Index support for compress chunk

[shhnwz]

Index support for compress chunk

It allows to override tuplesort with indexscan
if compression setting keys matches with Index keys.
Moreover this feature has Enable/Disable Toggle.
To Disable from the client use the following command,
SET timescaledb.enable_compression_indexscan = 'OFF'

Fixes #4617

[CLAassistant]

All committers have signed the CLA.

[codecov]

Codecov Report

Merging #4821 (d1a5bb7) into main (cbf5180) will increase coverage by 0.03%.
The diff coverage is 94.65%.

❗ Current head d1a5bb7 differs from pull request most recent head e45a557. Consider uploading reports for the commit e45a557 to get more accurate results

@@            Coverage Diff             @@
##             main    #4821      +/-   ##
==========================================
+ Coverage   89.59%   89.63%   +0.03%     
==========================================
  Files         227      226       -1     
  Lines       51623    51220     -403     
==========================================
- Hits        46252    45910     -342     
+ Misses       5371     5310      -61     

Impacted Files	Coverage Δ
tsl/src/compression/compression.c	96.09% <94.61%> (-0.12%)	⬇️
src/guc.c	94.11% <100.00%> (ø)
tsl/src/remote/copy_fetcher.c	78.39% <0.00%> (-6.81%)	⬇️
src/loader/bgw_message_queue.c	86.36% <0.00%> (-2.28%)	⬇️
src/nodes/chunk_dispatch.c	92.85% <0.00%> (-0.90%)	⬇️
src/time_bucket.c	95.28% <0.00%> (-0.54%)	⬇️
tsl/src/remote/connection.c	85.10% <0.00%> (-0.36%)	⬇️
tsl/src/fdw/modify_exec.c	84.97% <0.00%> (-0.26%)	⬇️
tsl/src/continuous_aggs/options.c	95.10% <0.00%> (-0.25%)	⬇️
tsl/src/continuous_aggs/create.c	87.68% <0.00%> (-0.22%)	⬇️
... and 47 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update cbf5180...e45a557. Read the comment docs.

[akuzm]

Needs more tests, probably good to put them into a separate test file.

1. some way to tell from the tests which code is used -- maybe a debug message? I don't really like them because then suddenly other debug messages pop up from unrelated parts of code, and your test becomes flaky. Maybe we can add a debug option which enables an INFO message about which code path is taken.
2. all combinations of desc/asc and nulls first/last that match the index or not.
3. different index type -- we can only use btree here, not gist and so on.

[svenklemm]

I think we are missing check for collation. Can you add tests with different collation. In the index check the collation must match for an index to be elegible for datatypes that are collation sensitive.

[akuzm]

Looks like we still need more tests:

1. When collation doesn't match
2. Two segmentby columns, both the case where they require the same order and the different one. Codecov seems to complain that this is not covered.
3. Would be good to construct some kind of test that shows the compression went correctly, and the compressed groups are the same as w/o the index. Maybe select segmentby + metadata columns from the compressed chunk?

[akuzm]

You're not resetting current_direction here, this will lead to an index still being chosen if the last column didn't match, right?

We have to fix some things:

1. add a test for this bug. Try to reproduce it on the unfixed version and verify that the last column is chosen incorrectly.
2. Reset current_direction at the start of the loop to avoid bugs like this.
3. Rewrite this as early exit from the loop:

					if (att_num == 0 || index_info->ii_IndexAttrNumbers[i] != att_num)
					{
					    break;
					}

Long if branches are hard to read, so early exit with break/return is preferrable.

[akuzm]

Attnum can't be zero here, right? The orderby/segmentby keys must be present in the table. We should check this with an assertion.

[akuzm]

Can att_num == 0 happen in practice? This means that the orderby/segmentby column is not found in the uncompressed chunk. That would be either a program error or a data corruption. Let's use Ensure macro to check this, not if.