SERVER-64730 Interrupt ongoing refreshes after entering into the critical section

(cherry picked from commit 343108041c5b3570e97418ee3204804535fbde4d)

5 files changed, 22 insertions, 8 deletions

diff --git a/jstests/sharding/libs/resharding_test_fixture.js b/jstests/sharding/libs/resharding_test_fixture.js
index ac6efd7b0d1..815eeb5bbc9 100644
--- a/jstests/sharding/libs/resharding_test_fixture.js
+++ b/jstests/sharding/libs/resharding_test_fixture.js
@@ -627,7 +627,11 @@ var ReshardingTest = class {
 
     /** @private */
     _checkConsistency() {
-        const nsCursor = this._st.s.getCollection(this._ns).find().sort({_id: 1});
+        // The "available" read concern level won't block this find cmd behind the critical section.
+        // Tests for resharding are not expected to have unowned documents in the collection being
+        // resharded.
+        const nsCursor =
+            this._st.s.getCollection(this._ns).find().readConcern("available").sort({_id: 1});
         const tempNsCursor = this._st.s.getCollection(this._tempNs).find().sort({_id: 1});
 
         const diff = ((diff) => {
diff --git a/src/mongo/db/s/collection_sharding_runtime.cpp b/src/mongo/db/s/collection_sharding_runtime.cpp
index 8334f99b74b..ccf7c28de65 100644
--- a/src/mongo/db/s/collection_sharding_runtime.cpp
+++ b/src/mongo/db/s/collection_sharding_runtime.cpp
@@ -163,6 +163,10 @@ void CollectionShardingRuntime::checkShardVersionOrThrow(OperationContext* opCtx
 void CollectionShardingRuntime::enterCriticalSectionCatchUpPhase(const CSRLock&,
                                                                  const BSONObj& reason) {
     _critSec.enterCriticalSectionCatchUpPhase(reason);
+
+    if (_shardVersionInRecoverOrRefresh) {
+        _shardVersionInRecoverOrRefresh->cancellationSource.cancel();
+    }
 }
 
 void CollectionShardingRuntime::enterCriticalSectionCommitPhase(const CSRLock&,
diff --git a/src/mongo/db/s/collection_sharding_runtime.h b/src/mongo/db/s/collection_sharding_runtime.h
index d97deb9bee6..9d35dded956 100644
--- a/src/mongo/db/s/collection_sharding_runtime.h
+++ b/src/mongo/db/s/collection_sharding_runtime.h
@@ -113,10 +113,10 @@ public:
      * Marks the collection's filtering metadata as UNKNOWN, meaning that all attempts to check for
      * shard version match will fail with StaleConfig errors in order to trigger an update.
      *
+     * Interrupts any ongoing shard metadata refresh.
+     *
      * It is safe to call this method with only an intent lock on the collection (as opposed to
-     * setFilteringMetadata which requires exclusive), however note that clearing a collection's
-     * filtering metadata will interrupt all in-progress orphan cleanups in which case orphaned data
-     * will remain behind on disk.
+     * setFilteringMetadata which requires exclusive).
      */
     void clearFilteringMetadata(OperationContext* opCtx);
 
@@ -125,6 +125,8 @@ public:
      * with both the collection lock and CSRLock held in exclusive mode.
      *
      * In these methods, the CSRLock ensures concurrent access to the critical section.
+     *
+     * Entering into the Critical Section interrupts any ongoing filtering metadata refresh.
      */
     void enterCriticalSectionCatchUpPhase(const CSRLock&, const BSONObj& reason);
     void enterCriticalSectionCommitPhase(const CSRLock&, const BSONObj& reason);
diff --git a/src/mongo/db/s/recoverable_critical_section_service.h b/src/mongo/db/s/recoverable_critical_section_service.h
index d1953b27942..d9ef1c75195 100644
--- a/src/mongo/db/s/recoverable_critical_section_service.h
+++ b/src/mongo/db/s/recoverable_critical_section_service.h
@@ -57,6 +57,8 @@ public:
      * specified namespace and reason. It works even if the namespace's current metadata are
      * UNKNOWN.
      *
+     * Entering into the Critical Section interrupts any ongoing filtering metadata refresh.
+     *
      * It adds a doc to config.collectionCriticalSections with with writeConcern write concern.
      *
      * Do nothing if the collection critical section is taken for that nss and reason, and will
@@ -87,7 +89,9 @@ public:
     /**
      * Releases the recoverable critical section for the given nss and reason.
      *
-     * It removes a doc from config.collectionCriticalSections with writeConcern write concern.
+     * It removes a doc from config.collectionCriticalSections with writeConcern write concern. As
+     * part of the removal, the filtering information is cleared on secondary nodes. It is
+     * responsability of the caller to properly set the filtering information on the primary node.
      *
      * Do nothing if the collection critical section is not taken for that nss and reason.
      */
diff --git a/src/mongo/db/s/shard_filtering_metadata_refresh.cpp b/src/mongo/db/s/shard_filtering_metadata_refresh.cpp
index 0a7cecfe632..802bddbc016 100644
--- a/src/mongo/db/s/shard_filtering_metadata_refresh.cpp
+++ b/src/mongo/db/s/shard_filtering_metadata_refresh.cpp
@@ -115,8 +115,7 @@ bool joinShardVersionOperation(OperationContext* opCtx,
             try {
                 inRecoverOrRefresh->get(opCtx);
             } catch (const ExceptionFor<ErrorCodes::ShardVersionRefreshCanceled>&) {
-                // The ongoing refresh has finished, although it was canceled by a
-                // 'clearFilteringMetadata'.
+                // The ongoing refresh has finished, although it was interrupted.
             }
         }
 
@@ -223,7 +222,8 @@ SharedSemiFuture<void> recoverRefreshShardVersion(ServiceContext* serviceContext
             if (cancellationToken.isCanceled() &&
                 (status.isOK() || status == ErrorCodes::Interrupted)) {
                 uasserted(ErrorCodes::ShardVersionRefreshCanceled,
-                          "Shard version refresh canceled by a 'clearFilteringMetadata'");
+                          "Shard version refresh canceled by an interruption, probably due to a "
+                          "'clearFilteringMetadata'");
             }
             return status;
         })