Hi,
On Wed, Feb 05, 2025 at 03:12:53PM +0900, Michael Paquier wrote:
> On Thu, Oct 31, 2024 at 08:47:04AM +0900, Michael Paquier wrote:
> > Thanks for the review. Applied that, then.
>
> I was looking at src/test/isolation/README, and based on what is
> described for parenthesized markers (which I didn't know about so we
> learn a new thing every day), it is possible to force a strict
> ordering of the test. Quote that matters:
> "A marker consisting solely of a step name indicates that this step
> may not be reported as completing until that other step has
> completed."
Oh, I did not know about it too (thanks for sharing).
> In our case, the problem that caused 9f00edc22888 is that the "wait"
> step could be so slow that the "detach" step could report before the
> "wait" step.
Yup.
> So I'd like to propose that we revisit 9f00edc22888, and re-enable the
> permutation with a marker so as the "detach" step waits for the "wait"
> step to complete before printing its own report, as per the attached.
> That should make the permutation safe to use.
>
> The output of the attached patch is the same as what has been removed
> in 9f00edc22888 except that the permutation is changed from that:
> permutation wait1 wakeup2 detach2
> To that, making sure that detach2 report happens only after wait1
> has returned its own result:
> permutation wait1 wakeup2 detach2(wait1)
>
> Thoughts?
I think that makes sense and the patch LGTM.
A few tests are already using this technique (including injection_points in
inplace.spec).
Regards,
--
Bertrand Drouvot
PostgreSQL Contributors Team
RDS Open Source Databases
Amazon Web Services: https://aws.amazon.com
