longer timeout for slaves and separate cache files per slave for improved scaling -- Phillip Moore

4 files changed, 23 insertions, 23 deletions

diff --git a/CHANGES b/CHANGES
index 4d2c27b..d0ed715 100644
--- a/CHANGES
+++ b/CHANGES
@@ -1,3 +1,6 @@
+* use separate slave cache and longer slave timeout. this
+  helps with slave congestion -- Phillip Moore  pdm yahoo-inc.com
+
 * add an optional --pid-dir option to specify the pid directory 
   when running as a slave. --niko
 * slaves were not polling targets where nomasterpoll was set --tobi
diff --git a/lib/Smokeping.pm b/lib/Smokeping.pm
index c7726ca..b3e37b9 100644
--- a/lib/Smokeping.pm
+++ b/lib/Smokeping.pm
@@ -1849,22 +1849,15 @@ sub update_rrds($$$$$$) {
         my $probeobj = $probes->{$probe};
         my $pings = $probeobj->_pings($tree);
         if ($prop eq 'host' and check_filter($cfg,$name) and $tree->{$prop} !~ m|^/|) { # skip multihost
-            my %slave_test;
-            my $slaveupdates;
             my @updates;
             if (not $tree->{nomasterpoll} or $tree->{nomasterpoll} eq 'no'){
                 @updates = ([ "", time, $probeobj->rrdupdate_string($tree) ]);
             }
             if ($tree->{slaves}){
-                %slave_test = ( map { $_,1 } split(/\s+/, $tree->{slaves}));
-                $slaveupdates = Smokeping::Master::get_slaveupdates($name);     
-                for my $slave (@{$slaveupdates}){
-                    if (not $slave_test{$slave->[0]}){
-                        warn "WARNING: skipping update for $slave->[0] since it is not configured for $name\n";
-                        next;
-                    }
-                    push @updates, $slave;
-                }
+                my @slaves = split(/\s+/, $tree->{slaves});
+                foreach my $slave (@slaves) {
+                    push @updates, Smokeping::Master::get_slaveupdates($name, $slave);
+                } #foreach my $checkslave
             }
             for my $update (sort {$a->[1] <=> $b->[1]}  @updates){ # make sure we put the updates in chronological order in
                 my $s = $update->[0] ? "~".$update->[0] : "";
diff --git a/lib/Smokeping/Master.pm b/lib/Smokeping/Master.pm
index 8560de3..d31c95b 100644
--- a/lib/Smokeping/Master.pm
+++ b/lib/Smokeping/Master.pm
@@ -97,16 +97,18 @@ sub save_updates {
     my $updates = shift;
     # name\ttime\tupdatestring
     # name\ttime\tupdatestring
+
     for my $update (split /\n/, $updates){
         my ($name, $time, $updatestring) = split /\t/, $update;
-        my $file = $cfg->{General}{datadir}."/${name}.slave_cache";
+        my $file = $cfg->{General}{datadir}."/${name}.${slave}.slave_cache";
         if ( ${name} =~ m{(^|/)\.\.($|/)} ){
-            warn "Skipping update for ${name}.slave_cache since ".
+            warn "Skipping update for ${name}.${slave}.slave_cache since ".
                  "you seem to try todo some directory magic here. Don't!";
         } 
         elsif ( open (my $lock, '>>' , "$file.lock") ) {
-            for (my $i = 10; $i > 0; $i--){
-                if ( flock $lock, LOCK_EX ){
+
+            for (my $i = 3; $i > 0; $i--){
+                if ( flock($lock, LOCK_EX) ){
                     my $existing = [];
 	            if ( -r $file ){
                         my $in = eval { retrieve $file };
@@ -116,12 +118,12 @@ sub save_updates {
 		            $existing = $in;
 			};
                     };
-                    push @{$existing}, [ $slave, $time, $updatestring];
+                    push @{$existing}, [ $slave, $time, $updatestring ];
                     nstore($existing, $file);		    
                     last;
                 } else {
-                    warn "Could not lock $file. Trying again for $i rounds.\n";
-                    sleep rand(3);
+                    warn "Could not lock $file ($!). Trying again for $i rounds.\n";
+                    sleep rand(2);
                 }
             }
             close $lock;
@@ -133,13 +135,15 @@ sub save_updates {
 
 =head3 get_slaveupdates
 
-Read in all updates provided by slaves and return an array reference.
+Read in all updates provided by the selected slave and return an array reference.
 
 =cut
 
 sub get_slaveupdates {
     my $name = shift;
-    my $file = $name.".slave_cache";
+    my $slave = shift;
+    my $file = $name . "." . $slave. ".slave_cache";
+    my @empty = ();
     my $data;
     if ( -r $file and open (my $lock, '>>', "$file.lock") ) {
         if ( flock $lock, LOCK_EX ){
@@ -147,7 +151,7 @@ sub get_slaveupdates {
             unlink $file;
             if ($@) { #error
                 warn "Loading $file: $@";  
-                return undef;
+                return @empty;
             }
         } else {
             warn "Could not lock $file. Will skip and try again in the next round. No harm done!\n";
@@ -155,7 +159,7 @@ sub get_slaveupdates {
         close $lock;        
         return $data;
     }
-    return;
+    return @empty;
 }
 
 
diff --git a/lib/Smokeping/Slave.pm b/lib/Smokeping/Slave.pm
index 6214817..0747860 100644
--- a/lib/Smokeping/Slave.pm
+++ b/lib/Smokeping/Slave.pm
@@ -72,7 +72,7 @@ sub submit_results {
     my $data_dump = join("\n",@{$restore}) || "";
     my $ua = LWP::UserAgent->new(
         agent => 'smokeping-slave/1.0',
-        timeout => 10,
+        timeout => 60,
         env_proxy => 1 );
 
     my $response = $ua->post(