Adaptive mark_segment_size for parallel replicas by nickitat · Pull Request #68424 · ClickHouse/ClickHouse

nickitat · 2024-08-15T17:36:50Z

Changelog category (leave one):

Performance Improvement

Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):

New algorithm to determine the unit of marks distribution between replicas by consistent hash. Different numbers of marks chosen for different read patterns to improve performance.

The essence of the changes:

new algorithm to choose unit of marks distribution by consistent hash (description)
changes in PR-s protocol, separate versioning is now actually versioning

CI Settings (Only check the boxes if you know what you are doing):

Allow: All Required Checks
Allow: Stateless tests
Allow: Stateful tests
Allow: Integration Tests
Allow: Performance tests
Allow: All Builds
Allow: batch 1, 2 for multi-batch jobs
Allow: batch 3, 4, 5, 6 for multi-batch jobs

Exclude: Style check
Exclude: Fast test
Exclude: All with ASAN
Exclude: All with TSAN, MSAN, UBSAN, Coverage
Exclude: All with aarch64, release, debug

Run only fuzzers related jobs (libFuzzer fuzzers, AST fuzzers, etc.)
Exclude: AST fuzzers

robot-ch-test-poll4 · 2024-08-15T17:39:50Z

This is an automated comment for commit 6357750 with description of existing statuses. It's updated for the latest CI running

❌ Click here to open a full report in a separate page

Check name	Description	Status
AST fuzzer	Runs randomly generated queries to catch program errors. The build type is optionally given in parenthesis. If it fails, ask a maintainer for help	❌ failure
Stateless tests	Runs stateless functional tests for ClickHouse binaries built in various configurations -- release, debug, with sanitizers, etc	❌ failure

Successful checks

Check name	Description	Status
Builds	There's no description for the check yet, please add it to tests/ci/ci_config.py:CHECK_DESCRIPTIONS	✅ success
ClickBench	Runs [ClickBench](https://github.com/ClickHouse/ClickBench/) with instant-attach table	✅ success
Compatibility check	Checks that clickhouse binary runs on distributions with old libc versions. If it fails, ask a maintainer for help	✅ success
Docker keeper image	The check to build and optionally push the mentioned image to docker hub	✅ success
Docker server image	The check to build and optionally push the mentioned image to docker hub	✅ success
Docs check	Builds and tests the documentation	✅ success
Fast test	Normally this is the first check that is ran for a PR. It builds ClickHouse and runs most of stateless functional tests, omitting some. If it fails, further checks are not started until it is fixed. Look at the report to see which tests fail, then reproduce the failure locally as described here	✅ success
Flaky tests	Checks if new added or modified tests are flaky by running them repeatedly, in parallel, with more randomization. Functional tests are run 100 times with address sanitizer, and additional randomization of thread scheduling. Integration tests are run up to 10 times. If at least once a new test has failed, or was too long, this check will be red. We don't allow flaky tests, read the doc	✅ success
Install packages	Checks that the built packages are installable in a clear environment	✅ success
Integration tests	The integration tests report. In parenthesis the package type is given, and in square brackets are the optional part/total tests	✅ success
Performance Comparison	Measure changes in query performance. The performance test report is described in detail here. In square brackets are the optional part/total tests	✅ success
Stateful tests	Runs stateful functional tests for ClickHouse binaries built in various configurations -- release, debug, with sanitizers, etc	✅ success
Stress test	Runs stateless functional tests concurrently from several clients to detect concurrency-related errors	✅ success
Style check	Runs a set of checks to keep the code style clean. If some of tests failed, see the related log from the report	✅ success
Unit tests	Runs the unit tests for different release types	✅ success
Upgrade check	Runs stress tests on server version from last release and then tries to upgrade it to the version from the PR. It checks if the new server can successfully startup without any errors, crashes or sanitizer asserts	✅ success

devcrafter

LGTM with comments

src/Storages/MergeTree/MergeTreeReadPoolParallelReplicas.cpp

devcrafter · 2024-09-10T11:39:15Z

src/Storages/MergeTree/MergeTreeReadPoolParallelReplicas.cpp

+    /// Here we take max of two numbers:
+    /// * (min_marks_per_task * threads) = the number of marks we request from the coordinator each time - there is no point to have segments smaller than one unit of work for a replica
+    /// * (sum_marks / number_of_replicas^2) - we use consistent hashing for work distribution (including work stealing). If we have a really slow replica
+    ///   everything up to (1/number_of_replicas) portion of its work will be stolen by other replicas. And it owns (1/number_of_replicas) share of total number of marks.


Shouldn't it be like this?

Suggested change

/// everything up to (1/number_of_replicas) portion of its work will be stolen by other replicas. And it owns (1/number_of_replicas) share of total number of marks.

/// everything up to ((number_of_replicas-1)/number_of_replicas) portion of its work will be stolen by other replicas. And it owns (1/number_of_replicas) share of total number of marks.

yeah, i should prob say 'except' instead of 'up to'

nickitat · 2024-09-10T15:24:05Z

Integration tests (aarch64) [5/6] - #69449

nickitat · 2024-09-13T16:04:44Z

waiting for #69602 to merge with master

nickitat · 2024-09-16T10:11:36Z

AST fuzzer (debug) - #68296
Stateless tests (tsan) [1/4] - not sure if it the reason for t/o but there is sanitizer report in logs #49156

devcrafter · 2024-09-16T20:04:22Z

src/Core/ProtocolDefines.h


-static constexpr auto DBMS_PARALLEL_REPLICAS_PROTOCOL_VERSION = 3;
+static constexpr auto DBMS_MIN_SUPPORTED_PARALLEL_REPLICAS_PROTOCOL_VERSION = 3;
+static constexpr auto DBMS_PARALLEL_REPLICAS_PROTOCOL_VERSION = 4;


It looks like the previous approach - using only CH client protocol version, was the right one. Every time, we need to change the PR protocol, it's necessary to change the client protocol version anyway.
It's because of the case when first request is sent by a version with a bigger protocol number than the recipient.

Let's consider 2 replicas communication with PR protocol v2 and v1 (v2 > v1, v1 >= DBMS_MIN_SUPPORTED_PARALLEL_REPLICAS_PROTOCOL_VERSION).
In this case, v2 replica needs to communicate with v1 replicas by v1 protocol. But v2 replicas can understand it only from client protocol version of v1 replica. This means that every time we change PR protocol, we have to increase CH client protocol version to cover this case. Which, in turn, makes PR protocol version unnecessary

By using purely PR protocol, we'd have to introduce a handshake phase, during which peers agree which protocol version to use - which is an additional RTT, i.e. waste of time/network.

Sorry, miss that the parallel replicas protocol version within this PR is part of client protocol now. So, we're getting it during client protocol handshake, - so, the PR protocol version of remote peer is known at first PR request to the peer.

Fix parallel replicas protocol after #68424

…_replicas Adaptive mark_segment_size for parallel replicas

Fix parallel replicas protocol after ClickHouse#68424

…_replicas Adaptive mark_segment_size for parallel replicas

Fix parallel replicas protocol after ClickHouse#68424

impl

cb03354

robot-ch-test-poll1 added the pr-performance Pull request with some performance improvements label Aug 15, 2024

nickitat added 2 commits August 15, 2024 18:44

fix typo

891f9c5

add setting change

80d985a

devcrafter self-assigned this Aug 15, 2024

nickitat added 5 commits August 16, 2024 23:23

add bw-compatibility test

e7fc89b

fix

628a430

Merge branch 'master' into adaptive_parallel_replicas

8a0f41d

better

30229a3

fix build

c252b3c

nickitat marked this pull request as ready for review August 18, 2024 22:41

Merge branch 'master' into adaptive_parallel_replicas

d4a3a03

nikitamikhaylov self-assigned this Aug 21, 2024

devcrafter and others added 4 commits September 6, 2024 23:08

Merge branch 'master' into adaptive_parallel_replicas

f5d49f8

fix settings changes

8cdc10c

Merge branch 'master' into adaptive_parallel_replicas

1df897d

fix

61ebcdc

devcrafter approved these changes Sep 10, 2024

View reviewed changes

nickitat added 5 commits September 10, 2024 17:18

fix wording

8d5d7dd

use final task size in segment size calculation

fc83c1c

log mark_segment_size on initiator

1e3bc6d

revive separate protocol versioning for PRs

16f93ea

Merge branch 'master' into adaptive_parallel_replicas

7b2810b

nickitat added 2 commits September 14, 2024 19:31

Merge branch 'master' into adaptive_parallel_replicas

9eb7877

fix build

6357750

devcrafter self-requested a review September 16, 2024 19:27

devcrafter reviewed Sep 16, 2024

View reviewed changes

devcrafter approved these changes Sep 17, 2024

View reviewed changes

nickitat enabled auto-merge September 17, 2024 12:22

nickitat added this pull request to the merge queue Sep 17, 2024

Merged via the queue into master with commit ffaf97a Sep 17, 2024

nickitat deleted the adaptive_parallel_replicas branch September 17, 2024 14:35

robot-ch-test-poll2 added the pr-synced-to-cloud The PR is synced to the cloud repo label Sep 17, 2024

github-merge-queue bot pushed a commit that referenced this pull request Sep 19, 2024

Merge pull request #69744 from ClickHouse/fix_pr_protocol

61524aa

Fix parallel replicas protocol after #68424

zvonand pushed a commit to Altinity/ClickHouse that referenced this pull request Dec 2, 2024

Merge pull request ClickHouse#68424 from ClickHouse/adaptive_parallel…

99b3005

…_replicas Adaptive mark_segment_size for parallel replicas

zvonand pushed a commit to Altinity/ClickHouse that referenced this pull request Dec 3, 2024

Merge pull request ClickHouse#68424 from ClickHouse/adaptive_parallel…

6426caa

…_replicas Adaptive mark_segment_size for parallel replicas

zvonand mentioned this pull request Dec 3, 2024

24.8 Backport of #70332 - Passing external user roles from query originator to other nodes Altinity/ClickHouse#542

Merged

zvonand pushed a commit to Altinity/ClickHouse that referenced this pull request Dec 13, 2024

Merge pull request ClickHouse#68424 from ClickHouse/adaptive_parallel…

bcb0370

…_replicas Adaptive mark_segment_size for parallel replicas

zvonand pushed a commit to Altinity/ClickHouse that referenced this pull request Dec 17, 2024

Merge pull request ClickHouse#68424 from ClickHouse/adaptive_parallel…

06a4050

…_replicas Adaptive mark_segment_size for parallel replicas

zvonand pushed a commit to Altinity/ClickHouse that referenced this pull request Dec 17, 2024

Merge pull request ClickHouse#69744 from ClickHouse/fix_pr_protocol

556c04a

Fix parallel replicas protocol after ClickHouse#68424

Enmk pushed a commit to Altinity/ClickHouse that referenced this pull request Dec 22, 2024

Merge pull request ClickHouse#68424 from ClickHouse/adaptive_parallel…

cdc6380

…_replicas Adaptive mark_segment_size for parallel replicas

Enmk pushed a commit to Altinity/ClickHouse that referenced this pull request Dec 22, 2024

Merge pull request ClickHouse#69744 from ClickHouse/fix_pr_protocol

b5cf09b

Fix parallel replicas protocol after ClickHouse#68424

Enmk mentioned this pull request Jan 21, 2025

24.8.11 Altinity Stable Pre-release Altinity/ClickHouse#582

Merged

Enmk mentioned this pull request Jan 28, 2025

24.8.11 Altinity Stable Pre-release (2) Altinity/ClickHouse#607

Merged

This was referenced Feb 18, 2025

24.8.13 Altinity Stable Pre-release Altinity/ClickHouse#635

Closed

24.8.14 Altinity Stable Pre-release Altinity/ClickHouse#641

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Adaptive mark_segment_size for parallel replicas#68424

Adaptive mark_segment_size for parallel replicas#68424
nickitat merged 20 commits intomasterfrom
adaptive_parallel_replicas

nickitat commented Aug 15, 2024 •

edited

Loading

Uh oh!

robot-ch-test-poll4 commented Aug 15, 2024 •

edited by robot-clickhouse

Loading

Uh oh!

devcrafter left a comment

Uh oh!

Uh oh!

devcrafter Sep 10, 2024

Uh oh!

nickitat Sep 10, 2024

Uh oh!

nickitat commented Sep 10, 2024

Uh oh!

nickitat commented Sep 13, 2024

Uh oh!

nickitat commented Sep 16, 2024

Uh oh!

devcrafter Sep 16, 2024

Uh oh!

devcrafter Sep 17, 2024

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

	/// everything up to (1/number_of_replicas) portion of its work will be stolen by other replicas. And it owns (1/number_of_replicas) share of total number of marks.
	/// everything up to ((number_of_replicas-1)/number_of_replicas) portion of its work will be stolen by other replicas. And it owns (1/number_of_replicas) share of total number of marks.

Conversation

nickitat commented Aug 15, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Changelog category (leave one):

Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):

CI Settings (Only check the boxes if you know what you are doing):

Uh oh!

robot-ch-test-poll4 commented Aug 15, 2024 • edited by robot-clickhouse Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

devcrafter left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

devcrafter Sep 10, 2024

Choose a reason for hiding this comment

Uh oh!

nickitat Sep 10, 2024

Choose a reason for hiding this comment

Uh oh!

nickitat commented Sep 10, 2024

Uh oh!

nickitat commented Sep 13, 2024

Uh oh!

nickitat commented Sep 16, 2024

Uh oh!

devcrafter Sep 16, 2024

Choose a reason for hiding this comment

Uh oh!

devcrafter Sep 17, 2024

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

nickitat commented Aug 15, 2024 •

edited

Loading

robot-ch-test-poll4 commented Aug 15, 2024 •

edited by robot-clickhouse

Loading