-
Type: Bug
-
Resolution: Done
-
Priority: Minor - P4
-
Affects Version/s: None
-
Component/s: None
-
None
Reference WT-1919.
WT_PM_REC_REWRITE is a short-lived page state set in one specific path, when forced eviction of a leaf page fails because there's nothing much to write. The fact it must be a leaf page during eviction is because it is only set in a reconciliation call where WT_EVICT_UPDATE_RESTORE was set. It's short-lived because before we give up the exclusive lock on the page, we create a new page and swap it into place, and discard the page where WT_PM_REC_REWRITE was set.
For these reasons...
sync_file (the WT_SYNC_CHECKPOINT operation) should not be checking for WT_PM_REC_REWRITE, it should never see that flag as the checkpoint will be blocked from the leaf page until the eviction is complete.
rec_root_write should not be checking for WT_PM_REC_REWRITE, it should never see that flag because it's writing internal pages, not leaf pages (and it's not evicting, anyway).
rec_col_int should not be checking for WT_PM_REC_REWRITE, because any write of a page's column-store internal page parent will be blocked from evaluating the leaf child until the eviction is complete.
rec_write_wrapup should not be checking for WT_PM_REC_REWRITE, because any previous reconciliation that set WT_PM_REC_REWRITE will have cleared that flag before this reconciliation could proceed.
rec_write_wrapup_err should not be checking for WT_PM_REC_REWRITE because it's only set when a reconciliation is already considered successful.
This pretty much removes the need for the WT_PM_REC_REWRITE state – the additional error handling makes me think we'd be better off getting rid of it and using WT_PM_REC_MULTIBLOCK with a single test, if we ended up with 1 block or several blocks.
- is depended on by
-
SERVER-20771 WiredTiger changes for MongoDB 3.2.0-rc0
- Closed
- links to