Merge branch '11590-log-reuse'

[arvados.git] / services / api / lib / crunch_dispatch.rb

diff --git a/services/api/lib/crunch_dispatch.rb b/services/api/lib/crunch_dispatch.rb
index 21843de67e64cc045c71d73e8085e4ca063f800e..2ae99f01c59b7bef384e4e48011357a37f3363d3 100644 (file)
--- a/services/api/lib/crunch_dispatch.rb
+++ b/services/api/lib/crunch_dispatch.rb
@@ -684,17 +684,20 @@ class CrunchDispatch
     jobrecord = Job.find_by_uuid(job_done.uuid)
 
     if exit_status == EXIT_RETRY_UNLOCKED or (exit_tempfail and @job_retry_counts.include? jobrecord.uuid)
+      $stderr.puts("dispatch: job #{jobrecord.uuid} was interrupted by node failure")
       # Only this crunch-dispatch process can retry the job:
       # it's already locked, and there's no way to put it back in the
       # Queued state.  Put it in our internal todo list unless the job
       # has failed this way excessively.
       @job_retry_counts[jobrecord.uuid] += 1
       exit_tempfail = @job_retry_counts[jobrecord.uuid] <= RETRY_UNLOCKED_LIMIT
+      do_what_next = "give up now"
       if exit_tempfail
         @todo_job_retries[jobrecord.uuid] = jobrecord
-      else
-        $stderr.puts("dispatch: job #{jobrecord.uuid} exceeded node failure retry limit -- giving up")
+        do_what_next = "re-attempt"
       end
+      $stderr.puts("dispatch: job #{jobrecord.uuid} has been interrupted " +
+                   "#{@job_retry_counts[jobrecord.uuid]}x, will #{do_what_next}")
     end
 
     if !exit_tempfail
@@ -963,8 +966,11 @@ class CrunchDispatch
   # An array of job_uuids in squeue
   def squeue_jobs
     if Rails.configuration.crunch_job_wrapper == :slurm_immediate
-      File.popen(['squeue', '-a', '-h', '-o', '%j']).readlines.map do |line|
-        line.strip
+      p = IO.popen(['squeue', '-a', '-h', '-o', '%j'])
+      begin
+        p.readlines.map {|line| line.strip}
+      ensure
+        p.close
       end
     else
       []
@@ -973,7 +979,9 @@ class CrunchDispatch
 
   def scancel slurm_name
     cmd = sudo_preface + ['scancel', '-n', slurm_name]
-    puts File.popen(cmd).read
+    IO.popen(cmd) do |scancel_pipe|
+      puts scancel_pipe.read
+    end
     if not $?.success?
       Rails.logger.error "scancel #{slurm_name.shellescape}: $?"
     end