ ErrorBundle ErrorCodes[] = {
*/
{ 623, IS, "623" },
{ 624, IS, "624" },
{ 625, IS, "Out of memory in Ndb Kernel, index part (increase IndexMemory)" },
{ 625, IS, "Out of memory in Ndb Kernel, hash index part (increase IndexMemory)" },
{ 640, IS, "Too many hash indexes (should not happen)" },
{ 826, IS, "Too many tables and attributes (increase MaxNoOfAttributes or MaxNoOfTables)" },
{ 827, IS, "Out of memory in Ndb Kernel, data part (increase DataMemory)" },
{ 827, IS, "Out of memory in Ndb Kernel, table data (increase DataMemory)" },
{ 902, IS, "Out of memory in Ndb Kernel, ordered index data (increase DataMemory)" },
{ 902, IS, "Out of memory in Ndb Kernel, data part (increase DataMemory)" },
{ 903, IS, "Too many ordered indexes (increase MaxNoOfOrderedIndexes)" },
{ 904, IS, "Out of fragment records (increase MaxNoOfOrderedIndexes)" },
......
ndb/test/ndbapi/testOIBasic.cpp
View file @ 850211eb
......@@ -259,6 +259,8 @@ struct Par : public Opt {
bool m_verify;
// deadlock possible
bool m_deadlock;
// abort percentabge
unsigned m_abortpct;
NdbOperation::LockMode m_lockmode;
// ordered range scan
bool m_ordered;
......@@ -281,6 +283,7 @@ struct Par : public Opt {
m_randomkey(false),
m_verify(false),
m_deadlock(false),
m_abortpct(0),
m_lockmode(NdbOperation::LM_Read),
m_ordered(false),
m_descending(false) {
......@@ -1143,7 +1146,7 @@ struct Con {
NdbScanFilter* m_scanfilter;
enum ScanMode { ScanNo = 0, Committed, Latest, Exclusive };
ScanMode m_scanmode;
enum ErrType { ErrNone = 0, ErrDeadlock, ErrOther };
enum ErrType { ErrNone = 0, ErrDeadlock, ErrNospace, ErrOther };
ErrType m_errtype;
Con() :
m_ndb(0), m_dic(0), m_tx(0), m_op(0), m_indexop(0),
......@@ -1172,7 +1175,7 @@ struct Con {
int endFilter();
int setFilter(int num, int cond, const void* value, unsigned len);
int execute(ExecType t);
int execute(ExecType t, bool& deadlock);
int execute(ExecType t, bool& deadlock, bool& nospace);
int readTuples(Par par);
int readIndexTuples(Par par);
int executeScan();
......@@ -1354,17 +1357,21 @@ Con::execute(ExecType t)
}
int
Con::execute(ExecType t, bool& deadlock)
Con::execute(ExecType t, bool& deadlock, bool& nospace)
{
int ret = execute(t);
if (ret != 0) {
if (deadlock && m_errtype == ErrDeadlock) {
if (ret != 0 && deadlock && m_errtype == ErrDeadlock) {
LL3("caught deadlock");
ret = 0;
}
} else {
deadlock = false;
}
if (ret != 0 && nospace && m_errtype == ErrNospace) {
LL3("caught nospace");
ret = 0;
} else {
nospace = false;
}
CHK(ret == 0);
return 0;
}
......@@ -1475,6 +1482,8 @@ Con::printerror(NdbOut& out)
// 631 is new, occurs only on 4 db nodes, needs to be checked out
if (code == 266 || code == 274 || code == 296 || code == 297 || code == 499 || code == 631)
m_errtype = ErrDeadlock;
if (code == 826 || code == 827 || code == 902)
m_errtype = ErrNospace;
}
if (m_op && m_op->getNdbError().code != 0) {
LL0(++any << " op : error " << m_op->getNdbError());
......@@ -1644,6 +1653,36 @@ createindex(Par par)
// data sets
static unsigned
urandom(unsigned n)
{
if (n == 0)
return 0;
unsigned i = random() % n;
return i;
}
static int
irandom(unsigned n)
{
if (n == 0)
return 0;
int i = random() % n;
if (random() & 0x1)
i = -i;
return i;
}
static bool
randompct(unsigned pct)
{
if (pct == 0)
return false;
if (pct >= 100)
return true;
return urandom(100) < pct;
}
// Val - typed column value
struct Val {
......@@ -2480,8 +2519,8 @@ struct Set {
void dbsave(unsigned i);
void calc(Par par, unsigned i, unsigned mask = 0);
bool pending(unsigned i, unsigned mask) const;
void notpending(unsigned i);
void notpending(const Lst& lst);
void notpending(unsigned i, ExecType et = Commit);
void notpending(const Lst& lst, ExecType et = Commit);
void dbdiscard(unsigned i);
void dbdiscard(const Lst& lst);
const Row& dbrow(unsigned i) const;
......@@ -2831,7 +2870,33 @@ Set::putval(unsigned i, bool force, unsigned n)
return 0;
}
// verify
void
Set::notpending(unsigned i, ExecType et)
{
assert(m_row[i] != 0);
Row& row = *m_row[i];
if (et == Commit) {
if (row.m_pending == Row::InsOp)
row.m_exist = true;
if (row.m_pending == Row::DelOp)
row.m_exist = false;
} else {
if (row.m_pending == Row::InsOp)
row.m_exist = false;
if (row.m_pending == Row::DelOp)
row.m_exist = true;
}
row.m_pending = Row::NoOp;
}
void
Set::notpending(const Lst& lst, ExecType et)
{
for (unsigned j = 0; j < lst.m_cnt; j++) {
unsigned i = lst.m_arr[j];
notpending(i, et);
}
}
int
Set::verify(Par par, const Set& set2) const
......@@ -3213,14 +3278,20 @@ pkinsert(Par par)
lst.push(i);
if (lst.cnt() == par.m_batch) {
bool deadlock = par.m_deadlock;
CHK(con.execute(Commit, deadlock) == 0);
bool nospace = true;
ExecType et = randompct(par.m_abortpct) ? Rollback : Commit;
CHK(con.execute(et, deadlock, nospace) == 0);
con.closeTransaction();
if (deadlock) {
LL1("pkinsert: stop on deadlock [at 1]");
return 0;
}
if (nospace) {
LL1("pkinsert: cnt=" << j << " stop on nospace");
return 0;
}
set.lock();
set.notpending(lst);
set.notpending(lst, et);
set.unlock();
lst.reset();
CHK(con.startTransaction() == 0);
......@@ -3228,14 +3299,20 @@ pkinsert(Par par)
}
if (lst.cnt() != 0) {
bool deadlock = par.m_deadlock;
CHK(con.execute(Commit, deadlock) == 0);
bool nospace = true;
ExecType et = randompct(par.m_abortpct) ? Rollback : Commit;
CHK(con.execute(et, deadlock, nospace) == 0);
con.closeTransaction();
if (deadlock) {
LL1("pkinsert: stop on deadlock [at 2]");
return 0;
}
if (nospace) {
LL1("pkinsert: end: stop on nospace");
return 0;
}
set.lock();
set.notpending(lst);
set.notpending(lst, et);
set.unlock();
return 0;
}
......@@ -3253,6 +3330,7 @@ pkupdate(Par par)
CHK(con.startTransaction() == 0);
Lst lst;
bool deadlock = false;
bool nospace = false;
for (unsigned j = 0; j < par.m_rows; j++) {
unsigned j2 = ! par.m_randomkey ? j : urandom(par.m_rows);
unsigned i = thrrow(par, j2);
......@@ -3269,28 +3347,38 @@ pkupdate(Par par)
lst.push(i);
if (lst.cnt() == par.m_batch) {
deadlock = par.m_deadlock;
CHK(con.execute(Commit, deadlock) == 0);
nospace = true;
ExecType et = randompct(par.m_abortpct) ? Rollback : Commit;
CHK(con.execute(et, deadlock, nospace) == 0);
if (deadlock) {
LL1("pkupdate: stop on deadlock [at 1]");
break;
}
if (nospace) {
LL1("pkupdate: cnt=" << j << " stop on nospace [at 1]");
break;
}
con.closeTransaction();
set.lock();
set.notpending(lst);
set.notpending(lst, et);
set.dbdiscard(lst);
set.unlock();
lst.reset();
CHK(con.startTransaction() == 0);
}
}
if (! deadlock && lst.cnt() != 0) {
if (! deadlock && ! nospace && lst.cnt() != 0) {
deadlock = par.m_deadlock;
CHK(con.execute(Commit, deadlock) == 0);
nospace = true;
ExecType et = randompct(par.m_abortpct) ? Rollback : Commit;
CHK(con.execute(et, deadlock, nospace) == 0);
if (deadlock) {
LL1("pkupdate: stop on deadlock [at 1]");
LL1("pkupdate: stop on deadlock [at 2]");
} else if (nospace) {
LL1("pkupdate: end: stop on nospace [at 2]");
} else {
set.lock();
set.notpending(lst);
set.notpending(lst, et);
set.dbdiscard(lst);
set.unlock();
}
......@@ -3309,6 +3397,7 @@ pkdelete(Par par)
CHK(con.startTransaction() == 0);
Lst lst;
bool deadlock = false;
bool nospace = false;
for (unsigned j = 0; j < par.m_rows; j++) {
unsigned j2 = ! par.m_randomkey ? j : urandom(par.m_rows);
unsigned i = thrrow(par, j2);
......@@ -3323,27 +3412,31 @@ pkdelete(Par par)
lst.push(i);
if (lst.cnt() == par.m_batch) {
deadlock = par.m_deadlock;
CHK(con.execute(Commit, deadlock) == 0);
nospace = true;
ExecType et = randompct(par.m_abortpct) ? Rollback : Commit;
CHK(con.execute(et, deadlock, nospace) == 0);
if (deadlock) {
LL1("pkdelete: stop on deadlock [at 1]");
break;
}
con.closeTransaction();
set.lock();
set.notpending(lst);
set.notpending(lst, et);
set.unlock();
lst.reset();
CHK(con.startTransaction() == 0);
}
}
if (! deadlock && lst.cnt() != 0) {
if (! deadlock && ! nospace && lst.cnt() != 0) {
deadlock = par.m_deadlock;
CHK(con.execute(Commit, deadlock) == 0);
nospace = true;
ExecType et = randompct(par.m_abortpct) ? Rollback : Commit;
CHK(con.execute(et, deadlock, nospace) == 0);
if (deadlock) {
LL1("pkdelete: stop on deadlock [at 2]");
} else {
set.lock();
set.notpending(lst);
set.notpending(lst, et);
set.unlock();
}
}
......@@ -4094,6 +4187,10 @@ readverify(Par par)
if (par.m_noverify)
return 0;
par.m_verify = true;
if (par.m_abortpct != 0) {
LL2("skip verify in this version"); // implement in 5.0 version
par.m_verify = false;
}
par.m_lockmode = NdbOperation::LM_CommittedRead;
CHK(pkread(par) == 0);
CHK(scanreadall(par) == 0);
......@@ -4106,6 +4203,10 @@ readverifyfull(Par par)
if (par.m_noverify)
return 0;
par.m_verify = true;
if (par.m_abortpct != 0) {
LL2("skip verify in this version"); // implement in 5.0 version
par.m_verify = false;
}
par.m_lockmode = NdbOperation::LM_CommittedRead;
const Tab& tab = par.tab();
if (par.m_no == 0) {
......@@ -4457,11 +4558,11 @@ runstep(Par par, const char* fname, TFunc func, unsigned mode)
for (n = 0; n < threads; n++) {
LL4("start " << n);
Thr& thr = *g_thrlist[n];
thr.m_par.m_tab = par.m_tab;
thr.m_par.m_set = par.m_set;
thr.m_par.m_tmr = par.m_tmr;
thr.m_par.m_lno = par.m_lno;
thr.m_par.m_slno = par.m_slno;
Par oldpar = thr.m_par;
// update parameters
thr.m_par = par;
thr.m_par.m_no = oldpar.m_no;
thr.m_par.m_con = oldpar.m_con;
thr.m_func = func;
thr.start();
}
......@@ -4590,6 +4691,24 @@ tbusybuild(Par par)
return 0;
}
static int
trollback(Par par)
{
par.m_abortpct = 50;
RUNSTEP(par, droptable, ST);
RUNSTEP(par, createtable, ST);
RUNSTEP(par, invalidatetable, MT);
RUNSTEP(par, pkinsert, MT);
RUNSTEP(par, createindex, ST);
RUNSTEP(par, invalidateindex, MT);
RUNSTEP(par, readverify, ST);
for (par.m_slno = 0; par.m_slno < par.m_subloop; par.m_slno++) {
RUNSTEP(par, mixedoperations, MT);
RUNSTEP(par, readverify, ST);
}
return 0;
}
static int
ttimebuild(Par par)
{
......@@ -4712,6 +4831,7 @@ tcaselist[] = {
TCase("d", tpkopsread, "pk operations and scan reads"),
TCase("e", tmixedops, "pk operations and scan operations"),
TCase("f", tbusybuild, "pk operations and index build"),
TCase("g", trollback, "operations with random rollbacks"),
TCase("t", ttimebuild, "time index build"),
TCase("u", ttimemaint, "time index maintenance"),
TCase("v", ttimescan, "time full scan table vs index on pk"),
......
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