SERVER-52833 Perform capped deletes during startup recovery on documents inserted earlier in startup recovery

4 files changed, 94 insertions, 1 deletions

diff --git a/jstests/noPassthrough/capped_deletes_within_rollback.js b/jstests/noPassthrough/capped_deletes_within_rollback.js
new file mode 100644
index 00000000000..8ba6a13e1f7
--- /dev/null
+++ b/jstests/noPassthrough/capped_deletes_within_rollback.js
@@ -0,0 +1,38 @@
+/**
+ * Tests that capped deletes occur during rollback on documents inserted earlier in rollback.
+ *
+ * @tags: [
+ *   requires_replication,
+ * ]
+ */
+(function() {
+'use strict';
+
+load('jstests/replsets/libs/rollback_test.js');
+
+const rollbackTest = new RollbackTest(jsTestName());
+
+const testDB = function() {
+    return rollbackTest.getPrimary().getDB('test');
+};
+
+const coll = function() {
+    return testDB().getCollection(jsTestName());
+};
+
+assert.commandWorked(
+    testDB().createCollection(coll().getName(), {capped: true, size: 100, max: 1}));
+assert.commandWorked(coll().insert({a: 1}));
+
+rollbackTest.transitionToRollbackOperations();
+rollbackTest.transitionToSyncSourceOperationsBeforeRollback();
+
+assert.commandWorked(coll().insert([{b: 1}, {b: 2}]));
+
+rollbackTest.transitionToSyncSourceOperationsDuringRollback();
+rollbackTest.transitionToSteadyStateOperations();
+
+assert.eq(coll().find().itcount(), 1);
+
+rollbackTest.stop();
+})();
\ No newline at end of file
diff --git a/jstests/noPassthrough/capped_deletes_within_startup_recovery.js b/jstests/noPassthrough/capped_deletes_within_startup_recovery.js
new file mode 100644
index 00000000000..527d9bb0d76
--- /dev/null
+++ b/jstests/noPassthrough/capped_deletes_within_startup_recovery.js
@@ -0,0 +1,45 @@
+/**
+ * Tests that capped deletes occur during statup recovery on documents inserted earlier in startup
+ * recovery.
+ *
+ * @tags: [
+ *   requires_persistence,
+ *   requires_replication,
+ * ]
+ */
+(function() {
+'use strict';
+
+load('jstests/libs/fail_point_util.js');
+
+const replTest = new ReplSetTest({nodes: 1});
+replTest.startSet();
+replTest.initiate();
+
+const primary = function() {
+    return replTest.getPrimary();
+};
+
+const testDB = function() {
+    return primary().getDB('test');
+};
+
+const coll = function() {
+    return testDB().getCollection(jsTestName());
+};
+
+assert.commandWorked(
+    testDB().createCollection(coll().getName(), {capped: true, size: 100, max: 1}));
+
+const ts =
+    assert.commandWorked(testDB().runCommand({insert: coll().getName(), documents: [{a: 1}]}))
+        .operationTime;
+configureFailPoint(primary(), 'holdStableTimestampAtSpecificTimestamp', {timestamp: ts});
+
+assert.commandWorked(coll().insert([{b: 1}, {b: 2}]));
+replTest.restart(primary());
+
+assert.eq(coll().find().itcount(), 1);
+
+replTest.stopSet();
+})();
\ No newline at end of file
diff --git a/src/mongo/db/server_recovery.h b/src/mongo/db/server_recovery.h
index 3b9d87a8065..e78653d768a 100644
--- a/src/mongo/db/server_recovery.h
+++ b/src/mongo/db/server_recovery.h
@@ -34,6 +34,7 @@
 
 #include "mongo/db/service_context.h"
 #include "mongo/platform/mutex.h"
+#include "mongo/util/string_map.h"
 
 namespace mongo {
 /**
@@ -82,7 +83,7 @@ public:
 
 private:
     mutable Mutex _mutex = MONGO_MAKE_LATCH("SizeRecoveryState::_mutex");
-    std::set<std::string> _collectionsAlwaysNeedingSizeAdjustment;
+    StringSet _collectionsAlwaysNeedingSizeAdjustment;
 };
 
 /**
diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_record_store.cpp b/src/mongo/db/storage/wiredtiger/wiredtiger_record_store.cpp
index f5873ff5249..eed04767a6e 100644
--- a/src/mongo/db/storage/wiredtiger/wiredtiger_record_store.cpp
+++ b/src/mongo/db/storage/wiredtiger/wiredtiger_record_store.cpp
@@ -1731,6 +1731,15 @@ void WiredTigerRecordStore::_initNextIdIfNeeded(OperationContext* opCtx) {
         return;
     }
 
+    // During startup recovery, the collectionAlwaysNeedsSizeAdjustment flag is not set by default
+    // for the sake of efficiency. However, if we reach this point, we may need to set it in order
+    // to ensure that capped deletes can occur on documents inserted earlier in startup recovery.
+    if (inReplicationRecovery(getGlobalServiceContext()) &&
+        !sizeRecoveryState(getGlobalServiceContext())
+             .collectionAlwaysNeedsSizeAdjustment(getIdent())) {
+        checkSize(opCtx);
+    }
+
     // Need to start at 1 so we are always higher than RecordId::min()
     int64_t nextId = 1;