-
Type:
Bug
-
Resolution: Fixed
-
Priority:
Major - P3
-
Affects Version/s: None
-
Component/s: None
-
None
-
Storage Execution
-
Fully Compatible
-
ALL
-
Storage Execution 2026-11-09, Storage Execution 2026-11-23
-
200
-
None
-
None
-
None
-
None
-
None
-
None
-
None
Classic oplog application has the following locks to prevent fsync and validate from running mid batch:
// Don't allow the fsync+lock thread to see intermediate states of batch application.
std::lock_guard<std::mutex> fsynclk(oplogApplierLockedFsync);
// Obtain the validation lock to synchronise batch application with validation.
auto lk = collection_validation::obtainExclusiveValidationLock(&opCtx);
Does POA needs the same?
This ticket is to handle VALIDATE only. See SERVER-139205 for fsync.
Solution idea:
Each time a batch is dispatched, POA takes the validation lock in MODE_IX. The lock is stashed with the rest of the in-flight batch's metadata and handed to the batch tracker. When the batch has been completed, the batch tracker unlocks it for one MODE_IX.
Meanwhile, validate should start taking the validation lock in MODE_S.
This means that POA and validate can't run concurrently because MODE_IX and MODE_S conflict. However, several POA batches can be in-flight at once, because multiple callers can lock in MODE_IX at once. Also, multiple validates can run concurrently because multiple MODE_S can run at once.
- related to
-
SERVER-139205 [POA] Implement equivalent of fsync lock
-
- Closed
-