-
Type:
Task
-
Resolution: Fixed
-
Priority:
Critical - P2
-
Affects Version/s: None
-
Component/s: APIs
-
Security Level: Public (Available to anyone on the web)
-
Storage Engines, Storage Engines - Persistence
-
8,852.152
-
SE Persistence - 2025-08-15
-
8
We should stop writing table and turtle files to disk. This could cause WT to automatically load a checkpoint at startup before mongod explicitly tells it which one to load, leading to unexpected behavior. For example, if I'm primary and wrote CKPT 1-5, but only 4 made it to the log server while 5 didn't, Mongo should restart at 4. But right now, it looks like it might start at 5. If WT needs metadata and turtle files to be written to disk for internal testing, we could make this configurable.
UPDATE: This ticket addresses the problem of not being able to restart disagg WT if there are files from a previous run in the database directory. A follow-on ticket (WT-15163) addresses the more general question of what (if anything) WT should write to the local file system. At first glance, it's not as simple as saying WT shouldn't write the metadata file, since the metadata can include information about local files (e.g., ingest tables) that it doesn't make sense to push to shared storage and if that metadata gets sufficiently large, WT might want to spill it from cache.
- is depended on by
-
WT-14427 [ds-09.04][Storage Engines (Core)] 100% hygiene plan execution
-
- Open
-
- is related to
-
WT-15163 Revisit how WT writes local files in Disagg mode
-
- Open
-
-
-
- Closed
-
- related to
-
WT-14822 Crash in __clayered_copy_constituent_bound due to bound mismatch on layered cursor
-
- Closed
-
-
-
- Closed
-
-
WT-15097 Add a data handle benchmark for WT
-
- Closed
-
-
-
- Closed
-
-
-
- Closed
-
-
-
- Closed
-
-
-
- Open
-
-
-
- Closed
-