Michael Paquier
Michael Paquier
Hmm.. We could get in get_hints_from_table() only if pg_hint_plan_enable_hint_table is enabled, and it is disabled by default. I'm OK to be consistent with compute_query_id being disabled on the fly, and...
Could it be simpler to just do a get_extension_oid() and see if the extension is installed? The table depends on the extension to exist.
> What do you think? Yeah.. There was a patch in upstream to add a syscache to extensions, which I'm going to apply for v18~, first. I could live with...
And done on all stable branches, with some tests.
Could you produce a self-contained test case that could be used for investigation, please? I am not familar with the JOB benchmark nor any of its queries, hence it would...
> I want to create a pull request to Postgres and make PG consider disable_cost for the disabled operators in the first phase of cost estimation. Would you like to...
Ahh, so the increment of disabled_nodes for the disabled operators is what comes into help here. Yes, let's close this ticket if it has no reason to be. Please note...
Not sure if this is still relevant. If you would like to see a new feature in this area, feel free to provide a patch with a pull request. Closing...
Instability of pg_hint_plan.sql test in PG16 + master on a slow box (or when testing with valgrind).
It is a very bad idea to penalize the tests with some hardcoded pg_sleep() calls that would make the tests even slower for everybody. IMO, we should redesign the tests...
Instability of pg_hint_plan.sql test in PG16 + master on a slow box (or when testing with valgrind).
This issue has been fixed by ea21f9599c3e, where I have redesigned the tests to rely more on specific APIs to force stats flushes, with counter manipulations. In short, the coverage...