Merge branch '1968-monitor-disk-usage'
[arvados.git] / services / api / script / crunch-dispatch.rb
index 109fe74389579bb8b71502f443f24460afa9757c..43a527afac98c42285d507aded21dd1bf958146e 100755 (executable)
@@ -11,10 +11,6 @@ $signal = {}
     $signal[:term] = true
   end
 end
-Signal.trap('HUP') do
-  $stderr.puts "Received HUP signal"
-  $signal[:hup] = true
-end
 
 if ENV["CRUNCH_DISPATCH_LOCKFILE"]
   lockfilename = ENV.delete "CRUNCH_DISPATCH_LOCKFILE"
@@ -41,7 +37,31 @@ class Dispatcher
   end
 
   def refresh_todo
-    @todo = Job.queue
+    @todo = Job.queue.select do |j| j.repository end
+    @todo_pipelines = PipelineInstance.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
@@ -50,8 +70,7 @@ class Dispatcher
       @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
@@ -126,32 +145,38 @@ class Dispatcher
             api_client_id: 0)
       job_auth.save
 
-      cmd_args << (ENV['CRUNCH_JOB_BIN'] || `which crunch-job`.strip)
-      cmd_args << '--job-api-token'
-      cmd_args << job_auth.api_token
-      cmd_args << '--job'
-      cmd_args << job.uuid
-
-      if cmd_args[0] == ''
+      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
 
-      commit = Commit.where(sha1: job.script_version).first
-      if commit
-        cmd_args << '--git-dir'
-        if File.exists?(File.
-                        join(Rails.configuration.git_repositories_dir,
-                             commit.repository_name + '.git'))
-          cmd_args << File.
-            join(Rails.configuration.git_repositories_dir,
-                 commit.repository_name + '.git')
-        else
-          cmd_args << File.
-            join(Rails.configuration.git_repositories_dir,
-                 commit.repository_name, '.git')
-        end
+      require 'shellwords'
+
+      arvados_internal = Rails.configuration.git_internal_dir
+      if not File.exists? arvados_internal
+        $stderr.puts `mkdir -p #{arvados_internal.shellescape} && cd #{arvados_internal.shellescape} && git init --bare`
       end
 
+      src_repo = File.join(Rails.configuration.git_repositories_dir, job.repository + '.git')
+      src_repo = File.join(Rails.configuration.git_repositories_dir, job.repository, '.git') unless File.exists? src_repo
+
+      unless src_repo
+        $stderr.puts "dispatch: #{File.join Rails.configuration.git_repositories_dir, job.repository} doesn't exist"
+        sleep 1
+        untake(job)
+        next
+      end
+
+      $stderr.puts `cd #{arvados_internal.shellescape} && git fetch --no-tags #{src_repo.shellescape} && git tag #{job.uuid.shellescape} #{job.script_version.shellescape}`
+
+      cmd_args << crunch_job_bin
+      cmd_args << '--job-api-token'
+      cmd_args << job_auth.api_token
+      cmd_args << '--job'
+      cmd_args << job.uuid
+      cmd_args << '--git-dir'
+      cmd_args << arvados_internal
+
       $stderr.puts "dispatch: #{cmd_args.join ' '}"
 
       begin
@@ -278,7 +303,6 @@ class Dispatcher
     job_done = j_done[:job]
     $stderr.puts "dispatch: child #{pid_done} exit"
     $stderr.puts "dispatch: job #{job_done.uuid} end"
-    $redis.publish job_done.uuid, "end"
 
     # Ensure every last drop of stdout and stderr is consumed
     read_pipes
@@ -289,15 +313,51 @@ class Dispatcher
     # Wait the thread
     j_done[:wait_thr].value
 
+    jobrecord = Job.find_by_uuid(job_done.uuid)
+    if jobrecord.started_at
+      # Clean up state fields in case crunch-job exited without
+      # putting the job in a suitable "finished" state.
+      jobrecord.running = false
+      jobrecord.finished_at ||= Time.now
+      if jobrecord.success.nil?
+        jobrecord.success = false
+      end
+      jobrecord.save!
+    else
+      # Don't fail the job if crunch-job didn't even get as far as
+      # starting it. If the job failed to run due to an infrastructure
+      # issue with crunch-job or slurm, we want the job to stay in the
+      # queue.
+    end
+
     # Invalidate the per-job auth token
     j_done[:job_auth].update_attributes expires_at: Time.now
 
+    $redis.publish job_done.uuid, "end"
+
     @running.delete job_done.uuid
   end
 
+  def update_pipelines
+    expire_tokens = @pipe_auth_tokens.dup
+    @todo_pipelines.each do |p|
+      pipe_auth = (@pipe_auth_tokens[p.uuid] ||= ApiClientAuthorization.
+                   create(user: User.where('uuid=?', p.modified_by_user_uuid).first,
+                          api_client_id: 0))
+      puts `export ARVADOS_API_TOKEN=#{pipe_auth.api_token} && arv-run-pipeline-instance --run-here --no-wait --instance #{p.uuid}`
+      expire_tokens.delete p.uuid
+    end
+
+    expire_tokens.each do |k, v|
+      v.update_attributes expires_at: Time.now
+      @pipe_auth_tokens.delete k
+    end
+  end
+
   def run
     act_as_system_user
     @running ||= {}
+    @pipe_auth_tokens ||= { }
     $stderr.puts "dispatch: ready"
     while !$signal[:term] or @running.size > 0
       read_pipes
@@ -314,35 +374,14 @@ class Dispatcher
           end
         end
       else
-        if File.exists?(Rails.configuration.crunch_dispatch_hup_trigger)
-          begin
-            File.unlink(Rails.configuration.crunch_dispatch_hup_trigger)
-            $signal[:hup] = true
-          rescue Errno::ENOENT
-            $stderr.puts "Weird, hup_trigger file was deleted by someone else."
-          rescue Errno::EPERM
-            if not $warned[:hup_trigger_perm]
-              $warned[:hup_trigger_perm] = true
-              $stderr.puts "Install problem: I see the hup_trigger file but cannot delete it."
-            end
-          end
-        end
-        if $signal[:hup]
-          # Pass HUP through to all crunch-job processes.
-          @running.each do |uuid, j|
-            begin
-              Process.kill 'HUP', j[:wait_thr].pid
-            rescue Errno::ESRCH
-              # Process ended but hasn't been reaped. Nothing to do.
-            end
-          end
-          $signal.delete :hup
-        end
         refresh_todo unless did_recently(:refresh_todo, 1.0)
         update_node_status
         unless @todo.empty? or did_recently(:start_jobs, 1.0) or $signal[:term]
           start_jobs
         end
+        unless (@todo_pipelines.empty? and @pipe_auth_tokens.empty?) or did_recently(:update_pipelines, 5.0)
+          update_pipelines
+        end
       end
       reap_children
       select(@running.values.collect { |j| [j[:stdout], j[:stderr]] }.flatten,
@@ -363,4 +402,7 @@ 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