12032: Controller support for group trash.

[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..230f03e5777a56aec28b57bda56dcea4e82b065f 100644 (file)
--- a/services/api/lib/crunch_dispatch.rb
+++ b/services/api/lib/crunch_dispatch.rb
@@ -1,3 +1,7 @@
+# Copyright (C) The Arvados Authors. All rights reserved.
+#
+# SPDX-License-Identifier: AGPL-3.0
+
 require 'open3'
 require 'shellwords'
 
@@ -684,17 +688,20 @@ 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
@@ -889,7 +896,7 @@ class CrunchDispatch
 
   def check_orphaned_slurm_jobs
     act_as_system_user do
-      squeue_uuids = squeue_jobs.select{|uuid| uuid.match(HasUuid::UUID_REGEX)}.
+      squeue_uuids = squeue_jobs.select{|uuid| uuid.match(/^[0-9a-z]{5}-8i9sb-[0-9a-z]{15}$/)}.
                                   select{|uuid| !@running.has_key?(uuid)}
 
       return if squeue_uuids.size == 0
@@ -963,8 +970,11 @@ 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 +983,9 @@ 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