[fix](cloud) Fix cloud warm up balance slow scheduling#58962
Merged
gavinchou merged 2 commits intoapache:masterfrom Dec 24, 2025
Merged
[fix](cloud) Fix cloud warm up balance slow scheduling#58962gavinchou merged 2 commits intoapache:masterfrom
gavinchou merged 2 commits intoapache:masterfrom
Conversation
deardeng
requested review from
dataroaring,
gavinchou and
w41ter
as code owners
Collaborator
Author
|
run buildall |
deardeng
force-pushed
the
fix-warm-up-balance-slow
branch
from
ad2b9c6 to
06299fa
Compare
BE UT Coverage ReportIncrement line coverage Increment coverage report
|
Collaborator
Author
|
run buildall |
deardeng
force-pushed
the
fix-warm-up-balance-slow
branch
from
bb9e649 to
88ab2db
Compare
Collaborator
Author
|
run buildall |
TPC-H: Total hot run time: 35190 ms |
TPC-DS: Total hot run time: 178254 ms |
ClickBench: Total hot run time: 27.52 s |
BE UT Coverage ReportIncrement line coverage Increment coverage report
|
Contributor
BE Regression && UT Coverage ReportIncrement line coverage Increment coverage report
|
Contributor
FE Regression Coverage ReportIncrement line coverage |
gavinchou
approved these changes
Dec 23, 2025
Contributor
|
PR approved by at least one committer and no changes requested. |
github-actions
Bot
added
approved
Contributor
|
PR approved by anyone and no changes requested. |
16 tasks
deardeng
added a commit
to deardeng/incubator-doris
that referenced
this pull request
Dec 24, 2025
Currently, when performing tablet warm-up balancing in the cloud, the sequential execution of a single warm-up task leads to a series of problems, such as: 1. When scaling up a computer group to include beta nodes, with a large number of tables (millions of tablets), actual tests showed that scaling from 1 beta node to 10 beta nodes took more than 6 hours to reach a balanced state. Each warm-up task RPC took about 30ms. This means that even if a new node can handle the load, scaling up a new node in the cloud can still take up to 6 hours in the worst case. 2. Due to the same logic, decomission be is also relatively slow. Fixes: 1. Batch and pipeline warm-up tasks. Each batch can contain multiple warm-up tasks with the same source and destination (each task represents migrating one tablet). 2. Separate the warm-up task finish thread to prevent scheduling logic from affecting the logic that modifies tablet-to-tablet mappings. 3. Asynchronously fetch file cache meta in the warm_up_cache_async logic and add some bvars. Post-fix testing showed that in a scenario with 10 databases, 10,000 tables, 100,000 partitions, and 1 million tablets, the number of be nodes increased from 3 to 10 within 10 minutes.
deardeng
added a commit
to deardeng/incubator-doris
that referenced
this pull request
Dec 24, 2025
Currently, when performing tablet warm-up balancing in the cloud, the sequential execution of a single warm-up task leads to a series of problems, such as: 1. When scaling up a computer group to include beta nodes, with a large number of tables (millions of tablets), actual tests showed that scaling from 1 beta node to 10 beta nodes took more than 6 hours to reach a balanced state. Each warm-up task RPC took about 30ms. This means that even if a new node can handle the load, scaling up a new node in the cloud can still take up to 6 hours in the worst case. 2. Due to the same logic, decomission be is also relatively slow. Fixes: 1. Batch and pipeline warm-up tasks. Each batch can contain multiple warm-up tasks with the same source and destination (each task represents migrating one tablet). 2. Separate the warm-up task finish thread to prevent scheduling logic from affecting the logic that modifies tablet-to-tablet mappings. 3. Asynchronously fetch file cache meta in the warm_up_cache_async logic and add some bvars. Post-fix testing showed that in a scenario with 10 databases, 10,000 tables, 100,000 partitions, and 1 million tablets, the number of be nodes increased from 3 to 10 within 10 minutes.
yiguolei
pushed a commit
that referenced
this pull request
Dec 25, 2025
Contributor
|
branch-3.1: #59336 |
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
7 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.
What problem does this PR solve?
Currently, when performing tablet warm-up balancing in the cloud, the sequential execution of a single warm-up task leads to a series of problems, such as:
When scaling up a computer group to include beta nodes, with a large number of tables (millions of tablets), actual tests showed that scaling from 1 beta node to 10 beta nodes took more than 6 hours to reach a balanced state. Each warm-up task RPC took about 30ms. This means that even if a new node can handle the load, scaling up a new node in the cloud can still take up to 6 hours in the worst case.
Due to the same logic, decomission be is also relatively slow.
Fixes:
Batch and pipeline warm-up tasks. Each batch can contain multiple warm-up tasks with the same source and destination (each task represents migrating one tablet).
Separate the warm-up task finish thread to prevent scheduling logic from affecting the logic that modifies tablet-to-tablet mappings.
Asynchronously fetch file cache meta in the warm_up_cache_async logic and add some bvars.
Post-fix testing showed that in a scenario with 10 databases, 10,000 tables, 100,000 partitions, and 1 million tablets, the number of be nodes increased from 3 to 10 within 10 minutes.
Issue Number: close #xxx
Related PR: #xxx
Problem Summary:
Release note
None
Check List (For Author)
Test
Behavior changed:
Does this need documentation?
Check List (For Reviewer who merge this PR)