add option to force deduplicate children

[urykhy]

I hereby agree to the terms of the CLA available at: https://yandex.ru/legal/cla/?lang=en

Changelog category (leave one):

• Improvement

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

reduce dataloss chance on insert into table with mv.

Detailed description / Documentation draft:

if you insert data into replicated table with mv, there is a chance, that ZooKeeper exception can cause data inserted into table only, since if CH do not try to insert into mv if block already deduplicated by main table. Usually this works well, but if ZooKeeper error occures after insert into main table, this block will never reach MV. If u set force_deduplicate_childrens=1, then CH will try to insert block into MV even if it deduplicated by parent table.

partial fix for #2621

[akuzm]

Is it possible to add some kind of integration test for this?

[urykhy]

@akuzm i'm not sure

we arrange test environment to trigger this kind of error, but:

• it's build from closed components (uploader, tests, test data)
• not 100% accuracy, this patch only reduce rate of error. it's still possible that data will not reach MV or reach MV twice.
• time to run at least 15 minutes.

[den-crane]

@urykhy at least you can emulate MV insert fail
https://gist.github.com/den-crane/bc35a15a8d71899fe15e5ed36668d387

As far as I understand with your setting the second insert (retry) should insert rows into test12345mv and SELECT count() FROM test12345mv should return rows.

You can implement two test cases with your setting and without.
I assume these tests will be pretty stable and fast enough.

[alexey-milovidov]

01083_expressions_in_engine_arguments

• well known issue.