feat(NODE-7122): exponential backoff between retries in convenient transaction API

src/sessions.ts

@@ -1,3 +1,5 @@
+import { setTimeout } from 'timers/promises';
+
 import { Binary, type Document, Long, type Timestamp } from './bson';
 import type { CommandOptions, Connection } from './cmap/connection';
 import { ConnectionPoolMetrics } from './cmap/metrics';
@@ -729,10 +731,10 @@ export class ClientSession
     const startTime = this.timeoutContext?.csotEnabled() ? this.timeoutContext.start : now();
 
     let committed = false;
-    let result: any;
+    let result: T;
 
     try {
-      while (!committed) {
+      for (let retry = 0; !committed; ++retry) {
         this.startTransaction(options); // may throw on error
 
         try {
@@ -768,7 +770,7 @@ export class ClientSession
 
           if (
             fnError.hasErrorLabel(MongoErrorLabel.TransientTransactionError) &&
-            (this.timeoutContext != null || now() - startTime < MAX_TIMEOUT)
+            (this.timeoutContext?.csotEnabled() || now() - startTime < MAX_TIMEOUT)
           ) {
             continue;
           }
@@ -786,32 +788,57 @@ export class ClientSession
             await this.commitTransaction();
             committed = true;
           } catch (commitError) {
-            /*
-             * Note: a maxTimeMS error will have the MaxTimeMSExpired
-             * code (50) and can be reported as a top-level error or
-             * inside writeConcernError, ex.
-             * { ok:0, code: 50, codeName: 'MaxTimeMSExpired' }
-             * { ok:1, writeConcernError: { code: 50, codeName: 'MaxTimeMSExpired' } }
-             */
-            if (
-              !isMaxTimeMSExpiredError(commitError) &&
-              commitError.hasErrorLabel(MongoErrorLabel.UnknownTransactionCommitResult) &&
-              (this.timeoutContext != null || now() - startTime < MAX_TIMEOUT)
-            ) {
-              continue;
-            }
-
-            if (
-              commitError.hasErrorLabel(MongoErrorLabel.TransientTransactionError) &&
-              (this.timeoutContext != null || now() - startTime < MAX_TIMEOUT)
-            ) {
-              break;
+            // If CSOT is enabled, we repeatedly retry until timeoutMS expires.  This is enforced by providing a
+            // timeoutContext to each async API, which know how to cancel themselves (i.e., the next retry will
+            // abort the withTransaction call).
+            // If CSOT is not enabled, do we still have time remaining or have we timed out?
+            const hasTimedOut =
+              !this.timeoutContext?.csotEnabled() && now() - startTime >= MAX_TIMEOUT;
+
+            if (!hasTimedOut) {
+              if (
+                !isMaxTimeMSExpiredError(commitError) &&
+                commitError.hasErrorLabel(MongoErrorLabel.UnknownTransactionCommitResult)
+              ) {
+                /*
+                 * Note: a maxTimeMS error will have the MaxTimeMSExpired
+                 * code (50) and can be reported as a top-level error or
+                 * inside writeConcernError, ex.
+                 * { ok:0, code: 50, codeName: 'MaxTimeMSExpired' }
+                 * { ok:1, writeConcernError: { code: 50, codeName: 'MaxTimeMSExpired' } }
+                 */
+                continue;
+              }
+
+              if (commitError.hasErrorLabel(MongoErrorLabel.TransientTransactionError)) {
+                const BACKOFF_INITIAL_MS = 5;
+                const BACKOFF_MAX_MS = 500;
+                const BACKOFF_GROWTH = 1.5;
+                const jitter = Math.random();
+                const backoffMS =
+                  jitter * Math.min(BACKOFF_INITIAL_MS * BACKOFF_GROWTH ** retry, BACKOFF_MAX_MS);
+
+                const willExceedTransactionDeadline =
+                  (this.timeoutContext?.csotEnabled() &&
+                    backoffMS > this.timeoutContext.remainingTimeMS) ||
+                  now() + backoffMS > startTime + MAX_TIMEOUT;
+
+                if (willExceedTransactionDeadline) {
+                  break;
+                }
+
+                await setTimeout(backoffMS);
+
+                break;
+              }
             }
 
             throw commitError;
           }
         }
       }
+
+      // @ts-expect-error Result is always defined if we reach here, the for-loop above convinces TS it is not.
       return result;
     } finally {
       this.timeoutContext = null;

test/integration/transactions-convenient-api/transactions-convenient-api.prose.test.ts

@@ -0,0 +1,87 @@
+import { expect } from 'chai';
+import { test } from 'mocha';
+import * as sinon from 'sinon';
+
+import { type ClientSession, type Collection, type MongoClient } from '../../../src';
+import { configureFailPoint, type FailCommandFailPoint, measureDuration } from '../../tools/utils';
+
+const failCommand: FailCommandFailPoint = {
+  configureFailPoint: 'failCommand',
+  mode: {
+    times: 13
+  },
+  data: {
+    failCommands: ['commitTransaction'],
+    errorCode: 251 // no such transaction
+  }
+};
+
+describe('Retry Backoff is Enforced', function () {
+  // 1. let client be a MongoClient
+  let client: MongoClient;
+
+  // 2. let coll be a collection
+  let collection: Collection;
+
+  beforeEach(async function () {
+    client = this.configuration.newClient();
+    collection = client.db('foo').collection('bar');
+  });
+
+  afterEach(async function () {
+    sinon.restore();
+    await client?.close();
+  });
+
+  test(
+    'works',
+    {
+      requires: {
+        mongodb: '>=4.4', // failCommand
+        topology: '!single' // transactions can't run on standalone servers
+      }
+    },
+    async function () {
+      const randomStub = sinon.stub(Math, 'random');
+
+      // 3.i Configure the random number generator used for jitter to always return 0
+      randomStub.returns(0);
+
+      // 3.ii Configure a fail point that forces 13 retries
+      await configureFailPoint(this.configuration, failCommand);
+
+      // 3.iii
+      const callback = async (s: ClientSession) => {
+        await collection.insertOne({}, { session: s });
+      };
+
+      // 3.iv Let no_backoff_time be the duration of the withTransaction API call
+      const { duration: noBackoffTime } = await measureDuration(() => {
+        return client.withSession(async s => {
+          await s.withTransaction(callback);
+        });
+      });
+
+      // 4.i Configure the random number generator used for jitter to always return 1.
+      randomStub.returns(1);
+
+      // 4.ii Configure a fail point that forces 13 retries like in step 3.2.
+      await configureFailPoint(this.configuration, failCommand);
+
+      // 4.iii Use the same callback defined in 3.3.
+      // 4.iv Let with_backoff_time be the duration of the withTransaction API call
+      const { duration: fullBackoffDuration } = await measureDuration(() => {
+        return client.withSession(async s => {
+          await s.withTransaction(callback);
+        });
+      });
+
+      // 5. Compare the two time between the two runs.
+      // The sum of 13 backoffs is roughly 2.2 seconds. There is a 1-second window to account for potential variance between the two runs.
+      expect(fullBackoffDuration).to.be.within(
+        noBackoffTime + 2200 - 1000,
+        noBackoffTime + 2200 + 1000
+      );
+    }
+  );
+});