Details
-
Bug
-
Status: Resolved
-
Critical
-
Resolution: Fixed
-
None
-
None
Description
There is
- durable memory segment in PDS enabled data region
- page replacement started
then Ignite performance slow down occurs, JFR & Visual VM shows get&put from FullPageIdTable requires significiant time to be executed (~5-10% оf samples count and up to 40% CPU usage)
Avg steps in loaded pages table, get: 9456 steps, put: 20243 steps, capacity: 20242 size: 7978
Effect of influence of FullPageID table is cumulative, once it starts to grow, number of steps required grows also.
Attachments
Issue Links
- causes
-
IGNITE-7965 Robin-hood hashing may fail with negative index in case backward shift finished with full table scan
- Resolved
- relates to
-
IGNITE-7698 Page read during replacement should be outside of segment write lock
- Resolved
- links to