[Fix](group commit) Fix group commit block queue mem estimate faule #35314
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
|
Thank you for your contribution to Apache Doris. Since 2024-03-18, the Document has been moved to doris-website. |
Collaborator
Author
|
test will be added soon |
Collaborator
Author
|
run buildall |
Contributor
|
clang-tidy review says "All clean, LGTM! 👍" |
TPC-H: Total hot run time: 40651 ms |
TPC-DS: Total hot run time: 167833 ms |
|
TeamCity be ut coverage result: |
Contributor
|
clang-tidy review says "All clean, LGTM! 👍" |
1 similar comment
Contributor
|
clang-tidy review says "All clean, LGTM! 👍" |
Yukang-Lian
force-pushed
the
Fix-Group-Commit-Block-Queue-Mem-Estimate-Fault
branch
from
6e3fe51 to
c8b3869
Compare
Collaborator
Author
|
run buildall |
Contributor
|
clang-tidy review says "All clean, LGTM! 👍" |
TPC-H: Total hot run time: 40068 ms |
TPC-DS: Total hot run time: 169998 ms |
ClickBench: Total hot run time: 30.65 s |
|
TeamCity be ut coverage result: |
gavinchou
reviewed
May 28, 2024
… into a `variant` type column, it causes incorrect memory statistics for group commit backpressure, leading to a stuck issue. **Cause:** In group commit mode, blocks are first added to a queue in batches using `add block`, and then blocks are retrieved from the queue using `get block`. To track memory usage during backpressure, we add the block size to the memory statistics during `add block` and subtract the block size from the memory statistics during `get block`. However, for `variant` types, during the `add block` write to WAL, serialization occurs, which can merge types (e.g., merging `int` and `bigint` into `bigint`), thereby changing the block size. This results in a discrepancy between the block size during `get block` and `add block`, causing memory statistics to overflow. **Solution:** Record the block size at the time of `add block` and use this recorded size during `get block` instead of the actual block size. This ensures consistency in the memory addition and subtraction.
Yukang-Lian
force-pushed
the
Fix-Group-Commit-Block-Queue-Mem-Estimate-Fault
branch
from
c8b3869 to
1a44791
Compare
Collaborator
Author
|
run buildall |
Contributor
|
clang-tidy review says "All clean, LGTM! 👍" |
TPC-H: Total hot run time: 41449 ms |
|
TeamCity be ut coverage result: |
TPC-DS: Total hot run time: 169895 ms |
ClickBench: Total hot run time: 30.35 s |
mymeiyi
approved these changes
May 28, 2024
Contributor
|
PR approved by anyone and no changes requested. |
github-actions
bot
added
the
approved
Contributor
|
PR approved by at least one committer and no changes requested. |
dataroaring
pushed a commit
that referenced
this pull request
May 31, 2024
…35314) ## Proposed changes Issue Number: close #xxx <!--Describe your changes.--> **Problem:** When `group commit=async_mode` and NULL data is imported into a `variant` type column, it causes incorrect memory statistics for group commit backpressure, leading to a stuck issue. **Cause:** In group commit mode, blocks are first added to a queue in batches using `add block`, and then blocks are retrieved from the queue using `get block`. To track memory usage during backpressure, we add the block size to the memory statistics during `add block` and subtract the block size from the memory statistics during `get block`. However, for `variant` types, during the `add block` write to WAL, serialization occurs, which can merge types (e.g., merging `int` and `bigint` into `bigint`), thereby changing the block size. This results in a discrepancy between the block size during `get block` and `add block`, causing memory statistics to overflow. **Solution:** Record the block size at the time of `add block` and use this recorded size during `get block` instead of the actual block size. This ensures consistency in the memory addition and subtraction. ## Further comments If this is a relatively large or complex change, kick off the discussion at [dev@doris.apache.org](mailto:dev@doris.apache.org) by explaining why you chose the solution you did and what alternatives you considered, etc...
Yukang-Lian
added a commit
to Yukang-Lian/doris
that referenced
this pull request
Jul 7, 2024
…pache#35314) ## Proposed changes Issue Number: close #xxx <!--Describe your changes.--> **Problem:** When `group commit=async_mode` and NULL data is imported into a `variant` type column, it causes incorrect memory statistics for group commit backpressure, leading to a stuck issue. **Cause:** In group commit mode, blocks are first added to a queue in batches using `add block`, and then blocks are retrieved from the queue using `get block`. To track memory usage during backpressure, we add the block size to the memory statistics during `add block` and subtract the block size from the memory statistics during `get block`. However, for `variant` types, during the `add block` write to WAL, serialization occurs, which can merge types (e.g., merging `int` and `bigint` into `bigint`), thereby changing the block size. This results in a discrepancy between the block size during `get block` and `add block`, causing memory statistics to overflow. **Solution:** Record the block size at the time of `add block` and use this recorded size during `get block` instead of the actual block size. This ensures consistency in the memory addition and subtraction. ## Further comments If this is a relatively large or complex change, kick off the discussion at [dev@doris.apache.org](mailto:dev@doris.apache.org) by explaining why you chose the solution you did and what alternatives you considered, etc...
dataroaring
pushed a commit
that referenced
this pull request
Jul 7, 2024
…lock queue mem estimate fault" (#37379) Pick [Fix](group commit) Fix group commit block queue mem estimate faule #35314 ## Proposed changes Issue Number: close #xxx <!--Describe your changes.--> **Problem:** When `group commit=async_mode` and NULL data is imported into a `variant` type column, it causes incorrect memory statistics for group commit backpressure, leading to a stuck issue. **Cause:** In group commit mode, blocks are first added to a queue in batches using `add block`, and then blocks are retrieved from the queue using `get block`. To track memory usage during backpressure, we add the block size to the memory statistics during `add block` and subtract the block size from the memory statistics during `get block`. However, for `variant` types, during the `add block` write to WAL, serialization occurs, which can merge types (e.g., merging `int` and `bigint` into `bigint`), thereby changing the block size. This results in a discrepancy between the block size during `get block` and `add block`, causing memory statistics to overflow. **Solution:** Record the block size at the time of `add block` and use this recorded size during `get block` instead of the actual block size. This ensures consistency in the memory addition and subtraction. ## Further comments If this is a relatively large or complex change, kick off the discussion at [dev@doris.apache.org](mailto:dev@doris.apache.org) by explaining why you chose the solution you did and what alternatives you considered, etc... ## Proposed changes Issue Number: close #xxx <!--Describe your changes.-->
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 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.
Proposed changes
Issue Number: close #xxx
Problem: When
group commit=async_modeand NULL data is imported into avarianttype column, it causes incorrect memory statistics for group commit backpressure, leading to a stuck issue.Cause: In group commit mode, blocks are first added to a queue in batches using
add block, and then blocks are retrieved from the queue usingget block. To track memory usage during backpressure, we add the block size to the memory statistics duringadd blockand subtract the block size from the memory statistics duringget block. However, forvarianttypes, during theadd blockwrite to WAL, serialization occurs, which can merge types (e.g., mergingintandbigintintobigint), thereby changing the block size. This results in a discrepancy between the block size duringget blockandadd block, causing memory statistics to overflow.Solution: Record the block size at the time of
add blockand use this recorded size duringget blockinstead of the actual block size. This ensures consistency in the memory addition and subtraction.Further comments
If this is a relatively large or complex change, kick off the discussion at dev@doris.apache.org by explaining why you chose the solution you did and what alternatives you considered, etc...