recover better from mysql's deadlock stupidity, grr, RT#10574

[freeside.git] / FS / bin / freeside-torrus-srvderive

diff --git a/FS/bin/freeside-torrus-srvderive b/FS/bin/freeside-torrus-srvderive
index 6f53cdb..0441c47 100644 (file)
--- a/FS/bin/freeside-torrus-srvderive
+++ b/FS/bin/freeside-torrus-srvderive
@@ -8,7 +8,7 @@ use FS::UID qw( adminsuidsetup dbh driver_name );
 use FS::Record qw( qsearch str2time_sql str2time_sql_closing concat_sql );
 use FS::torrus_srvderive;
 
-our $DEBUG = 0;
+our $DEBUG = 2;
 
 my $user = shift or die &usage;
 $FS::Daemon::PID_NEWSTYLE = 1;
@@ -48,12 +48,17 @@ my $sql = "
     WHERE NOT EXISTS (
                        SELECT 1 FROM srvexport AS other
                          WHERE other.serviceid IN ( $in, $out )
+                           AND srvexport.srv_date = other.srv_date
                            AND ABS( $_date - $other_date ) <= 60
                      )
+";
+
+my $orderlimit = "
     ORDER BY srv_date, srv_time
-    LIMIT 100
+    LIMIT 50
 ";
 
+
 while (1) {
 
   my $found = 0;
@@ -62,22 +67,46 @@ while (1) {
 
     my $serviceid = $torrus_srvderive->serviceid;
 
-    warn $sql if $DEBUG > 2;
-    my $sth = dbh->prepare($sql) or die $DBI::errstr; #better recovery?
+    my @serviceids = $torrus_srvderive->component_serviceids;
+
+    my @in = ();
+    for my $dir ('_IN', '_OUT') {
+      push @in, map dbh->quote("$_$dir"), @serviceids;
+    }
+    my $in = join(',', @in);
+
+    my $ssql = "
+      $sql AND EXISTS (
+                       SELECT 1 FROM srvexport AS other
+                         WHERE other.serviceid IN ($in)
+                           AND srvexport.srv_date = other.srv_date
+                           AND ABS( $_date - $other_date ) <= 60
+                     )
+    ";
+
+    $ssql .= " AND srv_date >= '". $torrus_srvderive->last_srv_date. "' "
+      if $torrus_srvderive->last_srv_date;
+
+    $ssql .= $orderlimit;
+
+    warn "searching for times to add $serviceid\n" if $DEBUG;
+    warn $ssql if $DEBUG > 2;
+    my $sth = dbh->prepare($ssql) or die $DBI::errstr; #better recovery?
     $sth->execute($serviceid, $serviceid) or die $sth->errstr;
+
     my $prev = 0;
     while ( my $row = $sth->fetchrow_arrayref ) {
+      last if sigterm() || sigint();
+
       my( $srv_date, $srv_time ) = @$row;
       my $cur = str2time( "$srv_date $srv_time" );
       next if $cur-$prev <= 60;
       last if time - $cur <= 300;
-      
+
       warn "no $serviceid for $srv_date $srv_time; adding\n"
         if $DEBUG;
       $found++;
 
-      my @serviceids = $torrus_srvderive->component_serviceids;
-
       for my $dir ('_IN', '_OUT') {
 
         my $sin = join(',', map dbh->quote("$_$dir"), @serviceids);
@@ -98,21 +127,35 @@ while (1) {
                       300, #intvl ... 
                     );
         warn $isql. ' with param '. join(',',@param). "\n"
-          if $DEBUG > 1;
+          if $DEBUG > 2;
 
         my $isth = dbh->prepare($isql) or die $DBI::errstr; #better recovery?
-        $isth->execute( @param )       or die $isth->errstr;
-        dbh->commit                    or die dbh->errstr;
+
+        #stupid mysql deadlocks all the time here, so we need to recover
+        $isth->execute( @param ) or do {
+          warn "Error inserting data for $serviceid$dir (restarting): ".
+               $isth->errstr;
+        }
                         
       }
 
+      if ( $srv_date ne $torrus_srvderive->last_srv_date ) {
+        warn "updating last_srv_date of $serviceid to $srv_date\n" if $DEBUG;
+        $torrus_srvderive->last_srv_date($srv_date);
+        my $error = $torrus_srvderive->replace;
+        die $error if $error;
+      }
+      dbh->commit or die dbh->errstr;
+
       $prev = $cur;
     }
+    warn "done with $serviceid\n" if $DEBUG;
 
   }
-
+  dbh->commit or die dbh->errstr;
 
   myexit() if sigterm() || sigint();
+  warn "restarting main loop\n" if $DEBUG > 1;
   sleep 60 unless $found;
 }