raftstore-v2: support column family based write buffer manager #15453
Conversation
Signed-off-by: SpadeA-Tang <[email protected]>
|
[REVIEW NOTIFICATION] This pull request has been approved by:
To complete the pull request process, please ask the reviewers in the list to review by filling The full list of commands accepted by this bot can be found here. Reviewer can indicate their review by submitting an approval review. |
ti-chi-bot
bot
added
do-not-merge/needs-linked-issue
release-note
ti-chi-bot
bot
added
needs-cherry-pick-release-7.1
Signed-off-by: SpadeA-Tang <[email protected]>
ti-chi-bot
bot
added
size/M
Signed-off-by: SpadeA-Tang <[email protected]>
src/config/mod.rs
Outdated
| @@ -1330,6 +1336,7 @@ impl Default for DbConfig { | |||
| enable_unordered_write: false, | |||
| allow_concurrent_memtable_write: None, | |||
| write_buffer_limit: None, | |||
| lock_write_buffer_limit: None, | |||
There was a problem hiding this comment.
Putting it in CfConfig makes more sense.
Signed-off-by: SpadeA-Tang <[email protected]>
ti-chi-bot
bot
added
size/L
src/config/mod.rs
Outdated
| || self.defaultcf.write_buffer_limit.is_some() | ||
| || self.raftcf.write_buffer_limit.is_some() | ||
| { | ||
| return Err( |
There was a problem hiding this comment.
We can support them all, as long as write_buffer_stall_ratio is 0.
There was a problem hiding this comment.
It's better to not support now? Expose this when we have use cases and relevant test.
There was a problem hiding this comment.
The code is very simple (one modification in build_cf_resources) compared to the benefit. When you forbid a configuration it would be hard to work around it when needed.
Signed-off-by: SpadeA-Tang <[email protected]>
components/engine_rocks/Cargo.toml
Outdated
| @@ -57,7 +57,8 @@ tracker = { workspace = true } | |||
| txn_types = { workspace = true } | |||
|
|
|||
| [dependencies.rocksdb] | |||
| git = "https://github.com/tikv/rust-rocksdb.git" | |||
| git = "https://github.com/SpadeA-Tang/rust-rocksdb.git" | |||
There was a problem hiding this comment.
make sure this is updated.
ti-chi-bot
bot
added
status/LGT2
Signed-off-by: SpadeA-Tang <[email protected]>
|
/run-test retry=5 |
|
/merge |
|
@SpadeA-Tang: It seems you want to merge this PR, I will help you trigger all the tests: /run-all-tests You only need to trigger
Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the ti-community-infra/tichi repository. |
|
This pull request has been accepted and is ready to merge. Commit hash: f668a00
|
ti-chi-bot
bot
added
the
status/can-merge
|
In response to a cherrypick label: new pull request created to branch |
ref tikv#12842 Signed-off-by: ti-chi-bot <[email protected]>
What is changed and how it works?
Issue Number: Ref #12842
What's Changed:
This PR (include related PRs) enables cf based write buffer manager, so that we can use a
separate write buffer manager to mange lock cf.
Compared with default and write cfs, lock cf normally uses and should use significantly less memory, and making all cfs
use a single write buffer manager actually means lock cf has no limit.
Sometimes especially when there's heavy coprocessor requests with insert request, we wish lock cf memtable flushes and compactions timely to make tombstone be removed quickly which can be done by using a relatively small lock write buffer limit after this PR. YCSB workload e demonstrates good for this senario:
ycsb workload e comparison:
qps increases around 15.2%. Both p99 latency and tombstone met reduced a lot.
metrics listed below:
master:
command per second
coprocessor p99 duration
tombstone
this PR
command per second
coprocessor p99 duration
tombstone
Related changes
tikv/rust-rocksdb#763
tikv/rocksdb#343
Check List
Tests
Release note