Respect insert_allow_materialized_columns for INSERT into Distributed()

[azat]

Changelog category (leave one):

• Improvement

Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):
Respect insert_allow_materialized_columns (allows materialized columns) for INSERT into Distributed table.

Follow-up for: #5429
Refs: #16897

[alexey-milovidov]

Respect insert_allow_materialized_columns for INSERT into Distributed()

Also not a good enough changelog entry.
I will fix it.

[azat]

Also not a good enough changelog entry.

@alexey-milovidov I've adjusted it a little

[alexey-milovidov]

@Mergifyio update

[alexey-milovidov]

@azat Do you think, maybe Distributed table should always insert into materialized columns?

[mergify]

Command update: success

Branch has been successfully updated

[azat]

@azat Do you think, maybe Distributed table should always insert into materialized columns?

This an option, but why? Especially if there is insert_allow_materialized_columns setting.
Plus underlying table will still require this setting, and the INSERT will fail

[alexey-milovidov]

The user only see the structure of Distributed table - if it can successfully insert data into it, materialized columns in remote table should not prevent sending the data.

[azat]

materialized columns in remote table should not prevent sending the data.

So what you are suggesting?
Always avoid sending them into underlying table? (like w/o this PR)
Or force insert_allow_materialized_columns for INSERTs to the underlying table of Distributed?

[azat]

Functional stateless tests (release, DatabaseReplicated) — fail: 1, passed: 2921, skipped: 108
Integration tests (thread) — Timeout, fail: 1, passed: 1053, flaky: 0
PVS check — Have 2 new errors, 22 total errors
SQLancer test — SQLancer test run. See report

Unrelated.

Style Check — Style check failed.

Unrelated (it reports an issue in a file that this PR does not touches - PocoHTTPClient.cpp)

Yandex third-party checks (only for Yandex employees)

Does not looks related either.

[azat]

maybe Distributed table should always insert into materialized columns?

I have the following use case, I have something like distributed dictionary (since overall dictionary is quite large and will not fit into memory of one node).
To materialize values, I'm using Distributed table to deliver data into the shards, each shard contain part of that dictionary, and the column is materialized on the shard correctly (via Buffer engine with MATERIALIZED column).
However if Distributed table will materialize this column it will be incorrect.

Materialized columns can be removed from Distributed table, but then you cannot query them via Distributed table.
Yes you can do this via separate Distributed table, but looks a little bit hacky.