-
Type:
Task
-
Resolution: Fixed
-
Priority:
Major - P3
-
Affects Version/s: None
-
Component/s: Checkpoints
-
Security Level: Public (Available to anyone on the web)
-
Storage Engines - Persistence
-
2,636.197
-
SE Persistence backlog
-
None
Checkpoint currently reconciles all pages sequentially, which leads to potentially long and inefficient checkpoints, especially for single-table workloads (e.g., YCSB), and with continuous checkpointing (which we would like to use in SLS).
One option for addressing this would be for the checkpoint code to add leaf pages from the btree traversal to a queue, and then have several threads to pick them up from the queue and then reconcile them in parallel. The checkpoint thread would then wait for these other threads to finish and reconcile the internal pages serially.
- is depended on by
-
WT-16768 Implement parallel checkpoint
-
- Open
-
- is duplicated by
-
-
- Closed
-
- is related to
-
-
- Closed
-
- related to
-
-
- Backlog
-
-
-
- Closed
-
-
-
- Investigating
-