Better handle interrupting TAP tests

Set up a signal handler for INT/TERM so that we run our END block if we
get them.  In END, if the exit status indicates a problem, call
_update_pid(-1) to improve chances of the stop working in case start()
hasn't returned yet.

Also, change END's teardown_node() so that it passes fail_ok=>1, so that
if a node fails to stop, we still stop the other nodes in the same test.

Per complaint from Andres Freund.

This doesn't seem important enough to backpatch, at least for now.

Discussion: https://postgr.es/m/20220930040734.mbted42oiynhn2t6@awork3.anarazel.de
This commit is contained in:
Alvaro Herrera 2022-10-19 17:09:51 +02:00
parent 342bb38bfe
commit 460c0076e8
No known key found for this signature in database
GPG Key ID: 1C20ACB9D5C564AE
1 changed files with 19 additions and 3 deletions

View File

@ -1545,7 +1545,14 @@ END
foreach my $node (@all_nodes)
{
$node->teardown_node;
# During unclean termination (which could be a signal or some
# other failure), we're not sure that the status of our nodes
# has been correctly set up already, so try and update it to
# improve our chances of shutting them down.
$node->_update_pid(-1) if $exit_code != 0;
# If that fails, don't let that foil other nodes' shutdown
$node->teardown_node(fail_ok => 1);
# skip clean if we are requested to retain the basedir
next if defined $ENV{'PG_TEST_NOCLEAN'};
@ -1564,13 +1571,15 @@ END
Do an immediate stop of the node
Any optional extra parameter is passed to ->stop.
=cut
sub teardown_node
{
my $self = shift;
my ($self, %params) = @_;
$self->stop('immediate');
$self->stop('immediate', %params);
return;
}
@ -2922,6 +2931,13 @@ sub corrupt_page_checksum
return;
}
#
# Signal handlers
#
$SIG{TERM} = $SIG{INT} = sub {
die "death by signal";
};
=pod
=back