Support read-committed isolation level

knizhnik · kelvich · commit 1161aab5db80 · 2017-11-13T02:17:39.000+03:00
diff --git a/multimaster.c b/multimaster.c
@@ -504,10 +504,22 @@ csn_t MtmDistributedTransactionSnapshot(TransactionId xid, int nodeId, nodemask_
     return snapshot;
 }
 
+void MtmSetSnapshot(csn_t globalSnapshot)
+{
+	MtmLock(LW_EXCLUSIVE);
+	MtmSyncClock(globalSnapshot);	
+	MtmTx.snapshot = globalSnapshot;	
+	MtmUnlock();
+}
 
+		
 Snapshot MtmGetSnapshot(Snapshot snapshot)
 {
 	snapshot = PgGetSnapshotData(snapshot);
+	if (XactIsoLevel == XACT_READ_COMMITTED && MtmTx.snapshot != INVALID_CSN && TransactionIdIsValid(GetCurrentTransactionIdIfAny())) { 
+		MtmTx.snapshot = MtmGetCurrentTime();
+		LogLogicalMessage("S", (char*)&MtmTx.snapshot, sizeof(MtmTx.snapshot), true);
+	}
 	RecentGlobalDataXmin = RecentGlobalXmin = Mtm->oldestXid;
 	return snapshot;
 }
@@ -540,7 +552,7 @@ bool MtmXidInMVCCSnapshot(TransactionId xid, Snapshot snapshot)
 
     Assert(xid != InvalidTransactionId);
 	
-	if (!MtmUseDtm) { 
+	if (!MtmUseDtm || TransactionIdPrecedes(xid, Mtm->oldestXid)) { 
 		return PgXidInMVCCSnapshot(xid, snapshot);
 	}
 	MtmLock(LW_SHARED);
@@ -681,6 +693,10 @@ MtmAdjustOldestXid(TransactionId xid)
 				hash_search(MtmGid2State, &prev->gid, HASH_REMOVE, NULL);
 			}
 		}
+		if (ts != NULL) { 
+			MTM_LOG2("Adjust(%lld) stop at snashot %lld, xid %lld, pinned=%d, oldestSnaphsot=%lld\n",
+					 (long64)xid, ts->csn, (long64)ts->xid, ts->isPinned, oldestSnapshot);
+		}					 
 	} 
 
 	if (MtmUseDtm && !MtmVolksWagenMode) 
@@ -2827,11 +2843,13 @@ static bool ConfigIsSane(void)
 {
 	bool ok = true;
 
+#if 0
 	if (DefaultXactIsoLevel != XACT_REPEATABLE_READ)
 	{
 		MTM_ELOG(WARNING, "multimaster requires default_transaction_isolation = 'repeatable read'");
 		ok = false;
 	}
+#endif
 
 	if (MtmMaxNodes < 1)
 	{
@@ -5123,11 +5141,11 @@ static void MtmProcessUtility(Node *parsetree, const char *queryString,
 		standard_ProcessUtility(parsetree, queryString, context,
 								params, dest, completionTag);
 	}
-	
+#if 0
     if (!MtmVolksWagenMode && MtmTx.isDistributed && XactIsoLevel != XACT_REPEATABLE_READ) { 
 		MTM_ELOG(ERROR, "Isolation level %s is not supported by multimaster", isoLevelStr[XactIsoLevel]);
 	}
-
+#endif
 	if (MyXactAccessedTempRel)
 	{
 		MTM_LOG1("Xact accessed temp table, stopping replication");
diff --git a/multimaster.h b/multimaster.h
@@ -424,5 +424,6 @@ extern bool MtmTransIsActive(void);
 extern MtmTransState* MtmGetActiveTransaction(MtmL2List* list);
 extern void MtmReleaseLocks(void);
 extern void MtmInitMessage(MtmArbiterMessage* msg, MtmMessageCode code);
+extern void MtmSetSnapshot(csn_t snapshot);
 
 #endif
diff --git a/pglogical_apply.c b/pglogical_apply.c
@@ -483,8 +483,15 @@ process_remote_message(StringInfo s)
 			standalone = true;
 			break;
 		}
-
-	}
+	    case 'S':
+		{
+  		    Assert(messageSize == sizeof(csn_t));
+		    MtmSetSnapshot(*(csn_t*)messageBody);
+			break;
+		}
+	    default:
+		    Assert(false);
+ 	}
 	return standalone;
 }
 	
diff --git a/pglogical_proto.c b/pglogical_proto.c
@@ -180,6 +180,11 @@ pglogical_write_message(StringInfo out,
 			MTM_LOG1("Send deadlock message to node %d", MtmReplicationNodeId);
 		}
 		break;
+	  case 'S':
+		if (MtmIsFilteredTxn) {
+			return;
+		}
+		break;
 	  case 'D':
 		if (MtmIsFilteredTxn) {
 			MTM_LOG2("%d: pglogical_write_message filtered", MyProcPid);
diff --git a/tests/dtmbench.cpp b/tests/dtmbench.cpp
@@ -69,6 +69,7 @@ struct config
     int updatePercent;
     vector<string> connections;
 	bool scatter;
+	bool avoidDeadlocks;
 
     config() {
         nReaders = 1;
@@ -77,6 +78,7 @@ struct config
         nAccounts = 100000;
         updatePercent = 100;
 		scatter = false;
+		avoidDeadlocks = false;
     }
 };
 
@@ -157,6 +159,12 @@ void* writer(void* arg)
 		if (cfg.scatter) { 
 			srcAcc = srcAcc/cfg.nWriters*cfg.nWriters + t.id;
 			dstAcc = dstAcc/cfg.nWriters*cfg.nWriters + t.id;
+		} else if (cfg.avoidDeadlocks) { 
+			if (dstAcc < srcAcc) { 
+				int tmp = srcAcc;
+				srcAcc = dstAcc;
+				dstAcc = tmp;
+			}
 		}
         try {            
             if (random() % 100 < cfg.updatePercent) { 
@@ -240,6 +248,9 @@ int main (int argc, char* argv[])
             case 'i':
                 initialize = true;
                 continue;
+            case 'd':
+			    cfg.avoidDeadlocks = true;
+                continue;
             }
         }
         printf("Options:\n"
@@ -249,6 +260,8 @@ int main (int argc, char* argv[])
                "\t-n N\tnumber of iterations (1000)\n"
                "\t-p N\tupdate percent (100)\n"
                "\t-c STR\tdatabase connection string\n"
+               "\t-s\tscattern avoid deadlocks\n"
+               "\t-d\tavoid deadlocks\n"
                "\t-i\tinitialize database\n");
         return 1;
     }

Original file line number	Diff line number	Diff line change
`@@ -504,10 +504,22 @@ csn_t MtmDistributedTransactionSnapshot(TransactionId xid, int nodeId, nodemask_`
`504`	`504`	`return snapshot;`
`505`	`505`	`}`
`506`	`506`
	`507`	`+void MtmSetSnapshot(csn_t globalSnapshot)`
	`508`	`+{`
	`509`	`+ MtmLock(LW_EXCLUSIVE);`
	`510`	`+ MtmSyncClock(globalSnapshot);`
	`511`	`+ MtmTx.snapshot = globalSnapshot;`
	`512`	`+ MtmUnlock();`
	`513`	`+}`
`507`	`514`
	`515`	`+`
`508`	`516`	`Snapshot MtmGetSnapshot(Snapshot snapshot)`
`509`	`517`	`{`
`510`	`518`	`snapshot = PgGetSnapshotData(snapshot);`
	`519`	`+ if (XactIsoLevel == XACT_READ_COMMITTED && MtmTx.snapshot != INVALID_CSN && TransactionIdIsValid(GetCurrentTransactionIdIfAny())) {`
	`520`	`+ MtmTx.snapshot = MtmGetCurrentTime();`
	`521`	`+ LogLogicalMessage("S", (char*)&MtmTx.snapshot, sizeof(MtmTx.snapshot), true);`
	`522`	`+ }`
`511`	`523`	`RecentGlobalDataXmin = RecentGlobalXmin = Mtm->oldestXid;`
`512`	`524`	`return snapshot;`
`513`	`525`	`}`
`@@ -540,7 +552,7 @@ bool MtmXidInMVCCSnapshot(TransactionId xid, Snapshot snapshot)`
`540`	`552`
`541`	`553`	`Assert(xid != InvalidTransactionId);`
`542`	`554`
`543`		`- if (!MtmUseDtm) {`
	`555`	`+ if (!MtmUseDtm \|\| TransactionIdPrecedes(xid, Mtm->oldestXid)) {`
`544`	`556`	`return PgXidInMVCCSnapshot(xid, snapshot);`
`545`	`557`	`}`
`546`	`558`	`MtmLock(LW_SHARED);`
`@@ -681,6 +693,10 @@ MtmAdjustOldestXid(TransactionId xid)`
`681`	`693`	`hash_search(MtmGid2State, &prev->gid, HASH_REMOVE, NULL);`
`682`	`694`	`}`
`683`	`695`	`}`
	`696`	`+ if (ts != NULL) {`
	`697`	`+ MTM_LOG2("Adjust(%lld) stop at snashot %lld, xid %lld, pinned=%d, oldestSnaphsot=%lld\n",`
	`698`	`+ (long64)xid, ts->csn, (long64)ts->xid, ts->isPinned, oldestSnapshot);`
	`699`	`+ }`
`684`	`700`	`}`
`685`	`701`
`686`	`702`	`if (MtmUseDtm && !MtmVolksWagenMode)`
`@@ -2827,11 +2843,13 @@ static bool ConfigIsSane(void)`
`2827`	`2843`	`{`
`2828`	`2844`	`bool ok = true;`
`2829`	`2845`
	`2846`	`+#if 0`
`2830`	`2847`	`if (DefaultXactIsoLevel != XACT_REPEATABLE_READ)`
`2831`	`2848`	`{`
`2832`	`2849`	`MTM_ELOG(WARNING, "multimaster requires default_transaction_isolation = 'repeatable read'");`
`2833`	`2850`	`ok = false;`
`2834`	`2851`	`}`
	`2852`	`+#endif`
`2835`	`2853`
`2836`	`2854`	`if (MtmMaxNodes < 1)`
`2837`	`2855`	`{`
`@@ -5123,11 +5141,11 @@ static void MtmProcessUtility(Node parsetree, const char queryString,`
`5123`	`5141`	`standard_ProcessUtility(parsetree, queryString, context,`
`5124`	`5142`	`params, dest, completionTag);`
`5125`	`5143`	`}`
`5126`		`-`
	`5144`	`+#if 0`
`5127`	`5145`	`if (!MtmVolksWagenMode && MtmTx.isDistributed && XactIsoLevel != XACT_REPEATABLE_READ) {`
`5128`	`5146`	`MTM_ELOG(ERROR, "Isolation level %s is not supported by multimaster", isoLevelStr[XactIsoLevel]);`
`5129`	`5147`	`}`
`5130`		`-`
	`5148`	`+#endif`
`5131`	`5149`	`if (MyXactAccessedTempRel)`
`5132`	`5150`	`{`
`5133`	`5151`	`MTM_LOG1("Xact accessed temp table, stopping replication");`