Fix excessive memory usage for FINAL (due to too much streams usage) #50429

azat · 2023-06-01T12:42:39Z

Changelog category (leave one):

Performance Improvement

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

Fix excessive memory usage for FINAL (due to too much streams usage)

Previously it could create MergeTreeInOrder for each mark, however this could be very suboptimal, due to each MergeTreeInOrder has some memory overhead.

Now, by collapsing all marks for one part together it is more memory effiecient.

I've tried the query from the altinity wiki 1 and it decreases memory usage twice:

SELECT * FROM repl_tbl FINAL WHERE key IN (SELECT toUInt32(number) FROM numbers(1000000) WHERE number % 50000 = 0) FORMAT Null

upstream: MemoryTracker: Peak memory usage (for query): 520.27 MiB.
patched: MemoryTracker: Peak memory usage (for query): 260.95 MiB.

And it could be not 2x and even more or less, it depends on the gaps in marks for reading (for example in my setup the memory usage increased a lot, from ~16GiB of RAM to >64GiB due to lots of marks and gaps).

Cc: @nickitat
Cc: @UnamedRus (you may want to run your queries)

Note: this pops up after a proper marks skipping for FINAL in #47801

P.S. Marked as bug fix to run tests on previous version.

robot-clickhouse-ci-2 · 2023-06-01T12:43:07Z

This is an automated comment for commit 228ebab with description of existing statuses. It's updated for the latest CI running
The full report is available here
The overall status of the commit is 🔴 failure

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	🟢 success
Bugfix validate check	Checks that either a new test (functional or integration) or there some changed tests that fail with the binary built on master branch	🟢 success
CI running	A meta-check that indicates the running CI. Normally, it's in success or pending state. The failed status indicates some problems with the PR	🟡 pending
ClickHouse build check	Builds ClickHouse in various configurations for use in further steps. You have to fix the builds that fail. Build logs often has enough information to fix the error, but you might have to reproduce the failure locally. The cmake options can be found in the build log, grepping for cmake. Use these options and follow the general build process	🟢 success
Compatibility check	Checks that clickhouse binary runs on distributions with old libc versions. If it fails, ask a maintainer for help	🟢 success
Docker image for servers	The check to build and optionally push the mentioned image to docker hub	🟢 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. Integrational 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
Mergeable Check	Checks if all other necessary checks are successful	🟢 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
Push to Dockerhub	The check for building and pushing the CI related docker images to docker hub	🟢 success
SQLancer	Fuzzing tests that detect logical bugs with SQLancer tool	🟢 success
Sqllogic	Run clickhouse on the sqllogic test set against sqlite and checks that all statements are passed	🟢 success
Stateful tests	Runs stateful functional tests for ClickHouse binaries built in various configurations -- release, debug, with sanitizers, etc	🟢 success
Stateless tests	Runs stateless functional tests for ClickHouse binaries built in various configurations -- release, debug, with sanitizers, etc	🔴 failure
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	🔴 failure

Previously it could create MergeTreeInOrder for each mark, however this could be very suboptimal, due to each MergeTreeInOrder has some memory overhead. Now, by collapsing all marks for one part together it is more memory effiecient. I've tried the query from the altinity wiki [1] and it decreases memory usage twice: SELECT * FROM repl_tbl FINAL WHERE key IN (SELECT toUInt32(number) FROM numbers(1000000) WHERE number % 50000 = 0) FORMAT Null - upstream: MemoryTracker: Peak memory usage (for query): 520.27 MiB. - patched: MemoryTracker: Peak memory usage (for query): 260.95 MiB. [1]: https://kb.altinity.com/engines/mergetree-table-engine-family/replacingmergetree/#multiple-keys And it could be not 2x and even more or less, it depends on the gaps in marks for reading (for example in my setup the memory usage increased a lot, from ~16GiB of RAM to >64GiB due to lots of marks and gaps). Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>

azat · 2023-06-03T18:58:42Z

Stateless tests (release) — fail: 1, passed: 5315, skipped: 6

Ignore QEMU logging regarding IFA #50442

Stateless tests (release, wide parts enabled) — fail: 1, passed: 5313, skipped: 8

02780_final_streams_data_skipping_index - related, fixed

azat · 2023-06-04T09:03:02Z

Stateless tests (release, analyzer) — fail: 1, passed: 5310, skipped: 16

Fix flaky tests caused by OPTIMIZE FINAL failing memory budget check #49764

Stateless tests (release, s3 storage) [2/2] — fail: 1, passed: 2662, skipped: 28

01565_query_loop_after_client_error - interesting failure - Make 01565_query_loop_after_client_error slightly more robust #50542

nickitat · 2023-06-07T10:24:26Z

src/Processors/QueryPlan/PartsSplitter.cpp

-        result_layers.emplace_back();
+        auto & current_layer = result_layers.emplace_back();
+        /// Map part_idx into index inside layer, used to merge marks from the same part into one reader
+        std::unordered_map<size_t, size_t> part_idx_in_layer;


shouldn't it be just ssize_t last_part_idx?

Looks like this will not be enough, since parts_ranges_queue is sorted by primary key value not by the part, plus there are modifications of this queue inside this loop.

yep, I misunderstood your approach initially.

nickitat

nice improvement, thanks

nickitat · 2023-06-07T11:34:48Z

src/Processors/QueryPlan/PartsSplitter.cpp

-        result_layers.emplace_back();
+        auto & current_layer = result_layers.emplace_back();
+        /// Map part_idx into index inside layer, used to merge marks from the same part into one reader
+        std::unordered_map<size_t, size_t> part_idx_in_layer;


yep, I misunderstood your approach initially.

li-zhao-hotstar · 2023-08-21T06:43:54Z

thanks for the fix!

do we know the plan this can be released / backported? since the LTS release 23.3 contains #47801, we had observed critical memory issues with it.

…lickHouse#50429) Previously it could create MergeTreeInOrder for each mark, however this could be very suboptimal, due to each MergeTreeInOrder has some memory overhead. Now, by collapsing all marks for one part together it is more memory effiecient. I've tried the query from the altinity wiki [1] and it decreases memory usage twice: SELECT * FROM repl_tbl FINAL WHERE key IN (SELECT toUInt32(number) FROM numbers(1000000) WHERE number % 50000 = 0) FORMAT Null - upstream: MemoryTracker: Peak memory usage (for query): 520.27 MiB. - patched: MemoryTracker: Peak memory usage (for query): 260.95 MiB. [1]: https://kb.altinity.com/engines/mergetree-table-engine-family/replacingmergetree/#multiple-keys And it could be not 2x and even more or less, it depends on the gaps in marks for reading (for example in my setup the memory usage increased a lot, from ~16GiB of RAM to >64GiB due to lots of marks and gaps). Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>

…l-memory-usage 23.3 Backport of ClickHouse#50429 - Fix excessive memory usage for FINAL (due to too much streams usage)

…lickHouse#50429) Previously it could create MergeTreeInOrder for each mark, however this could be very suboptimal, due to each MergeTreeInOrder has some memory overhead. Now, by collapsing all marks for one part together it is more memory effiecient. I've tried the query from the altinity wiki [1] and it decreases memory usage twice: SELECT * FROM repl_tbl FINAL WHERE key IN (SELECT toUInt32(number) FROM numbers(1000000) WHERE number % 50000 = 0) FORMAT Null - upstream: MemoryTracker: Peak memory usage (for query): 520.27 MiB. - patched: MemoryTracker: Peak memory usage (for query): 260.95 MiB. [1]: https://kb.altinity.com/engines/mergetree-table-engine-family/replacingmergetree/#multiple-keys And it could be not 2x and even more or less, it depends on the gaps in marks for reading (for example in my setup the memory usage increased a lot, from ~16GiB of RAM to >64GiB due to lots of marks and gaps). Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>

alexey-milovidov · 2023-10-18T22:40:56Z

@azat This isn't a bug fix, and it shouldn't have been backported.

azat · 2023-10-19T09:18:31Z

@alexey-milovidov I've marked it as bug fix to run tests on previous version, and AFAIK there is no automatic backports of the bug fixes anyway, so this had been done explicitly and not by me.

azat · 2023-10-19T09:19:22Z

And what was the problem with the backport? Did it create any bugs?

alexey-milovidov · 2023-10-20T02:09:13Z

@azat excessive backports compromise stability by stealing attention from important changes.

UnamedRus · 2023-10-20T02:28:33Z

AFAIK, there was a discussion and complain, about why people stuck to old versions of ClickHouse and not newer one, which is still in 1 year LTS support.

And if "newer" version of ClickHouse have regressions compared to older ones, which makes them 10-100 times slower for them, it's the answer why this (keeping to older versions) will continue to happen and people will look away from using newer releases.

And no, upgrade to latest version, which have all fixes is not an answer for them, #49410 #55643

alexey-milovidov · 2023-10-20T02:34:20Z

We should not do these types of backports.

robot-clickhouse-ci-2 added the pr-bugfix Pull request with bugfix, not backported by default label Jun 1, 2023

nickitat self-assigned this Jun 1, 2023

azat force-pushed the fix-final-memory-usage branch from 162ebc8 to 228ebab Compare June 3, 2023 18:58

azat requested a review from nickitat June 7, 2023 09:41

nickitat reviewed Jun 7, 2023

View reviewed changes

nickitat approved these changes Jun 7, 2023

View reviewed changes

nickitat merged commit 036ddcd into ClickHouse:master Jun 7, 2023
259 of 264 checks passed

azat deleted the fix-final-memory-usage branch June 7, 2023 12:15

nickitat added the v23.3-must-backport label Aug 21, 2023

robot-clickhouse-ci-1 mentioned this pull request Aug 21, 2023

Cherry pick #50429 to 23.3: Fix excessive memory usage for FINAL (due to too much streams usage) #53632

Closed

robot-ch-test-poll1 added the pr-backports-created Backport PRs are successfully created, it won't be processed by CI script anymore label Sep 6, 2023

Enmk mentioned this pull request Sep 20, 2023

23.3 Backport of #50429 - Fix excessive memory usage for FINAL (due to too much streams usage) Altinity/ClickHouse#307

Merged

Enmk added a commit to Altinity/ClickHouse that referenced this pull request Sep 21, 2023

Merge pull request #307 from Altinity/backports/23.3.8/50429_fix-fina…

c709951

…l-memory-usage 23.3 Backport of ClickHouse#50429 - Fix excessive memory usage for FINAL (due to too much streams usage)

This was referenced Sep 25, 2023

23.3.13 Pre-release Altinity/ClickHouse#310

Merged

23.3 Fixed stateless test 02780_final_streams_data_skipping_index Altinity/ClickHouse#312

Merged

alexey-milovidov removed the pr-bugfix Pull request with bugfix, not backported by default label Oct 18, 2023

Enmk mentioned this pull request Mar 1, 2024

23.3.19 Pre-release PR for 23.3.19.33.altinitystable Altinity/ClickHouse#370

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fix excessive memory usage for FINAL (due to too much streams usage) #50429

Fix excessive memory usage for FINAL (due to too much streams usage) #50429

azat commented Jun 1, 2023 •

edited by alexey-milovidov

robot-clickhouse-ci-2 commented Jun 1, 2023 •

edited by robot-ch-test-poll

azat commented Jun 3, 2023 •

edited

azat commented Jun 4, 2023

nickitat Jun 7, 2023

azat Jun 7, 2023

nickitat Jun 7, 2023

nickitat left a comment

nickitat Jun 7, 2023

li-zhao-hotstar commented Aug 21, 2023

alexey-milovidov commented Oct 18, 2023

azat commented Oct 19, 2023

azat commented Oct 19, 2023

alexey-milovidov commented Oct 20, 2023

UnamedRus commented Oct 20, 2023

alexey-milovidov commented Oct 20, 2023

Fix excessive memory usage for FINAL (due to too much streams usage) #50429

Fix excessive memory usage for FINAL (due to too much streams usage) #50429

Conversation

azat commented Jun 1, 2023 • edited by alexey-milovidov

Changelog category (leave one):

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

robot-clickhouse-ci-2 commented Jun 1, 2023 • edited by robot-ch-test-poll

azat commented Jun 3, 2023 • edited

azat commented Jun 4, 2023

nickitat Jun 7, 2023

Choose a reason for hiding this comment

azat Jun 7, 2023

Choose a reason for hiding this comment

nickitat Jun 7, 2023

Choose a reason for hiding this comment

nickitat left a comment

Choose a reason for hiding this comment

nickitat Jun 7, 2023

Choose a reason for hiding this comment

li-zhao-hotstar commented Aug 21, 2023

alexey-milovidov commented Oct 18, 2023

azat commented Oct 19, 2023

azat commented Oct 19, 2023

alexey-milovidov commented Oct 20, 2023

UnamedRus commented Oct 20, 2023

alexey-milovidov commented Oct 20, 2023

azat commented Jun 1, 2023 •

edited by alexey-milovidov

robot-clickhouse-ci-2 commented Jun 1, 2023 •

edited by robot-ch-test-poll

azat commented Jun 3, 2023 •

edited