ping the database and retry rather before doing anything

[freeside.git] / FS / bin / freeside-queued

diff --git a/FS/bin/freeside-queued b/FS/bin/freeside-queued
index 67e5e2b..0be3d9d 100644 (file)
--- a/FS/bin/freeside-queued
+++ b/FS/bin/freeside-queued
@@ -1,21 +1,21 @@
 #!/usr/bin/perl -w
 
 use strict;
-use vars qw( $log_file $sigterm $sigint $kids $max_kids );
+use vars qw( $log_file $sigterm $sigint $kids $max_kids %kids );
 use subs qw( _die _logmsg );
 use Fcntl qw(:flock);
-use POSIX qw(setsid);
+use POSIX qw(:sys_wait_h setsid);
 use Date::Format;
 use IO::File;
-use FS::UID qw(adminsuidsetup forksuidsetup driver_name dbh);
-use FS::Record qw(qsearchs);
+use FS::UID qw(adminsuidsetup forksuidsetup driver_name dbh myconnect);
+use FS::Record qw(qsearch qsearchs);
 use FS::queue;
 use FS::queue_depend;
 
 # no autoloading just yet
 use FS::cust_main;
 use FS::svc_acct;
-use Net::SSH 0.05;
+use Net::SSH 0.07;
 use FS::part_export;
 
 $max_kids = '10'; #guess it should be a config file...
@@ -28,16 +28,28 @@ my $pid_file = "/var/run/freeside-queued.pid";
 
 &daemonize1;
 
-sub REAPER { my $pid = wait; $SIG{CHLD} = \&REAPER; $kids--; }
-$SIG{CHLD} =  \&REAPER;
+#sub REAPER { my $pid = wait; $SIG{CHLD} = \&REAPER; $kids--; }
+#$SIG{CHLD} =  \&REAPER;
 
 $sigterm = 0;
 $sigint = 0;
 $SIG{INT} = sub { warn "SIGINT received; shutting down\n"; $sigint++; };
 $SIG{TERM} = sub { warn "SIGTERM received; shutting down\n"; $sigterm++; };
 
-$> = $FS::UID::freeside_uid unless $>;
-$< = $>;
+my $freeside_gid = scalar(getgrnam('freeside'))
+  or die "can't setgid to freeside group\n";
+$) = $freeside_gid;
+$( = $freeside_gid;
+#if freebsd can't setuid(), presumably it can't setgid() either.  grr fleabsd
+($(,$)) = ($),$();
+$) = $freeside_gid;
+
+$> = $FS::UID::freeside_uid;
+$< = $FS::UID::freeside_uid;
+#freebsd is sofa king broken, won't setuid()
+($<,$>) = ($>,$<);
+$> = $FS::UID::freeside_uid;
+
 $ENV{HOME} = (getpwuid($>))[7]; #for ssh
 adminsuidsetup $user;
 
@@ -53,52 +65,67 @@ warn "freeside-queued starting\n";
 my $warnkids=0;
 while (1) {
 
+  &reap_kids;
   #prevent runaway forking
   if ( $kids >= $max_kids ) {
     warn "WARNING: maximum $kids children reached\n" unless $warnkids++;
+    &reap_kids;
     sleep 1; #waiting for signals is cheap
     next;
   }
   $warnkids=0;
 
-  my $nodepend = driver name eq 'mysql'
+  my $dbh = dbh;
+  unless ( $dbh->ping ) {
+    warn "WARNING: connection to database lost, reconnecting...\n";
+    myconnect;
+    unless ( $dbh->ping ) {
+      warn "WARNING: still no connection to database, sleeping for retry...\n";
+      sleep 10;
+      next;
+    }
+  }
+
+  #my($job, $ljob);
+  #{
+  #  my $oldAutoCommit = $FS::UID::AutoCommit;
+  #  local $FS::UID::AutoCommit = 0;
+  $FS::UID::AutoCommit = 0;
+  
+  my $nodepend = driver_name eq 'mysql'
    ? ''
    : 'AND 0 = ( SELECT COUNT(*) FROM queue_depend'.
      ' WHERE queue_depend.jobnum = queue.jobnum ) ';
+  my $job = qsearchs(
+    'queue',
+    { 'status' => 'new' },
+    '',
+    driver_name eq 'mysql'
+      ? "$nodepend ORDER BY jobnum LIMIT 1 FOR UPDATE"
+      : "$nodepend ORDER BY jobnum FOR UPDATE LIMIT 1"
+  ) or do {
+    $dbh->commit or die $dbh->errstr; #if $oldAutoCommit;
+    sleep 5; #connecting to db is expensive
+    next;
+  };
 
-  my($job, $ljob);
-  {
-    my $oldAutoCommit = $FS::UID::AutoCommit;
-    local $FS::UID::AutoCommit = 0;
-    my $dbh = dbh; 
-  
-    $job = qsearchs(
-      'queue',
-      { 'status' => 'new' },
-      '',
-      driver_name eq 'mysql'
-        ? "$nodepend ORDER BY jobnum LIMIT 1 FOR UPDATE"
-        : "$nodepend ORDER BY jobnum FOR UPDATE LIMIT 1"
-    ) or do {
-      $dbh->commit or die $dbh->errstr if $oldAutoCommit;
-      sleep 5; #connecting to db is expensive
-      next;
-    };
+  if ( driver_name eq 'mysql'
+       && qsearch('queue_depend', { 'jobnum' => $job->jobnum } ) ) {
+    $dbh->commit or die $dbh->errstr; #if $oldAutoCommit;
+    sleep 5; #would be better if mysql could do everything in query above
+    next;
+  }
 
-    if ( driver_name eq 'mysql'
-         && qsearch('queue_depend', { 'jobnum' => $job->jobnum } ) ) {
-      $dbh->commit or die $dbh->errstr if $oldAutoCommit;
-      next;
-    }
+  my %hash = $job->hash;
+  $hash{'status'} = 'locked';
+  my $ljob = new FS::queue ( \%hash );
+  my $error = $ljob->replace($job);
+  die $error if $error;
 
-    my %hash = $job->hash;
-    $hash{'status'} = 'locked';
-    $ljob = new FS::queue ( \%hash );
-    my $error = $ljob->replace($job);
-    die $error if $error;
+  $dbh->commit or die $dbh->errstr; #if $oldAutoCommit;
 
-    $dbh->commit or die $dbh->errstr if $oldAutoCommit;
-  } 
+  $FS::UID::AutoCommit = 1;
+  #} 
 
   my @args = $ljob->args;
 
@@ -115,6 +142,7 @@ while (1) {
 
   if ( $pid ) {
     $kids++;
+    $kids{$pid} = 1;
   } else { #kid time
 
     #get new db handle
@@ -214,6 +242,16 @@ sub daemonize2 {
   open STDERR, '>&STDOUT' or die "Can't dup stdout: $!";
 }
 
+sub reap_kids {
+  foreach my $pid ( keys %kids ) {
+    my $kid = waitpid($pid, WNOHANG);
+    if ( $kid > 0 ) {
+      $kids--;
+      delete $kids{$kid};
+    }
+  }
+}
+
 =head1 NAME
 
 freeside-queued - Job queue daemon