Ignore index_wrong_build_id error in correctness. (#121)

Also

* UpdateIndexPlan should always get a valid buildId
* Adding few more TraceEvent on error messages.

Author: apkar

src/MetadataManager.actor.cpp

@@ -254,12 +254,13 @@ ACTOR static Future<Void> buildIndex_impl(bson::BSONObj indexObj,
 
 		state Reference<Plan> finalizePlan = ec->isolatedWrapOperationPlan(
 		    ref(new UpdateIndexStatusPlan(ns, encodedIndexId, ec->mm,
-		                                  std::string(DocLayerConstants::INDEX_STATUS_READY), info.buildId)),
+		                                  std::string(DocLayerConstants::INDEX_STATUS_READY), build_id)),
 		    0, -1);
 		int64_t _ = wait(executeUntilCompletionTransactionally(finalizePlan, ec->getOperationTransaction()));
 
 		return Void();
 	} catch (Error& e) {
+		TraceEvent(SevError, "indexRebuildFailed").error(e);
 		state Error err = e;
 		// try forever to set the index into an error status (unless somebody comes along before us and starts a
 		// different build)
@@ -270,7 +271,7 @@ ACTOR static Future<Void> buildIndex_impl(bson::BSONObj indexObj,
 			// buildId field does not exist (as is the case for 'ready' indexes).
 			state Reference<Plan> errorPlan = ec->isolatedWrapOperationPlan(
 			    ref(new UpdateIndexStatusPlan(ns, encodedIndexId, ec->mm,
-			                                  std::string(DocLayerConstants::INDEX_STATUS_ERROR), info.buildId)),
+			                                  std::string(DocLayerConstants::INDEX_STATUS_ERROR), build_id)),
 			    0, -1);
 			try {
 				int64_t _ = wait(executeUntilCompletionTransactionally(errorPlan, ec->getOperationTransaction()));

src/QLPlan.actor.cpp

@@ -1552,7 +1552,7 @@ ACTOR static Future<Void> updateIndexStatus(PlanCheckpoint* checkpoint,
                                             Standalone<StringRef> encodedIndexId,
                                             Reference<MetadataManager> mm,
                                             std::string newStatus,
-                                            Optional<UID> buildId,
+                                            UID buildId,
                                             PromiseStream<Reference<ScanReturnedContext>> output) {
 	state bool okay;
 	state FlowLock* flowControlLock = checkpoint->getDocumentFinishedLock();
@@ -1563,21 +1563,23 @@ ACTOR static Future<Void> updateIndexStatus(PlanCheckpoint* checkpoint,
 		    indexCollection->bindCollectionContext(tr)->cx->getSubContext(encodedIndexId);
 		Reference<UnboundCollectionContext> ucx = wait(mm->getUnboundCollectionContext(tr, ns));
 		state Reference<CollectionContext> mcx = ucx->bindCollectionContext(tr);
-		if (buildId.present()) {
-			Optional<DataValue> dv =
-			    wait(indexDoc->get(DataValue(DocLayerConstants::BUILD_ID_FIELD, DVTypeCode::STRING).encode_key_part()));
-			if (dv.present()) {
-				UID currId = UID::fromString(dv.get().getString());
-				if (currId == buildId.get()) {
-					okay = true;
-				} else {
-					okay = false;
-				}
+		Optional<DataValue> dv =
+		    wait(indexDoc->get(DataValue(DocLayerConstants::BUILD_ID_FIELD, DVTypeCode::STRING).encode_key_part()));
+		if (dv.present()) {
+			UID currId = UID::fromString(dv.get().getString());
+			if (currId == buildId) {
+				okay = true;
 			} else {
+				TraceEvent(SevError, "MismatchedIndexBuildID")
+				    .detail("indexDoc", indexDoc->toDbgString())
+				    .detail("newStatus", newStatus);
 				okay = false;
 			}
 		} else {
-			okay = true;
+			TraceEvent(SevError, "MissingIndexBuildID")
+			    .detail("indexDoc", indexDoc->toDbgString())
+			    .detail("newStatus", newStatus);
+			okay = false;
 		}
 
 		if (okay) {

src/QLPlan.h

@@ -669,13 +669,13 @@ struct UpdateIndexStatusPlan : ConcretePlan<UpdateIndexStatusPlan> {
 	Standalone<StringRef> encodedIndexId;
 	Reference<MetadataManager> mm;
 	std::string newStatus;
-	Optional<UID> buildId;
+	UID buildId;
 
 	UpdateIndexStatusPlan(Namespace const& ns,
 	                      Standalone<StringRef> encodedIndexId,
 	                      Reference<MetadataManager> mm,
 	                      std::string newStatus,
-	                      Optional<UID> buildId = Optional<UID>())
+	                      UID buildId)
 	    : ns(ns), encodedIndexId(encodedIndexId), mm(mm), newStatus(newStatus), buildId(buildId) {}
 
 	bson::BSONObj describe() override {

test/correctness/document-correctness.py

@@ -424,9 +424,12 @@ def _run_operation_(op1, op2):
 
 ignored_exceptions = [
     "Multi-multikey index size exceeds maximum value",
-    "key too large to index", # it's hard to estimate the exact byte size of KVS key to be inserted, ignore for now.
+    "key too large to index",  # it's hard to estimate the exact byte size of KVS key to be inserted, ignore for now.
     "Key length exceeds limit",
     "Operation aborted because the transaction timed out",
+    # This is another variant of "Operation aborted" error. For now, ignoring this error. We have to fix this part of
+    # transaction handling redesign.
+    "Attempting to change status of a different index build",
 ]