SERVER-52722 Handle cloning restart in TenantMigrationRecipientService

1 files changed, 12 insertions, 12 deletions

diff --git a/jstests/replsets/tenant_migration_resume_collection_cloner_after_rename.js b/jstests/replsets/tenant_migration_resume_collection_cloner_after_rename.js
index 7f89c304df2..7e9a602f9a5 100644
--- a/jstests/replsets/tenant_migration_resume_collection_cloner_after_rename.js
+++ b/jstests/replsets/tenant_migration_resume_collection_cloner_after_rename.js
@@ -44,8 +44,9 @@ const docs = [{_id: 0}, {_id: "string"}, {_id: UUID()}, {_id: new Date()}];
 tenantMigrationTest.insertDonorDB(dbName, collName, docs);
 
 const migrationId = UUID();
+const migrationIdString = extractUUIDFromObject(migrationId);
 const migrationOpts = {
-    migrationIdString: extractUUIDFromObject(migrationId),
+    migrationIdString: migrationIdString,
     recipientConnString: tenantMigrationTest.getRecipientConnString(),
     tenantId: tenantId,
 };
@@ -66,31 +67,30 @@ migrationThread.start();
 hangDuringCollectionClone.wait();
 assert.eq(2, recipientColl.find().itcount());
 
+// Insert some documents that will be fetched by the recipient. This is to test that on failover,
+// the fetcher will resume fetching from where it left off. The system is expected to crash if
+// the recipient fetches a duplicate oplog entry upon resuming the migration.
+tenantMigrationTest.insertDonorDB(dbName, "aNewColl", [{_id: "docToBeFetched"}]);
+assert.soon(() => {
+    const configDb = recipientPrimary.getDB("config");
+    const oplogBuffer = configDb.getCollection("repl.migration.oplog_" + migrationIdString);
+    return oplogBuffer.find({"entry.o._id": "docToBeFetched"}).count() === 1;
+});
+
 // Step up a new node in the recipient set and trigger a failover. The new primary should resume
 // cloning starting from the third document.
 recipientRst.awaitLastOpCommitted();
 const newRecipientPrimary = recipientRst.getSecondaries()[0];
 const newRecipientDb = newRecipientPrimary.getDB(dbName);
-const hangAfterStartingOplogFetcher = configureFailPoint(
-    newRecipientDb, "fpAfterStartingOplogFetcherMigrationRecipientInstance", {action: "hang"});
 assert.commandWorked(newRecipientPrimary.adminCommand({replSetStepUp: 1}));
 hangDuringCollectionClone.off();
 recipientRst.getPrimary();
 
-// Wait until the new recipient primary starts the oplog fetcher. This failpoint is needed only
-// because tenant migration didn't fully support resuming after a failover (specifically, not
-// preserving the startApplyingDonorOpTime) when this test was written. Otherwise, we can simply
-// rename the collection on the donor before stepping up the new recipient primary.
-hangAfterStartingOplogFetcher.wait();
-
 // Rename the collection on the donor.
 const donorColl = donorPrimary.getDB(dbName).getCollection(collName);
 const collNameRenamed = collName + "_renamed";
 assert.commandWorked(donorColl.renameCollection(collNameRenamed));
 
-// Resume collection cloning on the new recipient primary.
-hangAfterStartingOplogFetcher.off();
-
 // The migration should go through after recipient failover.
 assert.commandWorked(migrationThread.returnData());