#!/usr/athena/bin/perl # $Header: /afs/sipb.mit.edu/machine/charon2/src/RCS/nightly_jobs.pl,v 1.3 1999/10/01 05:26:07 jweiss Exp $ # require 'syslog.pl'; use Sys::Syslog; require 'errno.ph'; eval "require 'sys/resource.ph';"; # Takes a list of jobs and runs them one after the other, in a subshell, # with a separate AFS process group for each one. Each job also has an # owner, and if the job has any output, it is sent to the owner. $tf = "/var/tmp/nightly.$$"; # for output of each job $jobfile = '/etc/nightly_jobs.conf'; $pidfile = '/etc/nightly_jobs.pid'; $debug = 0; $usage = "Usage: $0 [ -conf conf-file ] [ -pidfile pid-file ] [ -start job-number] [ -debug ]"; $ONE_MINUTE = 60; $ONE_HOUR = $ONE_MINUTE * 60; $ONE_DAY = $ONE_HOUR * 24; $wait_interval = $ONE_MINUTE; $max_wait_interval = $ONE_HOUR; $wait_maximum = $ONE_DAY; $starting_job = 1; $sendmail = "/usr/lib/sendmail"; if ( -f "/usr/sbin/sendmail" ) { $sendmail = "/usr/sbin/sendmail"; } &openlog($0, 'pid', 'local1'); eval "setpriority(&PRIO_PROCESS, 0, 10);"; &syslog('warning', "Couldn't setpriority: %s", $@) if ($@); while (($_ = $ARGV[0]) && /^-/) { shift; if (/^-conf$/) { if (! ($jobfile = shift)) { &die("Missing argument to -conf command-line option.\n$usage\n"); } next; } elsif (/^-pidfile$/) { if (! ($pidfile = shift)) { &die("Missing argument to -pidfile command-line option.\n$usage\n"); } next; } elsif (/^-start$/) { if (! ($starting_job = shift)) { &die("Missing argument to $_ command-line option.\n$usage\n"); } next; } elsif (/^-debug$/) { $debug++; next; } elsif (/^--$/) { last; } else { &die("Unknown command-line option $_.\n$usage\n"); } } &die($usage) if (@ARGV > 0); open(NEWPID, ">$pidfile.$$") || &die("Opening $pidfile.$$ for write: $!"); print NEWPID "$$\n" || &die("Writing to $pidfile.$$: $!"); close(NEWPID) || &die("Closing $pidfile.$$: $!"); while (! link("$pidfile.$$", $pidfile)) { &die("Link to $pidfile.$$ from $pidfile failed: $!") if ($! != &EEXIST); if (! ($oldpid = do $pidfile)) { if (-f $pidfile) { &die("$pidfile is corrupt or unreadable (can't get PID from it)"); } next; } if (! kill(0, $oldpid)) { &die("Couldn't signal $oldpid: $!") if ($! != &ESRCH); &syslog('warning', "Old run (pid %d) appears to have died. Removing %s and continuing.", $oldpid, $pidfile); unlink($pidfile); last; } $important_only++; &die("Waited $wait_total seconds for PID $oldpid to die. Giving up and exiting.") if ($wait_total > $wait_maximum); &syslog('notice', "Old run (pid %d) is still running. Sleeping %d seconds.", $oldpid, $wait_interval); sleep $wait_interval; $wait_total += $wait_interval; $wait_interval *= 2 if ($wait_interval < $max_wait_interval); } unlink("$pidfile.$$"); $we_own_the_lock++; open(JOBS, $jobfile) || &die("Can't open $jobfile: $!"); while (&next_job()) { unlink($tf); if ($job_number < $starting_job) { &syslog('info', "Skipping job %d (%s) because it's lower than requested starting job %d", $job_number, $name, $starting_job); if ($debug) { print STDERR "Skipping job $job_number ($name) because it's lower than requested starting job $starting_job\n"; } next; } if ($important_only && !$important) { &syslog('notice', "Skipping unimportant job %d (%s)", $job_number, $name); next; } &syslog('info', "Starting job %d (%s)", $job_number, $name); open(SHELL, "|$shell > $tf 2>&1") || &die("Running $shell > $tf: $!"); print SHELL $job_text || &die("Writing to $shell: $!"); close(SHELL); $status = $? >> 8; $output = (! -z $tf); if ($status || $output) { open(MAIL, "| $sendmail -t") || &die("Running $sendmail: $!"); if ($debug) { print STDERR "Job $job_number ($name) exited with status $status.\n" if ($status); if ($output) { print STDERR "Job $job_number ($name) had output, starting with:\n"; open(HEAD, "head -10 $tf|"); for (1..10) { $_ = ; print "\t$_"; } close(HEAD); } } eval { print(MAIL "To: $owner\n", "Subject: Report on your nightly job", $name ? " ($name)\n\n" : "\n\n") || die; print MAIL "Report on your job:\n\n$job_text\n" || die; if ($status) { print MAIL "The job exited with status $status.\n\n" || die; } if ($output) { print MAIL "The job produced the following output:\n\n" || die; open(OUTPUT, $tf) || &die("Opening $tf: $!"); while () { print MAIL $_ || die; } close(OUTPUT); } }; &die("Writing to $sendmail: $!") if ($@); close(MAIL); &die("$sendmail exited with non-zero status (job number $job_number ($name))") if ($?); } &syslog('info', "Finished job %d (%s)", $job_number, $name); } close(JOBS); unlink($pidfile, $tf); sub next_job { $owner = "root"; $job_text = ""; $name = undef; $job_number++; $prefix = "/\^/"; $terminator = "/\^\$/"; $shell = "/bin/sh"; $important = 0; local($_); while () { next if (/^\#/ || /^\s*$/); chop; eval; &die("Error in Perl configuration line $_: $@") if ($@); last; } while () { $ret = eval $terminator; &die("Error evaluating terminator \"$terminator\" for job $job_number ($name): $@") if ($@); if ($ret) { last; } $ret = eval $prefix; &die("Error evaluating prefix \"$prefix\" for job $job_number ($name): $@") if ($@); if ($ret) { $job_text .= $_; next; } } if ($debug && $job_text) { local($text) = $job_text; local($*) = 1; $text =~ s/^/\t/g; print STDERR <