SERVER-42864 change index build initial write timestamp logic

(cherry picked from commit 99593e1255dde3320ac330599311e3481a1ac0e0)

1 files changed, 22 insertions, 10 deletions

diff --git a/src/mongo/db/catalog/multi_index_block.cpp b/src/mongo/db/catalog/multi_index_block.cpp
index 1e7c8795349..76dad0efcc2 100644
--- a/src/mongo/db/catalog/multi_index_block.cpp
+++ b/src/mongo/db/catalog/multi_index_block.cpp
@@ -233,17 +233,29 @@ MultiIndexBlock::OnInitFn MultiIndexBlock::kNoopOnInitFn =
 MultiIndexBlock::OnInitFn MultiIndexBlock::makeTimestampedIndexOnInitFn(OperationContext* opCtx,
                                                                         const Collection* coll) {
     return [opCtx, ns = coll->ns()](std::vector<BSONObj>& specs) -> Status {
-        auto replCoord = repl::ReplicationCoordinator::get(opCtx);
-        if (opCtx->recoveryUnit()->getCommitTimestamp().isNull() &&
-            replCoord->canAcceptWritesForDatabase(opCtx, "admin")) {
-            // Only primaries must timestamp this write. Secondaries run this from within a
-            // `TimestampBlock`. Primaries performing an index build via `applyOps` may have a
-            // wrapping commit timestamp that will be used instead.
-            opCtx->getServiceContext()->getOpObserver()->onOpMessage(
-                opCtx,
-                BSON("msg" << std::string(str::stream() << "Creating indexes. Coll: " << ns)));
-        }
+        // This function sets a timestamp for the initial catalog write when beginning an index
+        // build, if necessary.  There are four scenarios:
 
+        // 1. A timestamp is already set -- replication application sets a timestamp ahead of time.
+        // This could include the phase of initial sync where it applies oplog entries.  Also,
+        // primaries performing an index build via `applyOps` may have a wrapping commit timestamp.
+        if (!opCtx->recoveryUnit()->getCommitTimestamp().isNull())
+            return Status::OK();
+
+        // 2. If the node is initial syncing, we do not set a timestamp.
+        auto replCoord = repl::ReplicationCoordinator::get(opCtx);
+        if (replCoord->isReplEnabled() && replCoord->getMemberState().startup2())
+            return Status::OK();
+
+        // 3. If the index build is on the local database, do not timestamp.
+        if (ns.isLocal())
+            return Status::OK();
+
+        // 4. All other cases, we generate a timestamp by writing a no-op oplog entry.  This is
+        // better than using a ghost timestamp.  Writing an oplog entry ensures this node is
+        // primary.
+        opCtx->getServiceContext()->getOpObserver()->onOpMessage(
+            opCtx, BSON("msg" << std::string(str::stream() << "Creating indexes. Coll: " << ns)));
         return Status::OK();
     };
 }