reduce file lookups before purging and fix double referencing files #250
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
Signed-off-by: tabokie <xy.tao@outlook.com>
|
@yiwu-arbug @Connor1996 Since there's no feedback from upstream, shall we go ahead and do the review ourselves? You can comment on the upstream PR, I will synchronize them manually. |
|
yes. sad to see no response in upstream. I will take another look later the week. |
|
Agree |
This was referenced Jul 22, 2021
yiwu-arbug
reviewed
Jul 22, 2021
|
Commented on upstream PR. |
Signed-off-by: tabokie <xy.tao@outlook.com>
yiwu-arbug
approved these changes
Aug 4, 2021
Signed-off-by: tabokie <xy.tao@outlook.com>
Signed-off-by: Xinye Tao <xy.tao@outlook.com>
Connor1996
reviewed
Aug 12, 2021
tabokie
force-pushed
the
perf/purge-scan
branch
from
a584ade
to
2723739
Compare
Signed-off-by: Xinye Tao <xy.tao@outlook.com> Co-authored-by: Connor <zbk602423539@gmail.com>
Signed-off-by: tabokie <xy.tao@outlook.com>
14 tasks
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Outdated suggestions cannot be applied.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.
Signed-off-by: tabokie xy.tao@outlook.com
This PR contains a optimization to reduce DB mutex held time when generating obsolete files purging jobs:
Obsolete file candidates consist of
obsolete_files_that are no longer referenced by any version inVersionSet, or all physical files in DB directory ifdoing_full_scanis true. Consider the former case,obsolete_files_is guaranteed to be obsolete by protocol and thus no need to be checked against live files list. This PR avoids the unnecessary live file lookups when not doing full scan.The optimization assumes a solid reference counting for sst files, i.e. each physical SST file is represented by an unique
FileMetaDatastructure. But this assumption is actually broken in existing codebase: A trivially moved file will appear as a new file meta with the same file number. This patch also fix this issue by forwarding moved old files inVersionBuilder.Aside from that, this PR also assumes SST file won't be moved to a different path with the same file number, and will raise error in that case.
Finally, this PR shows impressive performance improvement in our own testing (thanks @5kbpers). We examine the patch in an artificial workload, where iterators are frequently created and released (active iterator count around 250, RocksDB write OPS around 20K). The logging shows before patching,
AddLiveFiles()call takes 100+ms and in that process copies 2500K file metas with 60K of them being unique.(left: patched, right: vanilla)
Test Plan:
Benchmark Results:
[TPC-C 5K WH 1 KV 512 threads]
Before: TPM: 23676.8, 50th(ms): 570.4, 90th(ms): 704.6, 95th(ms): 738.2, 99th(ms): 872.4, 99.9th(ms): 5637.1, Max(ms): 12884.9
After: TPM: 24395.1(3%+), 50th(ms): 570.4, 90th(ms): 704.6, 95th(ms): 738.2, 99th(ms): 838.9, 99.9th(ms): 1342.2(76%-), Max(ms): 2952.8(77%-)