Re: test: 'seen set from rename' potentially racy

[Eric Wong <e@80x24.org>]

Konstantin Ryabitsev <konstantin@linuxfoundation.org> wrote:
> Hello:
> 
> I was having some trouble using "make check" on the Fedora build system:
> 
> #   Failed test 'seen set from rename'
> #   at t/lei-watch.t line 61.
> #                   '/tmp/pi-lei-watch-11080-FYz2/lei-daemon/md2/cur/9bf1002c49eb075df47247b74d69bcd555e23422=99:2,'
> #     doesn't match '(?^:S\z)'
> # [
> #   '/tmp/pi-lei-watch-11080-FYz2/lei-daemon/md2/cur/9bf1002c49eb075df47247b74d69bcd555e23422=99:2,'
> # ]
> 
> This wasn't happening in my own chroot builds, so I'm suspecting that the
> faster system on the Fedora build infra side is causing this failure,
> therefore perhaps some kind of race condition. 

Hmm, I thought I fixed it...  Does increasing the tick sleep time help?

diff --git a/t/lei-watch.t b/t/lei-watch.t
index 9158571c..8126b707 100644
--- a/t/lei-watch.t
+++ b/t/lei-watch.t
@@ -49,13 +49,13 @@ test_lei(sub {
 	lei_ok 'ls-watch';
 	like($lei_out, qr/^\Qmaildir:$md\E$/sm, 'maildir shown');
 	lei_ok qw(q mid:testmessage@example.com -o), $md, '-I', "$ro_home/t1";
 	my @f = glob("$md/cur/*:2,");
 	is(scalar(@f), 1, 'got populated maildir with one result');
 	rename($f[0], "$f[0]S") or xbail "rename $!"; # set (S)een
-	tick($have_fast_inotify ? 0.1 : 2.1); # always needed for 1 CPU systems
+	tick($have_fast_inotify ? 1 : 5); # always needed for 1 CPU systems
 	lei_ok qw(note-event done); # flushes immediately (instead of 5s)
 
 	lei_ok qw(q mid:testmessage@example.com -o), $md2, '-I', "$ro_home/t1";
 	my @f2 = glob("$md2/*/*");
 	is(scalar(@f2), 1, 'got one result');
 	like($f2[0], qr/S\z/, 'seen set from rename') or diag explain(\@f2);

Maybe 1s is too much and 0.2s is enough.

> https://download.copr.fedorainfracloud.org/results/icon/b4/fedora-35-x86_64/02928611-public-inbox/build.log.gz

"prove --state=save -bvw -j3", so 2 cores?  Any idea what
CONFIG_HZ is on that system?

> Switching to "make test" seems to help, which is what I've done for now.

OK, so it could be load related.  There could probably be a way
to "subscribe" for updates to lei-daemon so sleeping wouldn't be
necessary.  I'll see abouit it next week, got some other stuff
to take care of...