Re: Still another race condition in recovery TAP tests

Michael Paquier <michael(dot)paquier(at)gmail(dot)com> writes:
> On Sun, Sep 10, 2017 at 12:38 AM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>> The specific case we need to allow is "ENOENT on a file/dir that was
>> there a moment ago". I think it still behooves us to complain about
>> anything else. If you think it's a simple fix, have at it. But
>> I see at least three ways for _copypath_recurse to fail depending on
>> exactly when the file disappears.

> With the check for -d and -f, this brings indeed the count to three
> code paths. With the patch attached, I have added some manual sleep
> calls in RecursiveCopy.pm and triggered manual deletions of the source
> repository. The copy is able to complete in any case, warning about
> missing directories or files. I have added warn messages in the patch
> when ENOENT is triggered, though I think that those should be removed
> in the final patch.

Hm, I don't much like having it silently ignore files that are present;
that seems like a foot-gun in the long run. What do you think of the
attached?

> By the way, 010_logical_decoding_timelines.pl does not need to include
> RecursiveCopy.

Good catch.

regards, tom lane