[RFC] Fix possible parts duplicates after ALTER TABLE MOVE PART TO SHARD

[azat]

The patch is pretty icky, hence - RFC

Changelog category (leave one):

• Improvement

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

Fix possible parts duplicates after ALTER TABLE MOVE PART TO SHARD (not a production feature)

Before this patch ALTER TABLE MOVE PART TO SHARD has a race, and that
was the reason for flaky test.

Parts moves between shards works as follow (some steps are omitted, only
relevant is here):

1. I: writes parts for migration written to zookeeper (/pinned_parts_uuids)
2. I: queue request to read this parts from zookeeper (SYNC_SOURCE)
3. R1/R2: process SYNC_SOURCE
4. R1/R2: process SYNC_PINNED_PARTS_UUIDS (read parts from /pinned_parts_uuids)
5. R1/R2: triggers DESTINATION_FETCH (after this destionation shard could clone the part)

Where:

• I - initiator
• R1/R2 - replica 1 and 2

Now imagine the following, SELECT query started before (4) finished, and
now (4) finished, and so now remote nodes could already fetch data from
the replicas, but those SELECT query may know nothing about
pinned_parts_uuids and in this case, the query will not filter out
duplicates (since it is done only if there are pinned_parts_uuids - to
avoid overhead for tables without active parts movements), and this will
lead to duplicates in the query.

So to fix this I see only one (icky) option - do not allow any queries
in parallel with updating pinned_parts_uuids. Note, that this should be
enough because SYNC_PINNED_PARTS_UUIDS processed synchronously on all
replicas, so it is not possible that parts will appears on the
destination shards before it is finished. And AFAICS dead lock should
not be possible, since acquiring locks in reverse order
(exclusive_lock, pinned_part_uuids_mutex) does not happens.

Refs: #49579 (cc @alexey-milovidov )
Fixes: #49574
Follow-up for: #17348 (cc @xjewer)
Follow-up for: #17871 (cc @nvartolomei)

Another option is to remove this feature all together, though it could be useful for automatic re-sharding

[robot-ch-test-poll4]

This is an automated comment for commit d89561c 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
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	🟢 success
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	🟢 success

[alesapin]

I would prefer to remove this experimental feature.

[azat]

I would prefer to remove this experimental feature.

@alesapin This could be one of possible building block for automatic data re-sharding, can you explain your concerns about this feature?

[alexey-milovidov]

This building block is too complex and fragile, it is better to remove it entirely and write a new one.

[azat]

Actually I don't like this patch, since it uses the lock that is not intended to leak into such places I would say, and I think that there will be a better way to distribute parts between replicas.

However I was thinking about keeping this feature until #45766 will be resolved.

So if you also don't like this patch as well, then let's close this, and I guess that this feature should be removed completely then, since it has this problem (even though it is pretty rare, but still) and it looks like original authors are not interested in it anymore.

[alexey-milovidov]

Ok. I also thought that we don't need part UUIDs. Instead, we should identify parts by their content hashes, as they are perfectly immutable.