SERVER-55557 Range deletion of aborted migration can fail after a refine shard key

4 files changed, 30 insertions, 2 deletions

diff --git a/src/mongo/base/error_codes.yml b/src/mongo/base/error_codes.yml
index 061dc16d8ff..5208c434cb4 100644
--- a/src/mongo/base/error_codes.yml
+++ b/src/mongo/base/error_codes.yml
@@ -440,6 +440,8 @@ error_codes:
     
     - {code: 348,name: ChangeStreamTopologyChange, extra: ChangeStreamTopologyChangeInfo}
 
+    - {code: 349, name: KeyPatternShorterThanBound}
+
     # Error codes 4000-8999 are reserved.
 
     # Non-sequential error codes for compatibility only)
diff --git a/src/mongo/db/keypattern.cpp b/src/mongo/db/keypattern.cpp
index bba2f3424e0..8e91e561c01 100644
--- a/src/mongo/db/keypattern.cpp
+++ b/src/mongo/db/keypattern.cpp
@@ -79,7 +79,7 @@ BSONObj KeyPattern::extendRangeBound(const BSONObj& bound, bool makeUpperInclusi
     BSONObjIterator pat(_pattern);
 
     while (src.more()) {
-        massert(16649,
+        massert(ErrorCodes::KeyPatternShorterThanBound,
                 str::stream() << "keyPattern " << _pattern << " shorter than bound " << bound,
                 pat.more());
         BSONElement srcElt = src.next();
diff --git a/src/mongo/db/s/migration_util.cpp b/src/mongo/db/s/migration_util.cpp
index e766e7ec99e..c6578126cf3 100644
--- a/src/mongo/db/s/migration_util.cpp
+++ b/src/mongo/db/s/migration_util.cpp
@@ -408,7 +408,32 @@ ExecutorFuture<void> submitRangeDeletionTask(OperationContext* opCtx,
                 }
             }
 
-            return cleanUpRange(serviceContext, executor, deletionTask);
+            return AsyncTry([=]() {
+                       return cleanUpRange(serviceContext, executor, deletionTask)
+                           .onError<ErrorCodes::KeyPatternShorterThanBound>([=](Status status) {
+                               ThreadClient tc(kRangeDeletionThreadName, serviceContext);
+                               {
+                                   stdx::lock_guard<Client> lk(*tc.get());
+                                   tc->setSystemOperationKillableByStepdown(lk);
+                               }
+                               auto uniqueOpCtx = tc->makeOperationContext();
+                               uniqueOpCtx->setAlwaysInterruptAtStepDownOrUp();
+
+                               LOGV2(55557,
+                                     "cleanUpRange failed due to keyPattern shorter than range "
+                                     "deletion bounds. Refreshing collection metadata to retry.",
+                                     "nss"_attr = deletionTask.getNss(),
+                                     "status"_attr = redact(status));
+
+                               onShardVersionMismatch(
+                                   uniqueOpCtx.get(), deletionTask.getNss(), boost::none);
+
+                               return status;
+                           });
+                   })
+                .until(
+                    [](Status status) { return status != ErrorCodes::KeyPatternShorterThanBound; })
+                .on(executor, CancellationToken::uncancelable());
         })
         .onError([=](const Status status) {
             ThreadClient tc(kRangeDeletionThreadName, serviceContext);
diff --git a/src/mongo/db/s/range_deletion_util.cpp b/src/mongo/db/s/range_deletion_util.cpp
index a8d29853478..8cf10df7718 100644
--- a/src/mongo/db/s/range_deletion_util.cpp
+++ b/src/mongo/db/s/range_deletion_util.cpp
@@ -352,6 +352,7 @@ ExecutorFuture<void> deleteRangeInBatches(const std::shared_ptr<executor::TaskEx
                 ErrorCodes::RangeDeletionAbandonedBecauseCollectionWithUUIDDoesNotExist ||
                 swNumDeleted.getStatus() ==
                 ErrorCodes::RangeDeletionAbandonedBecauseTaskDocumentDoesNotExist ||
+                swNumDeleted.getStatus().code() == ErrorCodes::KeyPatternShorterThanBound ||
                 ErrorCodes::isShutdownError(swNumDeleted.getStatus()) ||
                 ErrorCodes::isNotPrimaryError(swNumDeleted.getStatus());
         })