From a22f3c599e1e588cdb5916785f27af7cd5493d18 Mon Sep 17 00:00:00 2001
From: Tommaso Tocci <tommaso.tocci@mongodb.com>
Date: Fri, 4 Feb 2022 09:04:29 +0000
Subject: SERVER-63207  Speedup very slow move_primary_with_writes.js test

---
 jstests/sharding/move_primary_with_writes.js | 35 ++++++++++++++++++++++------
 src/mongo/db/s/sharding_ddl_coordinator.cpp  | 20 ++++++++++++----
 2 files changed, 44 insertions(+), 11 deletions(-)

diff --git a/jstests/sharding/move_primary_with_writes.js b/jstests/sharding/move_primary_with_writes.js
index b31647b5789..09783f8dc1d 100644
--- a/jstests/sharding/move_primary_with_writes.js
+++ b/jstests/sharding/move_primary_with_writes.js
@@ -4,7 +4,7 @@
 (function() {
 'use strict';
 
-load('jstests/sharding/libs/sharded_transactions_helpers.js');
+load('jstests/libs/fail_point_util.js');
 
 let st = new ShardingTest({
     mongos: 2,
@@ -177,15 +177,18 @@ function buildDDLCommands(collName) {
 }
 
 function testMovePrimary(failpoint, fromShard, toShard, db, shouldFail, sharded) {
+    jsTestLog("Testing move primary with FP: " + failpoint + " shouldFail: " + shouldFail +
+              " sharded: " + sharded);
+
     let codeToRunInParallelShell = '{ db.getSiblingDB("admin").runCommand({movePrimary: "' +
         dbName + '", to: "' + toShard.name + '"}); }';
 
-    assert.commandWorked(fromShard.adminCommand({configureFailPoint: failpoint, mode: 'alwaysOn'}));
+    let fp = configureFailPoint(fromShard, failpoint);
 
     let awaitShell = startParallelShell(codeToRunInParallelShell, st.s.port);
 
     jsTestLog("Waiting for failpoint " + failpoint);
-    waitForFailpoint("Hit " + failpoint, 1);
+    fp.wait();
     clearRawMongoProgramOutput();
 
     // Test DML
@@ -216,21 +219,24 @@ function testMovePrimary(failpoint, fromShard, toShard, db, shouldFail, sharded)
         }
     });
 
-    assert.commandWorked(fromShard.adminCommand({configureFailPoint: failpoint, mode: 'off'}));
+    fp.off();
 
     awaitShell();
 }
 
 function testMovePrimaryDDL(failpoint, fromShard, toShard, db, shouldFail, sharded) {
+    jsTest.log("Testing move primary DDL with FP: " + failpoint + " shouldFail: " + shouldFail +
+               " sharded: " + sharded);
+
     let codeToRunInParallelShell = '{ db.getSiblingDB("admin").runCommand({movePrimary: "' +
         dbName + '", to: "' + toShard.name + '"}); }';
 
-    assert.commandWorked(fromShard.adminCommand({configureFailPoint: failpoint, mode: 'alwaysOn'}));
+    let fp = configureFailPoint(fromShard, failpoint);
 
     let awaitShell = startParallelShell(codeToRunInParallelShell, st.s.port);
 
     jsTestLog("Waiting for failpoint " + failpoint);
-    waitForFailpoint("Hit " + failpoint, 1);
+    fp.wait();
     clearRawMongoProgramOutput();
 
     let collName;
@@ -253,11 +259,24 @@ function testMovePrimaryDDL(failpoint, fromShard, toShard, db, shouldFail, shard
         }
     });
 
-    assert.commandWorked(fromShard.adminCommand({configureFailPoint: failpoint, mode: 'off'}));
+    fp.off();
 
     awaitShell();
 }
 
+// Reduce DDL lock timeout to half a second to speedup testing command that are expected to fail
+// with lockbusy error
+let overrideDDLLockTimeoutFPs = [];
+st.forEachConnection(shard => {
+    try {
+        overrideDDLLockTimeoutFPs.push(
+            configureFailPoint(shard, "overrideDDLLockTimeout", {'timeoutMillisecs': 500}));
+    } catch (e) {
+        // The failpoint has been added in 5.3 so multiversion suite will fail to set this failpoint
+        jsTestLog("Failed to override DDL lock timeout: " + e);
+    }
+});
+
 createCollections();
 let fromShard = st.getPrimaryShard(dbName);
 let toShard = st.getOther(fromShard);
@@ -284,5 +303,7 @@ fromShard = st.getPrimaryShard(dbName);
 toShard = st.getOther(fromShard);
 testMovePrimary('hangInCleanStaleDataStage', fromShard, toShard, st.s.getDB(dbName), false, false);
 
+overrideDDLLockTimeoutFPs.forEach(fp => fp.off());
+
 st.stop();
 })();
diff --git a/src/mongo/db/s/sharding_ddl_coordinator.cpp b/src/mongo/db/s/sharding_ddl_coordinator.cpp
index 15e875918ac..e6da6be570a 100644
--- a/src/mongo/db/s/sharding_ddl_coordinator.cpp
+++ b/src/mongo/db/s/sharding_ddl_coordinator.cpp
@@ -50,6 +50,7 @@
 namespace mongo {
 
 MONGO_FAIL_POINT_DEFINE(hangBeforeRunningCoordinatorInstance);
+MONGO_FAIL_POINT_DEFINE(overrideDDLLockTimeout);
 
 namespace {
 
@@ -143,12 +144,23 @@ ExecutorFuture<void> ShardingDDLCoordinator::_acquireLockAsync(
 
                const auto coorName = DDLCoordinatorType_serializer(_coordId.getOperationType());
 
-               auto distLock = distLockManager->lockDirectLocally(
-                   opCtx, resource, DistLockManager::kDefaultLockTimeout);
+               const auto lockTimeOut = [&]() -> Milliseconds {
+                   if (auto sfp = overrideDDLLockTimeout.scoped(); MONGO_unlikely(sfp.isActive())) {
+                       if (auto timeoutElem = sfp.getData()["timeoutMillisecs"]; timeoutElem.ok()) {
+                           const auto timeoutMillisecs = Milliseconds(timeoutElem.safeNumberLong());
+                           LOGV2(6320700,
+                                 "Overriding DDL lock timeout",
+                                 "timeout"_attr = timeoutMillisecs);
+                           return timeoutMillisecs;
+                       }
+                   }
+                   return DistLockManager::kDefaultLockTimeout;
+               }();
+
+               auto distLock = distLockManager->lockDirectLocally(opCtx, resource, lockTimeOut);
                _scopedLocks.emplace(std::move(distLock));
 
-               uassertStatusOK(distLockManager->lockDirect(
-                   opCtx, resource, coorName, DistLockManager::kDefaultLockTimeout));
+               uassertStatusOK(distLockManager->lockDirect(opCtx, resource, coorName, lockTimeOut));
            })
         .until([this](Status status) { return (!_recoveredFromDisk) || status.isOK(); })
         .withBackoffBetweenIterations(kExponentialBackoff)
-- 
cgit v1.2.1