SERVER-56713 Avoid creating retryable findAndModify images while in initial sync

2 files changed, 128 insertions, 2 deletions

diff --git a/jstests/noPassthrough/store_retryable_find_and_modify_images_in_side_collection.js b/jstests/noPassthrough/store_retryable_find_and_modify_images_in_side_collection.js
index 26c49a816c4..b1bb008c285 100644
--- a/jstests/noPassthrough/store_retryable_find_and_modify_images_in_side_collection.js
+++ b/jstests/noPassthrough/store_retryable_find_and_modify_images_in_side_collection.js
@@ -44,8 +44,7 @@
 
     function checkImageCollection(conn, sessionInfo, expectedTs, expectedImage, expectedImageKind) {
         const coll = conn.getDB('config').image_collection;
-        const imageDoc =
-            coll.findOne({'_id.id': sessionInfo.sessionId});
+        const imageDoc = coll.findOne({'_id.id': sessionInfo.sessionId});
 
         assert.eq(sessionInfo.txnNum, imageDoc.txnNum, imageDoc);
         assert.eq(expectedImage, imageDoc.image, imageDoc);
diff --git a/jstests/replsets/retryable_writes_initial_sync.js b/jstests/replsets/retryable_writes_initial_sync.js
new file mode 100644
index 00000000000..b6a663dc319
--- /dev/null
+++ b/jstests/replsets/retryable_writes_initial_sync.js
@@ -0,0 +1,127 @@
+/**
+ * Tests that retryable findAndModify data stored in the `config.image_collection` side collection
+ * do not get populated by nodes doing oplog application while in initial sync.
+ */
+(function() {
+    'use strict';
+
+    load('jstests/replsets/rslib.js');
+
+    // A secondary cannot compute retryable images during initial sync. Thus we skip db hash checks
+    // as
+    // its expected for config.image_collection to not match.
+    TestData.skipCheckDBHashes = true;
+
+    // Start a single node replica set.
+    const rst = new ReplSetTest(
+        {nodes: 1, nodeOptions: {setParameter: {storeFindAndModifyImagesInSideCollection: true}}});
+    rst.startSet();
+    rst.initiateWithHighElectionTimeout();
+
+    const dbName = jsTest.name();
+    const primary = rst.getPrimary();
+    const primaryDB = primary.getDB(dbName);
+    const primaryColl = primaryDB['collection'];
+
+    primaryColl.insert({_id: 1});
+
+    let lsid = UUID();
+    let result = assert.commandWorked(primaryDB.runCommand({
+        findandmodify: primaryColl.getName(),
+        lsid: {id: lsid},
+        txnNumber: NumberLong(1),
+        stmtId: NumberInt(1),
+        query: {_id: 1},
+        new: false,
+        update: {$set: {preInitialSync: true}}
+    }));
+    jsTestLog("Pre-initial sync retryable findAndModify: " + tojson(result));
+
+    jsTestLog("Adding a new voting node (node1) to the replica set.");
+    const node1 = rst.add({
+        rsConfig: {priority: 1, votes: 1},
+        setParameter: {'failpoint.initialSyncHangAfterDataCloning': tojson({mode: 'alwaysOn'})}
+    });
+    rst.reInitiate();
+
+    jsTestLog("Wait for node1 to hang during initial sync.");
+    checkLog.contains(node1, 'initial sync - initialSyncHangAfterDataCloning fail point enabled.');
+
+    // Perform a findAndModify update that will not be retryable on the node that's concurrently
+    // initial syncing.
+    result = assert.commandWorked(primaryDB.runCommand({
+        findandmodify: primaryColl.getName(),
+        lsid: {id: lsid},
+        txnNumber: NumberLong(2),
+        stmtId: NumberInt(1),
+        query: {_id: 1},
+        new: false,
+        update: {$set: {duringInitialSync: true}}
+    }));
+    jsTestLog("Retryable findAndModify during initial sync" + tojson(result));
+
+    // With a separate logical session, perform a findAndModify removal that will not be retryable
+    // on the node that's concurrently initial syncing.
+    let otherLsid = UUID();
+    result = assert.commandWorked(primaryDB.runCommand({
+        findandmodify: primaryColl.getName(),
+        lsid: {id: otherLsid},
+        txnNumber: NumberLong(3),
+        stmtId: NumberInt(1),
+        query: {_id: 1},
+        new: false,
+        remove: true
+    }));
+    jsTestLog("Retryable findAndModify removal during initial sync: " + tojson(result));
+
+    jsTestLog("Resuming initial sync.");
+    assert.commandWorked(
+        node1.adminCommand({configureFailPoint: "initialSyncHangAfterDataCloning", mode: 'off'}));
+    rst.waitForState(node1, ReplSetTest.State.SECONDARY);
+
+    let initialSyncedNode = rst.getSecondary();
+    rst.stepUp(initialSyncedNode);
+
+    result = assert.commandFailedWithCode(initialSyncedNode.getDB(dbName).runCommand({
+        findandmodify: primaryColl.getName(),
+        lsid: {id: lsid},
+        txnNumber: NumberLong(2),
+        stmtId: NumberInt(1),
+        query: {_id: 1},
+        new: false,
+        update: {$set: {duringInitialSync: true}}
+    }),
+                                          ErrorCodes.IncompleteTransactionHistory);
+    // Assert that retrying the update fails.
+    jsTestLog(
+        "Secondary: " + initialSyncedNode + " Data: " +
+        tojson(initialSyncedNode.getDB(dbName)['collection'].findOne()) + " Image: " +
+        tojson(initialSyncedNode.getDB("config")['image_collection'].findOne({"_id.id": lsid})) +
+        " retried findAndModify against synced node: " + result);
+
+    result = assert.commandFailedWithCode(initialSyncedNode.getDB(dbName).runCommand({
+        findandmodify: primaryColl.getName(),
+        lsid: {id: otherLsid},
+        txnNumber: NumberLong(3),
+        stmtId: NumberInt(1),
+        query: {_id: 1},
+        new: false,
+        remove: true
+    }),
+                                          ErrorCodes.IncompleteTransactionHistory);
+    // Assert that retrying the delete fails.
+    jsTestLog("Secondary: " + initialSyncedNode + " Data: " +
+                  tojson(initialSyncedNode.getDB(dbName)['collection'].findOne()),
+              " Image: " + tojson(initialSyncedNode.getDB("config")['image_collection'].findOne(
+                               {"_id.id": otherLsid})) +
+                  " retried delete against synced node: " + result);
+
+    // There should be two sessions/image entries on the initial syncing node, and both should be
+    // flagged as invalidated.
+    assert.eq(2, initialSyncedNode.getDB('config')['image_collection'].count({invalidated: true}));
+    assert.eq(2,
+              initialSyncedNode.getDB('config')['image_collection'].count(
+                  {invalidatedReason: "initial sync"}));
+
+    rst.stopSet();
+}());