cope with empty virtual ports, RT#10574

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

diff --git a/FS/bin/freeside-torrus-srvderive b/FS/bin/freeside-torrus-srvderive
index 6708cc6..f6fc499 100644 (file)
--- a/FS/bin/freeside-torrus-srvderive
+++ b/FS/bin/freeside-torrus-srvderive
@@ -1,6 +1,7 @@
 #!/usr/bin/perl -w
 
 use strict;
+use Sys::SigAction qw( set_sig_handler );
 use Date::Parse;
 use Date::Format;
 use FS::Daemon ':all'; #daemonize1 drop_root daemonize2 myexit logfile sig*
@@ -8,7 +9,7 @@ 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 = 1;
+our $DEBUG = 2;
 
 my $user = shift or die &usage;
 $FS::Daemon::PID_NEWSTYLE = 1;
@@ -48,36 +49,105 @@ 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 10
+    LIMIT 50
 ";
 
-while (1) {
+
+MAIN: while (1) {
 
   my $found = 0;
 
-  foreach my $torrus_srvderive ( qsearch('torrus_srvderive', {}) ) {
+  SERVICEID: foreach my $torrus_srvderive ( qsearch('torrus_srvderive', {}) ) {
 
     my $serviceid = $torrus_srvderive->serviceid;
 
-    warn $sql if $DEBUG > 2;
-    my $sth = dbh->prepare($sql) or die $DBI::errstr; #better recovery?
-    $sth->execute($serviceid, $serviceid) or die $sth->errstr;
+    my @serviceids = $torrus_srvderive->component_serviceids;
+    next unless @serviceids; #don't try to search for empty virtual ports
+
+    my @in = ();
+    for my $dir ('_IN', '_OUT') {
+      push @in, map dbh->quote("$_$dir"), @serviceids;
+    }
+    my $in = join(',', @in);
+
+    if ( ! $torrus_srvderive->last_srv_date ) {
+      warn "finding initial last_srv_date for $serviceid\n" if $DEBUG;
+      my $dsql = "SELECT srv_date FROM srvexport WHERE serviceid IN ($in)
+                  ORDER BY srv_date LIMIT 1";
+      my $dsth = dbh->prepare($dsql) or die $DBI::errstr;
+      $dsth->execute or die $dsth->errstr;
+      my $date = $dsth->fetchrow_arrayref->[0];
+      if ( $date ) {
+        warn "found initial last_srv_date of $date; updating $serviceid\n"
+          if $DEBUG;
+        $torrus_srvderive->last_srv_date($date);
+        my $error = $torrus_srvderive->replace;
+        die $error if $error;
+      } else {
+        warn "no initial last_srv_date for $serviceid; skipping\n" if $DEBUG;
+        next;
+      }
+    }
+
+    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 here?
+
+    eval {
+      my $h = set_sig_handler( 'ALRM', sub { die "_timeout\n"; } );
+      alarm(10*60); #5*60); #$torrus_srvderive->last_srv_date ? 5*60 : 15*60);
+      $sth->execute($serviceid, $serviceid) or die $sth->errstr;
+      alarm(0);
+    };
+    alarm(0);
+    
+    if ( $@ && $@ eq "_timeout\n" ) {
+      warn "search timed out; reconnecting and restarting\n";
+      dbh->clone()->do("KILL QUERY ". dbh->{"mysql_thread_id"})
+        if driver_name eq 'mysql';
+      dbh->rollback; #or die dbh->errstr;
+      adminsuidsetup($user);
+      next SERVICEID; #MAIN;
+    } elsif ( $@ ) {
+      die $@;
+    }
+
+    warn "search finished; checking results\n" if $DEBUG;
+
     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 +168,38 @@ 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 on insert, so we need to recover
+        unless ( $isth->execute(@param) ) {
+          warn "Error inserting data for $serviceid$dir (restarting): ".
+               $isth->errstr;
+          dbh->rollback; #or die dbh->errstr;
+          sleep 5;
+          next SERVICEID; #MAIN;
+        }
                         
       }
 
+      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;
 
-  }
-
+  } #foreach my $torrus_srvderive
+  dbh->commit or die dbh->errstr;
 
   myexit() if sigterm() || sigint();
+  warn "restarting main loop\n" if $DEBUG > 1;
   sleep 60 unless $found;
 }