Merge perch.ndb.mysql.com:/home/jonas/src/mysql-5.0-ndb (4a4425cc) · Commits · Software / OSDI20 Artifacts / mariadb

storage/ndb/src/kernel/blocks/ERROR_codes.txt

+6 −1

Original line number	Diff line number	Diff line
		@@ -5,7 +5,7 @@ Next DBACC 3002
		Next DBTUP 4029
		Next DBLQH 5047
		Next DBDICT 6008
		Next DBDIH 7193
		Next DBDIH 7195
		Next DBTC 8054
		Next CMVMI 9000
		Next BACKUP 10038
		@@ -81,6 +81,11 @@ Delay GCP_SAVEREQ by 10 secs

		7185: Dont reply to COPY_GCI_REQ where reason == GCP

		7193: Dont send LCP_FRAG_ORD to self, and crash when sending first
		LCP_FRAG_ORD(last)

		7194: Force removeNodeFromStored to complete in the middle of MASTER_LCPCONF

		ERROR CODES FOR TESTING NODE FAILURE, LOCAL CHECKPOINT HANDLING:
		-----------------------------------------------------------------

storage/ndb/src/kernel/blocks/dbdih/Dbdih.hpp

+10 −0

Original line number	Diff line number	Diff line
		@@ -1310,7 +1310,17 @@ private:
		LcpStatus lcpStatus;
		Uint32 lcpStatusUpdatedPlace;

		struct Save {
		LcpStatus m_status;
		Uint32 m_place;
		} m_saveState[10];

		void setLcpStatus(LcpStatus status, Uint32 line){
		for (Uint32 i = 9; i > 0; i--)
		m_saveState[i] = m_saveState[i-1];
		m_saveState[0].m_status = lcpStatus;
		m_saveState[0].m_place = lcpStatusUpdatedPlace;

		lcpStatus = status;
		lcpStatusUpdatedPlace = line;
		}

storage/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp

+54 −6

Original line number	Diff line number	Diff line
		@@ -5181,10 +5181,18 @@ void Dbdih::startRemoveFailedNode(Signal* signal, NodeRecordPtr failedNodePtr)
		}

		jam();

		if (!ERROR_INSERTED(7194))
		{
		signal->theData[0] = DihContinueB::ZREMOVE_NODE_FROM_TABLE;
		signal->theData[1] = failedNodePtr.i;
		signal->theData[2] = 0; // Tab id
		sendSignal(reference(), GSN_CONTINUEB, signal, 3, JBB);
		}
		else
		{
		ndbout_c("7194 Not starting ZREMOVE_NODE_FROM_TABLE");
		}

		setLocalNodefailHandling(signal, failedNodePtr.i, NF_REMOVE_NODE_FROM_TABLE);
		}//Dbdih::startRemoveFailedNode()
		@@ -6115,11 +6123,21 @@ Dbdih::checkEmptyLcpComplete(Signal *signal){
		signal->theData[0] = 7012;
		execDUMP_STATE_ORD(signal);

		if (ERROR_INSERTED(7194))
		{
		ndbout_c("7194 starting ZREMOVE_NODE_FROM_TABLE");
		signal->theData[0] = DihContinueB::ZREMOVE_NODE_FROM_TABLE;
		signal->theData[1] = c_lcpMasterTakeOverState.failedNodeId;
		signal->theData[2] = 0; // Tab id
		sendSignal(reference(), GSN_CONTINUEB, signal, 3, JBB);
		}

		c_lcpMasterTakeOverState.set(LMTOS_INITIAL, __LINE__);
		MasterLCPReq * const req = (MasterLCPReq *)&signal->theData[0];
		req->masterRef = reference();
		req->failedNodeId = c_lcpMasterTakeOverState.failedNodeId;
		sendLoopMacro(MASTER_LCPREQ, sendMASTER_LCPREQ);

		} else {
		sendMASTER_LCPCONF(signal);
		}
		@@ -6432,6 +6450,15 @@ void Dbdih::execMASTER_LCPCONF(Signal* signal)
		{
		const MasterLCPConf * const conf = (MasterLCPConf *)&signal->theData[0];
		jamEntry();

		if (ERROR_INSERTED(7194))
		{
		ndbout_c("delaying MASTER_LCPCONF due to error 7194");
		sendSignalWithDelay(reference(), GSN_MASTER_LCPCONF, signal,
		300, signal->getLength());
		return;
		}

		Uint32 senderNodeId = conf->senderNodeId;
		MasterLCPConf::State lcpState = (MasterLCPConf::State)conf->lcpState;
		const Uint32 failedNodeId = conf->failedNodeId;
		@@ -6566,7 +6593,6 @@ void Dbdih::MASTER_LCPhandling(Signal* signal, Uint32 failedNodeId)
		#endif

		c_lcpState.keepGci = SYSFILE->keepGCI;
		c_lcpState.setLcpStatus(LCP_START_LCP_ROUND, __LINE__);
		startLcpRoundLoopLab(signal, 0, 0);
		break;
		}
		@@ -10538,6 +10564,8 @@ void Dbdih::sendLastLCP_FRAG_ORD(Signal* signal)
		if(ERROR_INSERTED(7075)){
		continue;
		}

		CRASH_INSERTION(7193);
		BlockReference ref = calcLqhBlockRef(nodePtr.i);
		sendSignal(ref, GSN_LCP_FRAG_ORD, signal,LcpFragOrd::SignalLength, JBB);
		}
		@@ -10765,6 +10793,13 @@ Dbdih::checkLcpAllTablesDoneInLqh(){
		CRASH_INSERTION2(7017, !isMaster());

		c_lcpState.setLcpStatus(LCP_TAB_COMPLETED, __LINE__);

		if (ERROR_INSERTED(7194))
		{
		ndbout_c("CLEARING 7194");
		CLEAR_ERROR_INSERT_VALUE;
		}

		return true;
		}

		@@ -10954,6 +10989,11 @@ Dbdih::sendLCP_FRAG_ORD(Signal* signal,

		BlockReference ref = calcLqhBlockRef(replicaPtr.p->procNode);

		if (ERROR_INSERTED(7193) && replicaPtr.p->procNode == getOwnNodeId())
		{
		return;
		}

		LcpFragOrd * const lcpFragOrd = (LcpFragOrd *)&signal->theData[0];
		lcpFragOrd->tableId = info.tableId;
		lcpFragOrd->fragmentId = info.fragId;
		@@ -14500,6 +14540,14 @@ Dbdih::execDUMP_STATE_ORD(Signal* signal)
		("immediateLcpStart = %d masterLcpNodeId = %d",
		c_lcpState.immediateLcpStart,
		refToNode(c_lcpState.m_masterLcpDihRef));

		for (Uint32 i = 0; i<10; i++)
		{
		infoEvent("%u : status: %u place: %u", i,
		c_lcpState.m_saveState[i].m_status,
		c_lcpState.m_saveState[i].m_place);
		}

		infoEvent("-- Node %d LCP STATE --", getOwnNodeId());
		}

storage/ndb/test/ndbapi/testNodeRestart.cpp

+48 −0

Original line number	Diff line number	Diff line
		@@ -1836,6 +1836,51 @@ runBug31525(NDBT_Context* ctx, NDBT_Step* step)
		return NDBT_OK;
		}

		int
		runBug32160(NDBT_Context* ctx, NDBT_Step* step)
		{
		int result = NDBT_OK;
		int loops = ctx->getNumLoops();
		int records = ctx->getNumRecords();
		Ndb* pNdb = GETNDB(step);
		NdbRestarter res;

		if (res.getNumDbNodes() < 2)
		{
		return NDBT_OK;
		}

		int master = res.getMasterNodeId();
		int next = res.getNextMasterNodeId(master);

		if (res.insertErrorInNode(next, 7194))
		{
		return NDBT_FAILED;
		}

		int val2[] = { DumpStateOrd::CmvmiSetRestartOnErrorInsert, 1 };
		if (res.dumpStateOneNode(master, val2, 2))
		return NDBT_FAILED;

		if (res.insertErrorInNode(master, 7193))
		return NDBT_FAILED;

		int val3[] = { 7099 };
		if (res.dumpStateOneNode(master, val3, 1))
		return NDBT_FAILED;

		if (res.waitNodesNoStart(&master, 1))
		return NDBT_FAILED;

		if (res.startNodes(&master, 1))
		return NDBT_FAILED;

		if (res.waitClusterStarted())
		return NDBT_FAILED;

		return NDBT_OK;
		}

		NDBT_TESTSUITE(testNodeRestart);
		TESTCASE("NoLoad",
		"Test that one node at a time can be stopped and then restarted "\
		@@ -2205,6 +2250,9 @@ TESTCASE("Bug28717", ""){
		TESTCASE("Bug29364", ""){
		INITIALIZER(runBug29364);
		}
		TESTCASE("Bug32160", ""){
		INITIALIZER(runBug32160);
		}
		NDBT_TESTSUITE_END(testNodeRestart);

		int main(int argc, const char** argv){

storage/ndb/test/run-test/daily-basic-tests.txt

+4 −0

Original line number	Diff line number	Diff line
		@@ -581,6 +581,10 @@ max-time: 1000
		cmd: testNodeRestart
		args: -n Bug29364 T1

		max-time: 300
		cmd: testNodeRestart
		args: -n Bug32160 T1

		#
		# DICT TESTS
		max-time: 500