percona/percona-toolkit
1
0
Fork 0
mirror of https://github.com/percona/percona-toolkit.git synced 2025-09-17 17:27:57 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add another test for reconnect, and fix failure/race condition in pt-slave-delay

Browse Source
This commit is contained in:
Baron Schwartz
2012-06-04 11:33:06 -04:00
parent 9aaafe8478
commit 56e20d269a
2 changed files with 40 additions and 40 deletions
Download Patch File Download Diff File Expand all files Collapse all files

73
bin/pt-slave-delay
View File

@@ -2117,44 +2117,43 @@ sub main {
$now = time();
# TODO: this is a race condition. See 0xdeadbeef below.
if ( !$slave_dbh || !$slave_dbh->ping() ) {
# Try 10 times, for about 2 minutes, to reconnect to the slave,
# increasing wait time from 3 to 15 seconds.
$o->set('ask-pass', 0); # don't ask again
my $tries = 10;
my $rt = new Retry();
$rt->retry(
tries => $tries,
retry_on_die => 1,
wait => sub {
my ( %args ) = @_;
return unless $oktorun;
my $t = min($args{tryno} * 3, 15);
info("Could not reconnect to slave, sleeping $t seconds "
. "and trying " . ($tries-$args{tryno}) . " more times")
if $tries - $args{tryno};
sleep $t;
},
try => sub {
return unless $oktorun;
info("Lost connection to slave, trying to reconnect");
# If the database connection is gone, we must live on!
# Try 10 times, for about 2 minutes, to reconnect to the slave,
# increasing wait time from 3 to 15 seconds.
$o->set('ask-pass', 0); # don't ask again
my $tries = 10;
my $rt = new Retry();
$rt->retry(
tries => $tries,
retry_on_die => 1,
wait => sub {
my ( %args ) = @_;
return unless $oktorun;
my $t = min($args{tryno} * 3, 15);
info("Lost connection, sleeping $t seconds "
. "and trying " . ($tries-$args{tryno}) . " more times")
if $tries - $args{tryno};
sleep $t;
info("Trying to reconnect");
eval {
$slave_dbh = get_dbh($dp, $slave_dsn);
return $slave_dbh;
},
on_success => sub {
info("Reconnected to slave");
},
on_failure => sub {
return unless $oktorun;
die "Failed to reconnect to slave";
},
);
last unless $oktorun; # might have gotten interrupt while waiting
}
# 0xdeadbeef (see above): just because we reconnected in the above Retry
# does not mean we have a connection here!
$status = $slave_dbh->selectrow_hashref("SHOW SLAVE STATUS");
};
},
try => sub {
return unless $oktorun;
$status = $slave_dbh->selectrow_hashref("SHOW SLAVE STATUS");
return $status;
},
on_success => sub {
info("Reconnected to slave");
},
on_failure => sub {
return unless $oktorun;
die "Failed to reconnect to slave";
},
);
last unless $oktorun; # might have gotten interrupt while waiting
if ( !$status || ! %$status ) {
die "No SLAVE STATUS found";
}