Fix recover ssd cache data by preserving file id map #10409
Closed
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
facebook-github-bot
added
the
CLA Signed
✅ Deploy Preview for meta-velox canceled.
|
In Meta interactive Prestissimo workload, we found there is not much read can hit in SSD cache after cluster upgrade or worker reboot if we rerun the same set of queries. Sometime this is because of the non-deterministic execution order of multiple filter conditions in table scan. If the filter rate is high, then the last few filter conditions might not be executed and the decision is local to a split which means that filter condition execution order varies across splits. This might cause the rerun query to read from remote storage instead of local SSD cache. But this only cause a small portion of reads go to remote storage. The main problem is due to the recover cache entries from checkpoint. Even though we guarantee that the same source file name maps to the same file id but we can't guarantee that the newly assigned file id will be mapped to the same SSD file shard. This will cause the cache space polution. Even if we recovered all the SSD cache entries from checkpoint file, the cache lookup can't find them if the newly assigned file id is mapped to a different SSD file shard. To fix this problem, we add a new constructor in StringIdMap which takes both id and string to recover the previous string id map. The caller guarantees there is no conflict for the passed id/string pairs, and StringIdMap will throw if breaks. Correspondingly, when SSD file recovers from the checkpoint file, it will use the persisted file id and name to rebuild the string id map. Verified on Meta interactive cluster with SSD and a second run after cluster reboot read all the data from local SSD. The end-to-end query latency has been reduced by half compared with the first run.
|
@xiaoxmeng has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
duanmeng
approved these changes
Jul 6, 2024
duanmeng
reviewed
Jul 6, 2024
duanmeng
reviewed
Jul 6, 2024
|
@xiaoxmeng merged this pull request in 770f2ad. |
|
Conbench analyzed the 1 benchmark run on commit There were no benchmark performance regressions. 🎉 The full Conbench report has more details. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
In Meta interactive Prestissimo workload, we found there is not much read can hit in SSD cache
after cluster upgrade or worker reboot if we rerun the same set of queries. Sometime this is
because of the non-deterministic execution order of multiple filter conditions in table scan. If the
filter rate is high, then the last few filter conditions might not be executed and the decision is
local to a split which means that filter condition execution order varies across splits. This might
cause the rerun query to read from remote storage instead of local SSD cache. But this only
cause a small portion of reads go to remote storage.
The main problem is due to the recover cache entries from checkpoint. Even though we guarantee
that the same source file name maps to the same file id but we can't guarantee that the newly
assigned file id will be mapped to the same SSD file shard. This will cause the cache space polution.
Even if we recovered all the SSD cache entries from checkpoint file, the cache lookup can't find them
if the newly assigned file id is mapped to a different SSD file shard.
To fix this problem, we add a new constructor in StringIdMap which takes both id and string to recover
the previous string id map. The caller guarantees there is no conflict for the passed id/string pairs, and
StringIdMap will throw if breaks. Correspondingly, when SSD file recovers from the checkpoint file, it
will use the persisted file id and name to rebuild the string id map.
Verified on Meta interactive cluster with SSD and a second run after cluster reboot read all the data from
local SSD. The end-to-end query latency has been reduced by half compared with the first run.