From c5f46b8b4325babe2ed18404b55c8d224452c696 Mon Sep 17 00:00:00 2001
From: Ondrej Holecek <oholecek@suse.com>
Date: Thu, 9 Apr 2015 15:23:57 +0200
Subject: [PATCH] properly exit when scheduler misbehaves

- reported by maxlin, worker api-call timed out on inactivity from
scheduler side. This does not solve the issue itself, but at least
worker should not force it through with undefined vars
- also die in timer only kills the timer, not the app itself
---
 lib/OpenQA/Worker/Common.pm | 9 +++++++--
 lib/OpenQA/Worker/Jobs.pm   | 4 ++--
 2 files changed, 9 insertions(+), 4 deletions(-)

diff --git a/lib/OpenQA/Worker/Common.pm b/lib/OpenQA/Worker/Common.pm
index 41a35b4482a..ea4676a2730 100644
--- a/lib/OpenQA/Worker/Common.pm
+++ b/lib/OpenQA/Worker/Common.pm
@@ -172,7 +172,7 @@ sub api_call {
     my ($method, $path, $params, $json_data, $ignore_errors) = @_;
     state $call_running;
 
-    return undef unless verify_workerid();
+    return unless verify_workerid();
 
     if ($call_running) {
         # quit immediately
@@ -293,6 +293,9 @@ sub _get_capabilities {
 }
 
 sub setup_websocket {
+    # no point in trying if we are not registered
+    return unless verify_workerid();
+
     # if there is an existing web socket connection wait until it finishes.
     if ($ws) {
         add_timer('setup_websocket', 2, \&setup_websocket, 1);
@@ -367,7 +370,9 @@ sub register_worker {
     my $tx = $ua->post($ua_url => json => $worker_caps);
     unless ($tx->success && $tx->success->json) {
         if ($tx->error && $tx->error->{code} && $tx->error->{code} =~ /^4\d\d$/) {
-            die sprintf "server refused with code %s: %s\n", $tx->error->{code}, $tx->res->body;
+            # don't retry when 4xx codes are returned. There is problem with scheduler
+            printf "server refused with code %s: %s\n", $tx->error->{code}, $tx->res->body;
+            Mojo::IOLoop->stop;
         }
         print "failed to register worker, retry ...\n" if $verbose;
         add_timer('register_worker', 10, \&register_worker, 1);
diff --git a/lib/OpenQA/Worker/Jobs.pm b/lib/OpenQA/Worker/Jobs.pm
index f2f37070071..ee4d24de40c 100644
--- a/lib/OpenQA/Worker/Jobs.pm
+++ b/lib/OpenQA/Worker/Jobs.pm
@@ -187,7 +187,7 @@ sub _stop_job($;$) {
             $job_done = 1;
         }
     }
-    unless ($job_done) {
+    unless ($job_done || $aborted eq 'api-failure') {
         # set job to done. if priority is less than threshold duplicate it
         # with worse priority so it can be picked up again.
         my %args;
@@ -208,7 +208,7 @@ sub _stop_job($;$) {
         return;
     }
     # immediatelly check for already scheduled job
-    add_timer('check_job', 0, \&check_job, 1) unless ($job);
+    add_timer('check_job', 0, \&check_job, 1);
 }
 
 sub start_job {