-
Type:
Technical Debt
-
Resolution: Fixed
-
Priority:
Major - P3
-
Affects Version/s: None
-
Component/s: Block Cache
-
Storage Engines, Storage Engines - Foundations, Storage Engines - Persistence
-
SE Persistence backlog
-
3
wt_blkcache_write contains a heuristic for whether it's worth compressing a block. Originally, this was whether the block was smaller than the minimum block allocation size. This makes sense, because we'd have to allocate the minimum block allocation size anyway.
But the presence of the disagg block manager changes that – this block manager doesn't allocate space in files so has no concept of a minimum allocation size. I changed the heuristic to be "double the header size" but that'll only work for disagg, and doesn't make sense for the existing block manager.
We should come up with a better heuristic, or change the heuristic based on the block manager, or something else (block manager to call to check if a page is worth compressing?)
- is related to
-
WT-14968 Remove the checkpoint cleanup wait after each file process
-
- Closed
-
-
WT-15037 maybe-uninitialized warning for uri treated as error in cur_hs.c
-
- Closed
-
-
WT-14932 Build Framework for Unique WT Verbose IDs & Update Critical Atlas Log Rules
-
- Closed
-
-
WT-15014 Forbid setting preserve prepare config if precise checkpoint is not enabled
-
- Closed
-
-
WT-14628 Check exception lists in dist/ for disagg-related items
-
- Closed
-
-
WT-15038 Track multiblock write stats at connection level
-
- Closed
-
-
WT-14957 Evg-task-timeout -- deadlock scenario b/w eviction and reconfigure
-
- Closed
-
- related to
-
WT-15032 Better compression heuristic in block cache
-
- Open
-