Make RocksDB delete entries batch size configurable #3646
Merged
Conversation
This file contains hidden or 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
wenbingshen
requested changes
Nov 15, 2022
| if (deletedEntriesInBatch > deleteEntriesBatchSize) { | ||
| batch.flush(); | ||
| batch.clear(); | ||
| deletedEntriesInBatch = 0; |
Member
There was a problem hiding this comment.
Nov 11, 2022 @ 05:49:39.336 05:49:39.336 [db-storage-cleanup-10-1] INFO org.apache.bookkeeper.bookie.storage.ldb.EntryLocationIndex - Deleted indexes for 0 entries from 16074 ledgers in 625.242 seconds
deletedEntries=0, does it mean that the condition of deletedEntriesInBatch > DELETE_ENTRIES_BATCH_SIZE has not entered yet, will adjusting the size of DELETE_ENTRIES_BATCH_SIZE affect this delay? It doesn't look like the read latency to a few seconds is caused by this situation.
Contributor
Author
There was a problem hiding this comment.
deletedEntries=0 is a bug in the log, we have fixed it in the master branch.
zymap
approved these changes
Nov 16, 2022
Member
|
@hangc0276 I'm not very clear about the current bookkeeper configuration management, including whether we need to add instructions for this new config in the docs? Personally, I think this configuration is helpful for performance. Should it be explained to users in the docs? |
hangc0276
added a commit
that referenced
this pull request
Nov 18, 2022
…ance (#3653) ### Motivation The entry log location index deletion is deleted in batches one by one currently, and it will have low performance. Refer to: #3646 Matteo has introduced deleteRange API a few years ago, but rollback due to RocksDB delete ranges bug. #1620. The RocksDB bug has been addressed since 5.18.0 https://github.com/facebook/rocksdb/blob/main/HISTORY.md#5180-2018-11-30. We can bring the `deleteRange` API back to improve the entry log location deletion performance. ### Changes Bring `deleteRange` API back for entry log location deletion.
hangc0276
added a commit
that referenced
this pull request
Nov 21, 2022
…ance (#3653) The entry log location index deletion is deleted in batches one by one currently, and it will have low performance. Refer to: #3646 Matteo has introduced deleteRange API a few years ago, but rollback due to RocksDB delete ranges bug. #1620. The RocksDB bug has been addressed since 5.18.0 https://github.com/facebook/rocksdb/blob/main/HISTORY.md#5180-2018-11-30. We can bring the `deleteRange` API back to improve the entry log location deletion performance. Bring `deleteRange` API back for entry log location deletion. (cherry picked from commit 696919c)
zhaohaidao
pushed a commit
to zhaohaidao/bookkeeper
that referenced
this pull request
Nov 21, 2022
zhaohaidao
pushed a commit
to zhaohaidao/bookkeeper
that referenced
this pull request
Nov 21, 2022
…ance (apache#3653) ### Motivation The entry log location index deletion is deleted in batches one by one currently, and it will have low performance. Refer to: apache#3646 Matteo has introduced deleteRange API a few years ago, but rollback due to RocksDB delete ranges bug. apache#1620. The RocksDB bug has been addressed since 5.18.0 https://github.com/facebook/rocksdb/blob/main/HISTORY.md#5180-2018-11-30. We can bring the `deleteRange` API back to improve the entry log location deletion performance. ### Changes Bring `deleteRange` API back for entry log location deletion.
nicoloboschi
pushed a commit
to datastax/bookkeeper
that referenced
this pull request
Jan 11, 2023
…ance (apache#3653) The entry log location index deletion is deleted in batches one by one currently, and it will have low performance. Refer to: apache#3646 Matteo has introduced deleteRange API a few years ago, but rollback due to RocksDB delete ranges bug. apache#1620. The RocksDB bug has been addressed since 5.18.0 https://github.com/facebook/rocksdb/blob/main/HISTORY.md#5180-2018-11-30. We can bring the `deleteRange` API back to improve the entry log location deletion performance. Bring `deleteRange` API back for entry log location deletion. (cherry picked from commit 696919c) (cherry picked from commit e159510)
yaalsn
pushed a commit
to yaalsn/bookkeeper
that referenced
this pull request
Jan 30, 2023
yaalsn
pushed a commit
to yaalsn/bookkeeper
that referenced
this pull request
Jan 30, 2023
…ance (apache#3653) ### Motivation The entry log location index deletion is deleted in batches one by one currently, and it will have low performance. Refer to: apache#3646 Matteo has introduced deleteRange API a few years ago, but rollback due to RocksDB delete ranges bug. apache#1620. The RocksDB bug has been addressed since 5.18.0 https://github.com/facebook/rocksdb/blob/main/HISTORY.md#5180-2018-11-30. We can bring the `deleteRange` API back to improve the entry log location deletion performance. ### Changes Bring `deleteRange` API back for entry log location deletion.
hangc0276
added a commit
that referenced
this pull request
Feb 8, 2023
### Motivations This PR is to resolve the issue #3734 in branch-4.14 by following this suggestion. #3734 (comment) ### Modifications 1. Revert #3653 2. Bring #3646 to branch-4.14 to make delete entries batch size configurable.
nicoloboschi
pushed a commit
to datastax/bookkeeper
that referenced
this pull request
Mar 13, 2023
…e#3768) ### Motivations This PR is to resolve the issue apache#3734 in branch-4.14 by following this suggestion. apache#3734 (comment) ### Modifications 1. Revert apache#3653 2. Bring apache#3646 to branch-4.14 to make delete entries batch size configurable. (cherry picked from commit e56d6d6)
nicoloboschi
pushed a commit
to datastax/bookkeeper
that referenced
this pull request
Mar 13, 2023
…e#3768) ### Motivations This PR is to resolve the issue apache#3734 in branch-4.14 by following this suggestion. apache#3734 (comment) ### Modifications 1. Revert apache#3653 2. Bring apache#3646 to branch-4.14 to make delete entries batch size configurable. (cherry picked from commit e56d6d6)
Ghatage
pushed a commit
to sijie/bookkeeper
that referenced
this pull request
Jul 12, 2024
Ghatage
pushed a commit
to sijie/bookkeeper
that referenced
this pull request
Jul 12, 2024
…ance (apache#3653) ### Motivation The entry log location index deletion is deleted in batches one by one currently, and it will have low performance. Refer to: apache#3646 Matteo has introduced deleteRange API a few years ago, but rollback due to RocksDB delete ranges bug. apache#1620. The RocksDB bug has been addressed since 5.18.0 https://github.com/facebook/rocksdb/blob/main/HISTORY.md#5180-2018-11-30. We can bring the `deleteRange` API back to improve the entry log location deletion performance. ### Changes Bring `deleteRange` API back for entry log location deletion.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.
Motivation
When there are a lot of ledgers to be deleted, the EntyLocationIndex in RocksDB will delete a large number of entries at one time.
The RocksDB location index deletes operation will cause RocksDB busy, and impact the read latency to a few seconds.
Currently, we delete RocksDB entries in batch, and the batch size is hard-coded. We can not change it.
Changes
Make the RocksDB delete entries in batch configurable.