Add test for repair table script

[mkindahl]

This commit copies the repair query into a separate file and turns
it into a procedure. This will allow the repair query to be tested
separately using pg_regress. It also adds a test script for the
repair procedure and will remove different sorts of dimensions and see
that they can be properly repaired.

It also contain a fix for the timestamp parsing query that prevented it
from handling all timestamp formats and a fix so that if many chunks refer to the same dimension slice, it might generate duplicate rows.

[codecov]

Codecov Report

Merging #2065 into master will increase coverage by 0.23%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #2065      +/-   ##
==========================================
+ Coverage   90.01%   90.24%   +0.23%     
==========================================
  Files         210      211       +1     
  Lines       33255    33643     +388     
==========================================
+ Hits        29933    30361     +428     
+ Misses       3322     3282      -40     

Impacted Files	Coverage Δ
tsl/src/fdw/shippable.c	82.85% <0.00%> (-11.43%)	⬇️
tsl/src/fdw/data_node_scan_exec.c	82.69% <0.00%> (-1.63%)	⬇️
src/loader/bgw_message_queue.c	84.51% <0.00%> (-1.02%)	⬇️
tsl/src/debug.c	39.89% <0.00%> (-0.33%)	⬇️
tsl/src/compression/create.c	95.16% <0.00%> (-0.26%)	⬇️
src/plan_agg_bookend.c	92.34% <0.00%> (-0.26%)	⬇️
src/hypertable.c	90.84% <0.00%> (-0.11%)	⬇️
src/chunk.c	95.29% <0.00%> (-0.05%)	⬇️
src/compat.h	100.00% <0.00%> (ø)
tsl/test/src/remote/stmt_params.c	100.00% <0.00%> (ø)
... and 27 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 41c89b5...f81ce58. Read the comment docs.

[mkindahl]

There is one unfortunate situation in that we need to have two copies of the repair script. I fear that might result in mistakes when one script is updated but not the other. Include commands do not work with the update scripts, but we could have version-specific pre- and post-scripts and use these to generate the actual scripts in a similar way to have latest-dev.sql is used.

[erimatnor]

It doesn't seem right to me that the update script is tested in a regression test. Why isn't it part of the update test?

The odd thing is that this is a script that needs to run only once on upgrades of specific versions of the database, so it doesn't seem right that this has to run as part of our standard regression suite. It also doesn't test that the update script actually works during an update to a new version of the extension.

[mkindahl]

It doesn't seem right to me that the update script is tested in a regression test. Why isn't it part of the update test?

The odd thing is that this is a script that needs to run only once on upgrades of specific versions of the database, so it doesn't seem right that this has to run as part of our standard regression suite. It also doesn't test that the update script actually works during an update to a new version of the extension.

Not sure I follow. Pg_regress will start a clean server and test that the repair script will repair tables, so it's perfect for testing the repair script. It is not a complete test that the update script works, it just test that it can repair the dimension slice table.

We already have update tests, but they cannot catch all kind of problems. They just test that a correct installation can be updated.

[svenklemm]

I agree with @erimatnor the script is run when upgrading from 1.7.1 to 1.7.2 so there is no need in maintaining it and adjusting it to catalog changes because it needs to run in the 1.7.1 environment therefore regression tests are not the right place.

[mkindahl]

Closing.