SERVER-44393 split IndexBuildsCoordinator::_buildIndex() into hybrid index build phases

2 files changed, 97 insertions, 0 deletions

diff --git a/src/mongo/db/index_builds_coordinator.cpp b/src/mongo/db/index_builds_coordinator.cpp
index aea56abc00b..a1bed562111 100644
--- a/src/mongo/db/index_builds_coordinator.cpp
+++ b/src/mongo/db/index_builds_coordinator.cpp
@@ -1248,6 +1248,22 @@ void IndexBuildsCoordinator::_buildIndex(
     std::shared_ptr<ReplIndexBuildState> replState,
     const IndexBuildOptions& indexBuildOptions,
     boost::optional<Lock::CollectionLock>* exclusiveCollectionLock) {
+    _scanCollectionAndInsertKeysIntoSorter(opCtx, dbAndUUID, replState, exclusiveCollectionLock);
+    auto nss = _insertKeysFromSideTablesWithoutBlockingWrites(opCtx, dbAndUUID, replState);
+    auto commitIndexBuildTimestamp = _waitForCommitOrAbort(opCtx, nss, replState);
+    _insertKeysFromSideTablesAndCommit(opCtx,
+                                       dbAndUUID,
+                                       replState,
+                                       indexBuildOptions,
+                                       exclusiveCollectionLock,
+                                       commitIndexBuildTimestamp);
+}
+
+void IndexBuildsCoordinator::_scanCollectionAndInsertKeysIntoSorter(
+    OperationContext* opCtx,
+    const NamespaceStringOrUUID& dbAndUUID,
+    std::shared_ptr<ReplIndexBuildState> replState,
+    boost::optional<Lock::CollectionLock>* exclusiveCollectionLock) {
 
     {
         auto nss = CollectionCatalog::get(opCtx).lookupNSSByUUID(replState->collectionUUID);
@@ -1290,7 +1306,16 @@ void IndexBuildsCoordinator::_buildIndex(
         log() << "Hanging after dumping inserts from bulk builder";
         hangAfterIndexBuildDumpsInsertsFromBulk.pauseWhileSet();
     }
+}
 
+/**
+ * Second phase is extracting the sorted keys and writing them into the new index table.
+ * Looks up collection namespace while holding locks.
+ */
+NamespaceString IndexBuildsCoordinator::_insertKeysFromSideTablesWithoutBlockingWrites(
+    OperationContext* opCtx,
+    const NamespaceStringOrUUID& dbAndUUID,
+    std::shared_ptr<ReplIndexBuildState> replState) {
     // Perform the first drain while holding an intent lock.
     {
         opCtx->recoveryUnit()->abandonSnapshot();
@@ -1330,6 +1355,16 @@ void IndexBuildsCoordinator::_buildIndex(
         hangAfterIndexBuildSecondDrain.pauseWhileSet();
     }
 
+    return nss;
+}
+
+/**
+ * Waits for commit or abort signal from primary.
+ */
+Timestamp IndexBuildsCoordinator::_waitForCommitOrAbort(
+    OperationContext* opCtx,
+    const NamespaceString& nss,
+    std::shared_ptr<ReplIndexBuildState> replState) {
     Timestamp commitIndexBuildTimestamp;
     if (shouldWaitForCommitOrAbort(opCtx, nss, *replState)) {
         log() << "Index build waiting for commit or abort before completing final phase: "
@@ -1361,7 +1396,20 @@ void IndexBuildsCoordinator::_buildIndex(
             invariant(!replState->isCommitReady, replState->buildUUID.toString());
         }
     }
+    return commitIndexBuildTimestamp;
+}
 
+/**
+ * Third phase is catching up on all the writes that occurred during the first two phases.
+ * Accepts a commit timestamp for the index (null if not available).
+ */
+void IndexBuildsCoordinator::_insertKeysFromSideTablesAndCommit(
+    OperationContext* opCtx,
+    const NamespaceStringOrUUID& dbAndUUID,
+    std::shared_ptr<ReplIndexBuildState> replState,
+    const IndexBuildOptions& indexBuildOptions,
+    boost::optional<Lock::CollectionLock>* exclusiveCollectionLock,
+    const Timestamp& commitIndexBuildTimestamp) {
     // Need to return the collection lock back to exclusive mode, to complete the index build.
     opCtx->recoveryUnit()->abandonSnapshot();
     exclusiveCollectionLock->emplace(opCtx, dbAndUUID, MODE_X);
diff --git a/src/mongo/db/index_builds_coordinator.h b/src/mongo/db/index_builds_coordinator.h
index 0cc29db2155..3c909c3f2c9 100644
--- a/src/mongo/db/index_builds_coordinator.h
+++ b/src/mongo/db/index_builds_coordinator.h
@@ -419,6 +419,55 @@ protected:
                      std::shared_ptr<ReplIndexBuildState> replState,
                      const IndexBuildOptions& indexBuildOptions,
                      boost::optional<Lock::CollectionLock>* collLock);
+
+    /**
+     * First phase is the collection scan and insertion of the keys into the sorter.
+     */
+    void _scanCollectionAndInsertKeysIntoSorter(
+        OperationContext* opCtx,
+        const NamespaceStringOrUUID& dbAndUUID,
+        std::shared_ptr<ReplIndexBuildState> replState,
+        boost::optional<Lock::CollectionLock>* exclusiveCollectionLock);
+
+    /**
+     * Second phase is extracting the sorted keys and writing them into the new index table.
+     * On completion, this function returns the namespace of the collection, which may have changed
+     * after the previous phase. The namespace is used in two phase index builds to determine the
+     * current replication state in _waitForCommitOrAbort().
+     */
+    NamespaceString _insertKeysFromSideTablesWithoutBlockingWrites(
+        OperationContext* opCtx,
+        const NamespaceStringOrUUID& dbAndUUID,
+        std::shared_ptr<ReplIndexBuildState> replState);
+
+    /**
+     * Waits for commit or abort signal from primary.
+     *
+     * On completion, this function returns a timestamp, which may be null, that may be used to
+     * update the mdb catalog as we commit the index build. The commit index build timestamp is
+     * obtained from a commitIndexBuild oplog entry during secondary oplog application.
+     * This function returns a null timestamp on receiving a abortIndexBuild oplog entry; or if we
+     * are currently a primary, in which case we do not need to wait any external signal to commit
+     * the index build.
+     */
+    Timestamp _waitForCommitOrAbort(OperationContext* opCtx,
+                                    const NamespaceString& nss,
+                                    std::shared_ptr<ReplIndexBuildState> replState);
+
+    /**
+     * Third phase is catching up on all the writes that occurred during the first two phases.
+     * Accepts a commit timestamp for the index, which could be null. See _waitForCommitOrAbort()
+     * comments. This timestamp is used only for committing the index, which sets the ready flag to
+     * true, to the catalog; it is not used for the catch-up writes during the final drain phase.
+     */
+    void _insertKeysFromSideTablesAndCommit(
+        OperationContext* opCtx,
+        const NamespaceStringOrUUID& dbAndUUID,
+        std::shared_ptr<ReplIndexBuildState> replState,
+        const IndexBuildOptions& indexBuildOptions,
+        boost::optional<Lock::CollectionLock>* exclusiveCollectionLock,
+        const Timestamp& commitIndexBuildTimestamp);
+
     /**
      * Returns total number of indexes in collection, including unfinished/in-progress indexes.
      *