SERVER-57090 Investigate tenant dbhash check behavior with concurrent failovers

1 files changed, 27 insertions, 8 deletions

diff --git a/jstests/hooks/run_check_tenant_migration_dbhash.js b/jstests/hooks/run_check_tenant_migration_dbhash.js
index f826dfa12af..779ab9d1795 100644
--- a/jstests/hooks/run_check_tenant_migration_dbhash.js
+++ b/jstests/hooks/run_check_tenant_migration_dbhash.js
@@ -2,7 +2,7 @@
 'use strict';
 
 (function() {
-load('jstests/libs/discover_topology.js');  // For Topology and DiscoverTopology.
+
 load("jstests/replsets/libs/tenant_migration_util.js");
 
 const excludedDBs = ["testTenantMigration"];
@@ -10,17 +10,36 @@ const testDBName = "testTenantMigration";
 const dbhashCollName = "dbhashCheck";
 const tenantId = TestData.tenantId;
 const migrationId = UUID(TestData.migrationIdString);
+const retryOnNetworkErrors = TestData.networkErrorAndTxnOverrideConfig &&
+    TestData.networkErrorAndTxnOverrideConfig.retryOnNetworkErrors;
 
-const donorRst = new ReplSetTest(TestData.donorConnectionString);
-const recipientRst = new ReplSetTest(TestData.recipientConnectionString);
+let donorRst;
+let recipientRst;
+let donorDB;
+while (true) {
+    try {
+        donorRst = new ReplSetTest(TestData.donorConnectionString);
+        recipientRst = new ReplSetTest(TestData.recipientConnectionString);
 
-const donorPrimary = donorRst.getPrimary();
-const donorPrimaryDB = donorPrimary.getDB(testDBName);
+        // In failover suites, we want to allow retryable writes in case there is a failover while
+        // running the tenant dbhash check. In non-failover suites we dont expect to see any
+        // failovers, but we run in a session to keep the code simple.
+        donorDB =
+            new Mongo(donorRst.getURL()).startSession({retryWrites: true}).getDatabase(testDBName);
+        break;
+    } catch (e) {
+        if (!TenantMigrationUtil.checkIfRetriableErrorForTenantDbHashCheck(e)) {
+            throw e;
+        }
+        print(`Got error: ${tojson(e)}. Retrying ReplSetTest setup on retriable error.`);
+    }
+}
 
 // We assume every db is under the tenant being migrated.
-TenantMigrationUtil.checkTenantDBHashes(donorRst, recipientRst, tenantId, excludedDBs);
+TenantMigrationUtil.checkTenantDBHashes(
+    donorRst, recipientRst, tenantId, excludedDBs, retryOnNetworkErrors);
 
 // Mark that we have completed the dbhash check.
-assert.commandWorked(
-    donorPrimaryDB[dbhashCollName].insert([{_id: migrationId}], {writeConcern: {w: "majority"}}));
+assert.commandWorked(donorDB.runCommand(
+    {insert: dbhashCollName, documents: [{_id: migrationId}], writeConcern: {w: "majority"}}));
 })();