Suggest adjust bookkeekper.conf default FlushEntryLogBytes to 256MB to improve bookie io throughput #6915
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
|
/pulsarbot run-failure-checks |
|
/pulsarbot run-failure-checks |
1 similar comment
|
/pulsarbot run-failure-checks |
sijie
approved these changes
May 9, 2020
Huanli-Meng
pushed a commit
to Huanli-Meng/pulsar
that referenced
this pull request
May 27, 2020
…o improve bookie io throughput (apache#6915) ### Motivation When AddEntry to bookie, entries will be written to memtable first and then written to the journal log. If it blocked in memtable write, the total AddEntry operation will be blocked, and the 99th AddEntry latency will increase. When memtable full or checkpoint scheduled by timer, memtable will be swapped with backup one. However, if the backup memtable has not been flushed to disk complete yet, the memtable swap will be blocked, and the `AddEntry` operation will be blocked. So in which situation the backup memtable will be blocked? When pulsar doing `catch-up read`, large amount of data will be read from ledger disk, and ledger disk io util will increase. At the same time, backup memtable will flush data from PageCache to disk triggered by OS (default is 30s) or rolling file no matter for HDD or SSD. It is hard to control the flush frequency and batch size in each flush cycle. In most cases, the flush batch size of each flush cycle is from 500MB to 1GB, which will lead to the ledger disk io util to 100% from 1s to 10s+, especially when large amount of read from the ledger disk in the same time. ### Changes In order to decrease ledger disk high io util duration, we can control the backup memtable flush batch size from pageCache to disk. In my practice, the flush batch set to 256MB may perform better for HDD disk. So I suggest adjust the default configuration `FlushEntryLogBytes` to 256MB in bookkeeper.conf to increase bookie io throughput.
huangdx0726
pushed a commit
to huangdx0726/pulsar
that referenced
this pull request
Aug 24, 2020
…o improve bookie io throughput (apache#6915) ### Motivation When AddEntry to bookie, entries will be written to memtable first and then written to the journal log. If it blocked in memtable write, the total AddEntry operation will be blocked, and the 99th AddEntry latency will increase. When memtable full or checkpoint scheduled by timer, memtable will be swapped with backup one. However, if the backup memtable has not been flushed to disk complete yet, the memtable swap will be blocked, and the `AddEntry` operation will be blocked. So in which situation the backup memtable will be blocked? When pulsar doing `catch-up read`, large amount of data will be read from ledger disk, and ledger disk io util will increase. At the same time, backup memtable will flush data from PageCache to disk triggered by OS (default is 30s) or rolling file no matter for HDD or SSD. It is hard to control the flush frequency and batch size in each flush cycle. In most cases, the flush batch size of each flush cycle is from 500MB to 1GB, which will lead to the ledger disk io util to 100% from 1s to 10s+, especially when large amount of read from the ledger disk in the same time. ### Changes In order to decrease ledger disk high io util duration, we can control the backup memtable flush batch size from pageCache to disk. In my practice, the flush batch set to 256MB may perform better for HDD disk. So I suggest adjust the default configuration `FlushEntryLogBytes` to 256MB in bookkeeper.conf to increase bookie io throughput.
15 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.
Motivation
When AddEntry to bookie, entries will be written to memtable first and then written to the journal log. If it blocked in memtable write, the total AddEntry operation will be blocked, and the 99th AddEntry latency will increase.
When memtable full or checkpoint scheduled by timer, memtable will be swapped with backup one. However, if the backup memtable has not been flushed to disk complete yet, the memtable swap will be blocked, and the
AddEntryoperation will be blocked. So in which situation the backup memtable will be blocked?When pulsar doing
catch-up read, large amount of data will be read from ledger disk, and ledger disk io util will increase. At the same time, backup memtable will flush data from PageCache to disk triggered by OS (default is 30s) or rolling file no matter for HDD or SSD. It is hard to control the flush frequency and batch size in each flush cycle. In most cases, the flush batch size of each flush cycle is from 500MB to 1GB, which will lead to the ledger disk io util to 100% from 1s to 10s+, especially when large amount of read from the ledger disk in the same time.Changes
In order to decrease ledger disk high io util duration, we can control the backup memtable flush batch size from pageCache to disk. In my practice, the flush batch set to 256MB may perform better for HDD disk.
So I suggest adjust the default configuration
FlushEntryLogBytesto 256MB in bookkeeper.conf to increase bookie io throughput.