X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/3687df5d70eb3112914b81c002c6fa10af3a1e8a..e46ae4015db8f166359515f2e9823ac1a2c7cab8:/services/api/script/crunch-dispatch.rb

diff --git a/services/api/script/crunch-dispatch.rb b/services/api/script/crunch-dispatch.rb
index a0ba9b1fde..ecc7f5d98e 100755
--- a/services/api/script/crunch-dispatch.rb
+++ b/services/api/script/crunch-dispatch.rb
@@ -2,6 +2,7 @@
 
 include Process
 
+$warned = {}
 $signal = {}
 %w{TERM INT}.each do |sig|
   signame = sig
@@ -39,14 +40,36 @@ class Dispatcher
     @todo = Job.queue
   end
 
+  def sinfo
+    @@slurm_version ||= Gem::Version.new(`sinfo --version`.match(/\b[\d\.]+\b/)[0])
+    if Gem::Version.new('2.3') <= @@slurm_version
+      `sinfo --noheader -o '%n:%t'`.strip
+    else
+      # Expand rows with hostname ranges (like "foo[1-3,5,9-12]:idle")
+      # into multiple rows with one hostname each.
+      `sinfo --noheader -o '%N:%t'`.split("\n").collect do |line|
+        tokens = line.split ":"
+        if (re = tokens[0].match /^(.*?)\[([-,\d]+)\]$/)
+          re[2].split(",").collect do |range|
+            range = range.split("-").collect(&:to_i)
+            (range[0]..range[-1]).collect do |n|
+              [re[1] + n.to_s, tokens[1..-1]].join ":"
+            end
+          end
+        else
+          tokens.join ":"
+        end
+      end.flatten.join "\n"
+    end
+  end
+
   def update_node_status
     if Server::Application.config.crunch_job_wrapper.to_s.match /^slurm/
       @nodes_in_state = {idle: 0, alloc: 0, down: 0}
       @node_state ||= {}
       node_seen = {}
       begin
-        `sinfo --noheader -o '%n:%t'`.
-          split("\n").
+        sinfo.split("\n").
           each do |line|
           re = line.match /(\S+?):+(idle|alloc|down)/
           next if !re
@@ -121,16 +144,17 @@ class Dispatcher
             api_client_id: 0)
       job_auth.save
 
-      cmd_args << (ENV['CRUNCH_JOB_BIN'] || `which crunch-job`.strip)
+      crunch_job_bin = (ENV['CRUNCH_JOB_BIN'] || `which arv-crunch-job`.strip)
+      if crunch_job_bin == ''
+        raise "No CRUNCH_JOB_BIN env var, and crunch-job not in path."
+      end
+
+      cmd_args << crunch_job_bin
       cmd_args << '--job-api-token'
       cmd_args << job_auth.api_token
       cmd_args << '--job'
       cmd_args << job.uuid
 
-      if cmd_args[0] == ''
-        raise "No CRUNCH_JOB_BIN env var, and crunch-job not in path."
-      end
-
       commit = Commit.where(sha1: job.script_version).first
       if commit
         cmd_args << '--git-dir'
@@ -163,6 +187,7 @@ class Dispatcher
       $stderr.puts start_banner
       $redis.set job.uuid, start_banner + "\n"
       $redis.publish job.uuid, start_banner
+      $redis.publish job.owner_uuid, start_banner
 
       @running[job.uuid] = {
         stdin: i,
@@ -218,8 +243,10 @@ class Dispatcher
           lines.each do |line|
             $stderr.print "#{job_uuid} ! " unless line.index(job_uuid)
             $stderr.puts line
-            $redis.publish job_uuid, "#{Time.now.ctime.to_s} #{line.strip}"
-            $redis.append job_uuid, "#{Time.now.ctime.to_s} #{line}"
+            pub_msg = "#{Time.now.ctime.to_s} #{line.strip}"
+            $redis.publish job.owner_uuid, pub_msg
+            $redis.publish job_uuid, pub_msg
+            $redis.append job_uuid, pub_msg + "\n"
             if LOG_BUFFER_SIZE < $redis.strlen(job_uuid)
               $redis.set(job_uuid,
                          $redis
@@ -308,7 +335,9 @@ class Dispatcher
       else
         refresh_todo unless did_recently(:refresh_todo, 1.0)
         update_node_status
-        start_jobs unless @todo.empty? or did_recently(:start_jobs, 1.0)
+        unless @todo.empty? or did_recently(:start_jobs, 1.0) or $signal[:term]
+          start_jobs
+        end
       end
       reap_children
       select(@running.values.collect { |j| [j[:stdout], j[:stderr]] }.flatten,
@@ -329,4 +358,7 @@ class Dispatcher
   end
 end
 
+# This is how crunch-job child procs know where the "refresh" trigger file is
+ENV["CRUNCH_REFRESH_TRIGGER"] = Rails.configuration.crunch_refresh_trigger
+
 Dispatcher.new.run