close($_);
}
fcntl ("writer", F_SETFL, 0) or croak ($!); # no close-on-exec
- open(STDOUT,">&writer");
- open(STDERR,">&writer");
+ open(STDOUT,">&writer") or croak ($!);
+ open(STDERR,">&writer") or croak ($!);
undef $dbh;
undef $sth;
delete $Jobstep->{tempfail};
$Jobstep->{'arvados_task'}->{started_at} = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($Jobstep->{starttime});
- $Jobstep->{'arvados_task'}->save;
+ retry_op(sub { $Jobstep->{'arvados_task'}->save; }, "job_tasks.update API");
splice @jobstep_todo, $todo_ptr, 1;
--$todo_ptr;
"ERROR: Task process exited %s, but never updated its task record to indicate success and record its output.",
exit_status_s($childstatus)));
$Jobstep->{'arvados_task'}->{success} = 0;
- $Jobstep->{'arvados_task'}->save;
+ retry_op(sub { $Jobstep->{'arvados_task'}->save; }, "job_tasks.update API");
$task_success = 0;
}
$Jobstep->{exitcode} = $childstatus;
$Jobstep->{finishtime} = time;
$Jobstep->{'arvados_task'}->{finished_at} = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($Jobstep->{finishtime});
- $Jobstep->{'arvados_task'}->save;
+ retry_op(sub { $Jobstep->{'arvados_task'}->save; }, "job_tasks.update API");
process_stderr_final ($jobstepidx);
Log ($jobstepidx, sprintf("task output (%d bytes): %s",
length($Jobstep->{'arvados_task'}->{output}),
close($stdout_r);
fcntl($stderr_w, F_SETFL, 0) or croak($!); # no close-on-exec
fcntl($stdout_w, F_SETFL, 0) or croak($!);
- open(STDERR, ">&", $stderr_w);
- open(STDOUT, ">&", $stdout_w);
+ open(STDERR, ">&", $stderr_w) or croak ($!);
+ open(STDOUT, ">&", $stdout_w) or croak ($!);
srun ($srunargs, $execargs, $opts, $stdin);
exit (1);
}
# that can be retried, the second function will be called with
# the current try count (0-based), next try time, and error message.
my $operation = shift;
- my $retry_callback = shift;
+ my $op_text = shift;
my $retries = retry_count();
+ my $retry_callback = sub {
+ my ($try_count, $next_try_at, $errmsg) = @_;
+ $errmsg =~ s/\s*\bat \Q$0\E line \d+\.?\s*//;
+ $errmsg =~ s/\s/ /g;
+ $errmsg =~ s/\s+$//;
+ my $retry_msg;
+ if ($next_try_at < time) {
+ $retry_msg = "Retrying.";
+ } else {
+ my $next_try_fmt = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($next_try_at);
+ $retry_msg = "Retrying at $next_try_fmt.";
+ }
+ Log(undef, "$op_text failed: $errmsg. $retry_msg");
+ };
foreach my $try_count (0..$retries) {
my $next_try = time + (2 ** $try_count);
my $result = eval { $operation->(@_); };
# This function will call that method, retrying as needed until
# the current retry_count is exhausted, with a log on the first failure.
my $method_name = shift;
- my $log_api_retry = sub {
- my ($try_count, $next_try_at, $errmsg) = @_;
- $errmsg =~ s/\s*\bat \Q$0\E line \d+\.?\s*//;
- $errmsg =~ s/\s/ /g;
- $errmsg =~ s/\s+$//;
- my $retry_msg;
- if ($next_try_at < time) {
- $retry_msg = "Retrying.";
- } else {
- my $next_try_fmt = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($next_try_at);
- $retry_msg = "Retrying at $next_try_fmt.";
- }
- Log(undef, "API method $method_name failed: $errmsg. $retry_msg");
- };
my $method = $arv;
foreach my $key (split(/\//, $method_name)) {
$method = $method->{$key};
}
- return retry_op(sub { $method->execute(@_); }, $log_api_retry, @_);
+ return retry_op(sub { $method->execute(@_); }, "API method $method_name", @_);
}
sub exit_status_s {
# Hide messages from the install script (unless it fails: shell_or_die
# will show $destdir.log in that case).
-open(STDOUT, ">>", "$destdir.log");
-open(STDERR, ">&", STDOUT);
+open(STDOUT, ">>", "$destdir.log") or die ($!);
+open(STDERR, ">&", STDOUT) or die ($!);
if (-e "$destdir/crunch_scripts/install") {
shell_or_die (undef, "$destdir/crunch_scripts/install", $install_dir);
sub can_run {
my $command_name = shift;
- open(my $which, "-|", "which", $command_name);
+ open(my $which, "-|", "which", $command_name) or die ($!);
while (<$which>) { }
close($which);
return ($? == 0);