-
Type:
New Feature
-
Status: Done
-
Priority:
Medium
-
Resolution: Fixed
-
Affects Version/s: 3.6.x
-
Fix Version/s: 4.0.19-12, 4.2.7-7, 3.6.18-6.0
-
Component/s: Core
-
Labels:None
-
Needs Review:Yes
When there are very high number of collections on a node (say >100,000) the iteration of all datahandles in early stages of a WiredTiger checkpoint can take a considerable amount of time (think > 10s).
At first this is not 100% blocking to client db ops but the dhandle (data-handle) lock grows to be 100% over the course of some time and then it is. So if this phase takes say 12 secs, there can be 4~5 seconds towards the end where the db ops are blocked (not sure if logical locking or just effective locking by CPU resource saturation).
- relates to
-
PSMDB-600 Fix leak of dhandle session_inuse counter in __evict_walk through 'error' early loop exit logic
-
- Done
-
-
PSMDB-616 Add "txn_checkpoint_prepare_time" and "txn_checkpoint_tree_helper_time" wiredTiger.transaction stats
-
- Done
-
-
PSMDB-730 Revert PSMDB-587 (and probably PSMDB-616) as WT-6598 arrives in 4.0.21, 4.2.10 and 4.4.2
-
- Done
-