From 215697368aa04b0d97ae2321ec321d9ca8e2cb91 Mon Sep 17 00:00:00 2001
From: Sven Nierlein <sven@nierlein.de>
Date: Mon, 27 Jun 2022 21:54:37 +0200
Subject: [PATCH] improve waiting for core reload

---
 lib/Thruk/Utils.pm                            | 23 ++++++++++---------
 .../business_process/lib/Thruk/BP/Utils.pm    | 14 +++++++++--
 .../conf/lib/Thruk/Controller/conf.pm         |  2 +-
 3 files changed, 25 insertions(+), 14 deletions(-)

diff --git a/lib/Thruk/Utils.pm b/lib/Thruk/Utils.pm
index 0fe5420c9c..6ae6014707 100644
--- a/lib/Thruk/Utils.pm
+++ b/lib/Thruk/Utils.pm
@@ -2347,19 +2347,20 @@ sub restart_later {
 
   wait_after_reload($c, [$backend], [$timestamp])
 
-wait up to 60 seconds till the core responds
+wait up to 30 seconds till the core responds
 
 =cut
 
 sub wait_after_reload {
-    my($c, $pkey, $time) = @_;
-    $c->stats->profile(begin => "wait_after_reload ($time)");
+    my($c, $pkey, $last_reload) = @_;
+    $c->stats->profile(begin => "wait_after_reload");
     $pkey = $c->stash->{'param_backend'} unless $pkey;
     my $start = time();
-    if(!$pkey && !$time) {
-        _debug('no peer key and time, waiting 3 seconds');
+    if(!$pkey) {
+        _debug('no peer key, waiting 3 seconds');
         sleep 3;
     }
+    $last_reload = time() unless $last_reload;
 
     # wait until core responds again
     my $procinfo = {};
@@ -2370,7 +2371,7 @@ sub wait_after_reload {
                 'header' => {
                     'WaitTimeout'   => 2000,
                     'WaitTrigger'   => 'all', # using something else seems not to work all the time
-                    'WaitCondition' => "program_start > ".$time,
+                    'WaitCondition' => "program_start > ".$last_reload,
                 },
         };
     }
@@ -2394,11 +2395,11 @@ sub wait_after_reload {
             $msg = 'still waiting for core reload for '.(time()-$start).'s: '.$c->stash->{'failed_backends'}->{$pkey};
             _debug($msg);
         }
-        elsif($pkey and $time) {
+        elsif($pkey and $last_reload) {
             # not yet restarted
             if($procinfo and $procinfo->{$pkey} and $procinfo->{$pkey}->{'program_start'}) {
                 $c->stats->profile(comment => "core program_start: ".$procinfo->{$pkey}->{'program_start'});
-                if($procinfo->{$pkey}->{'program_start'} > $time) {
+                if($procinfo->{$pkey}->{'program_start'} > $last_reload) {
                     $done = 1;
                     _debug('core reloaded after '.(time()-$start).'s, last program_start: '.(scalar localtime($procinfo->{$pkey}->{'program_start'})));
                     last;
@@ -2408,14 +2409,14 @@ sub wait_after_reload {
                 }
             }
         }
-        elsif($time) {
+        elsif($last_reload) {
             my $newest_core = 0;
             if($procinfo) {
                 for my $key (keys %{$procinfo}) {
                     if($procinfo->{$key}->{'program_start'} > $newest_core) { $newest_core = $procinfo->{$key}->{'program_start'}; }
                 }
                 $c->stats->profile(comment => "core program_start: ".$newest_core);
-                if($newest_core > $time) {
+                if($newest_core > $last_reload) {
                     $done = 1;
                     last;
                 } else {
@@ -2433,7 +2434,7 @@ sub wait_after_reload {
             sleep(1);
         }
     }
-    $c->stats->profile(end => "wait_after_reload ($time)");
+    $c->stats->profile(end => "wait_after_reload");
     if($done) {
         # clean up cached groups which may have changed
         $c->cache->clear();
diff --git a/plugins/plugins-available/business_process/lib/Thruk/BP/Utils.pm b/plugins/plugins-available/business_process/lib/Thruk/BP/Utils.pm
index 652b0c8515..2786960489 100644
--- a/plugins/plugins-available/business_process/lib/Thruk/BP/Utils.pm
+++ b/plugins/plugins-available/business_process/lib/Thruk/BP/Utils.pm
@@ -303,7 +303,6 @@ sub save_bp_objects {
         }
 
         # and reload
-        my $time = time();
         my $pkey;
         if($result_backend) {
             my $peer = $c->db->get_peer_by_key($result_backend);
@@ -317,6 +316,17 @@ sub save_bp_objects {
                 }
             }
         }
+
+        my $last_reload = time() - 1;
+        if($pkey) {
+            $last_reload = $c->stash->{'pi_detail'}->{$pkey}->{'program_start'};
+            if(!$last_reload) {
+                my $processinfo = $c->db->get_processinfo(backends => $pkey);
+                $last_reload = ($processinfo->{$pkey} && $processinfo->{$pkey}->{'program_start'}) || (time() - 1);
+            }
+            sleep(1) if $last_reload == time();
+        }
+
         my $cmd = $c->config->{'Thruk::Plugin::BP'}->{'objects_reload_cmd'};
         my $reloaded = 0;
         if($cmd) {
@@ -335,7 +345,7 @@ sub save_bp_objects {
             $reloaded = 1;
         }
         if($rc == 0 && $reloaded) {
-            my $core_reloaded = Thruk::Utils::wait_after_reload($c, $pkey, $time-1);
+            my $core_reloaded = Thruk::Utils::wait_after_reload($c, $pkey, $last_reload);
             if(!$core_reloaded) {
                 ($rc, $msg) = (1, 'business process saved but core failed to restart');
             }
diff --git a/plugins/plugins-available/conf/lib/Thruk/Controller/conf.pm b/plugins/plugins-available/conf/lib/Thruk/Controller/conf.pm
index af85f9a038..59f457cdeb 100644
--- a/plugins/plugins-available/conf/lib/Thruk/Controller/conf.pm
+++ b/plugins/plugins-available/conf/lib/Thruk/Controller/conf.pm
@@ -2737,9 +2737,9 @@ sub _config_reload {
     if(!$last_reload) {
         my $processinfo = $c->db->get_processinfo(backends => $pkey);
         $last_reload = ($processinfo->{$pkey} && $processinfo->{$pkey}->{'program_start'}) || (time() - 1);
+        sleep(1) if $last_reload == time();
     }
 
-    $c->stats->profile(comment => "program_start before reload: ".$last_reload);
     if($c->stash->{'peer_conftool'}->{'obj_reload_cmd'}) {
         if($c->{'obj_db'}->is_remote() && $c->{'obj_db'}->remote_config_reload($c)) {
             Thruk::Utils::set_message( $c, 'success_message', 'config reloaded successfully' );