Add support for splitting a chunk

[erimatnor]

A chunk can be split with a new procedure called split_chunk(). In this initial version, a chunk can only be split in two given a split point:

call split_chunk('chunk_1', split_at => '2025-03-01 00:00');

If no split point is given, the chunk is split in two equal size partition ranges. The partitioning dimension/column to split along can also be specified, but only the primary dimension (time) is supported. Future updates to split_chunk() can add the ability to split also along other dimensions, e.g., space partitions.

Currently, splitting a chunk takes an AccessExclusiveLock on the chunk being split. A lock is also taken on the hypertable root to prevent new chunks being created while the split is ongoing.

[erimatnor]

I want to add some concurrency/isolation tests. But might do it in another PR.

[codecov]

Codecov Report

Attention: Patch coverage is 84.54545% with 34 lines in your changes missing coverage. Please review.

Project coverage is 82.15%. Comparing base (59f50f2) to head (2807d8a).
Report is 908 commits behind head on main.

Files with missing lines	Patch %	Lines
tsl/src/chunk.c	84.47%	9 Missing and 25 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #7946      +/-   ##
==========================================
+ Coverage   80.06%   82.15%   +2.08%     
==========================================
  Files         190      250      +60     
  Lines       37181    46504    +9323     
  Branches     9450    11670    +2220     
==========================================
+ Hits        29770    38205    +8435     
- Misses       2997     3671     +674     
- Partials     4414     4628     +214     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[erimatnor]

Going to work on expanding tests to improve the code coverage.

[mkindahl]

There seems to be a few minor things that might be good to take a look at to make sure that they are not causing issues.

Looking for ways this could break, you might want to test these cases in the part of the test where you actually check the contents of the tuple (because there is potential for corrupting the data and printing out the tuples is more likely to reveal issues):

• No tuples in table and splitting
• Deleting and/or updating tuples in the table and splitting (you do vacuum as part of the split).
• Odd number of tuples in table and splitting
• Splitting table so that one part is empty and the other contains all of them (two cases, original is empty and new chunk is empty).
• Using split chunk with the start timestamp of the chunk
• Using split chunk with the end timestamp of the chunk
• Using both byval types (int, for example) and byref types (text, for example). You are using byval types, but I see no byref types.

[fetchezar]

Hi! I wanted to ask how this works regarding compression. For example I have a database with a chunk size that's too large for the resources it has, 180 days. This chunk ends by the end of the year.

Ideally I wouldn't want to wait until that chunk is done to compress it, and it seems this function would let me split it in two chunks of, say, 90 days, compress the first, adapt the chunk policy so it compresses and segments at 90 days, and all should work fine, right?

This is our use case for splitting and merging. Basically fixing bad estimations for hypertable chunks in either way (or sometimes cases where compression works really well and we could merge several old compressed chunks for performance reasons).

Does this splitting support both compressed and uncompressed chunks?

[erimatnor]

Does this splitting support both compressed and uncompressed chunks?

Thanks for asking. The first version won't support compressed chunks, but this will be added in a follw-up PR.

[mkindahl]

Might be good to add a test that splits into an empty new or empty old chunk. I did not see any such test.

[erimatnor]

Might be good to add a test that splits into an empty new or empty old chunk. I did not see any such test.

Added.

[fetchezar]

Does this splitting support both compressed and uncompressed chunks?

Thanks for asking. The first version won't support compressed chunks, but this will be added in a follw-up PR.

I see, thank you!