Merge branch 'master' into 2681-new-inactive-user-notification
[arvados.git] / sdk / cli / bin / arv-run-pipeline-instance
index 1e62a52c04dcf9edbd9106a42d9c2124d07967e3..fc636dff507213ac09bdfc2a9c90c4d7fd62a126 100755 (executable)
 # [--no-wait] Make only as much progress as possible without entering
 #             a sleep/poll loop.
 #
-# [--no-reuse-finished] Do not reuse existing outputs to satisfy
-#                       pipeline components. Always submit a new job
-#                       or use an existing job which has not yet
-#                       finished.
-#
 # [--no-reuse] Do not reuse existing jobs to satisfy pipeline
 #              components. Submit a new job for every component.
 #
@@ -79,6 +74,7 @@ $arvados_api_token = ENV['ARVADOS_API_TOKEN'] or
   abort "#{$0}: fatal: ARVADOS_API_TOKEN environment variable not set."
 
 begin
+  require 'arvados'
   require 'rubygems'
   require 'json'
   require 'pp'
@@ -89,7 +85,7 @@ rescue LoadError => l
   abort <<-EOS
 #{$0}: fatal: #{l.message}
 Some runtime dependencies may be missing.
-Try: gem install pp google-api-client json trollop
+Try: gem install arvados pp google-api-client json trollop
   EOS
 end
 
@@ -152,10 +148,6 @@ p = Trollop::Parser.new do
       "Do not wait for jobs to finish. Just look up status, submit new jobs if needed, and exit.",
       :short => :none,
       :type => :boolean)
-  opt(:no_reuse_finished,
-      "Do not reuse existing outputs to satisfy pipeline components. Always submit a new job or use an existing job which has not yet finished.",
-      :short => :none,
-      :type => :boolean)
   opt(:no_reuse,
       "Do not reuse existing jobs to satisfy pipeline components. Submit a new job for every component.",
       :short => :none,
@@ -225,6 +217,7 @@ $client ||= Google::APIClient.
       :application_name => File.split($0).last,
       :application_version => $application_version.to_s)
 $arvados = $client.discovered_api('arvados', $arvados_api_version)
+$arv = Arvados.new api_version: 'v1'
 
 
 class PipelineInstance
@@ -233,10 +226,10 @@ class PipelineInstance
                              :parameters => {
                                :uuid => uuid
                              },
-                             :body => {
-                               :api_token => ENV['ARVADOS_API_TOKEN']
-                             },
-                             :authenticated => false)
+                             :authenticated => false,
+                             :headers => {
+                               authorization: 'OAuth2 '+ENV['ARVADOS_API_TOKEN']
+                             })
     j = JSON.parse result.body, :symbolize_names => true
     unless j.is_a? Hash and j[:uuid]
       debuglog "Failed to get pipeline_instance: #{j[:errors] rescue nil}", 0
@@ -249,10 +242,12 @@ class PipelineInstance
   def self.create(attributes)
     result = $client.execute(:api_method => $arvados.pipeline_instances.create,
                              :body => {
-                               :api_token => ENV['ARVADOS_API_TOKEN'],
                                :pipeline_instance => attributes
                              },
-                             :authenticated => false)
+                             :authenticated => false,
+                             :headers => {
+                               authorization: 'OAuth2 '+ENV['ARVADOS_API_TOKEN']
+                             })
     j = JSON.parse result.body, :symbolize_names => true
     unless j.is_a? Hash and j[:uuid]
       abort "Failed to create pipeline_instance: #{j[:errors] rescue nil} #{j.inspect}"
@@ -266,10 +261,12 @@ class PipelineInstance
                                :uuid => @pi[:uuid]
                              },
                              :body => {
-                               :api_token => ENV['ARVADOS_API_TOKEN'],
                                :pipeline_instance => @attributes_to_update.to_json
                              },
-                             :authenticated => false)
+                             :authenticated => false,
+                             :headers => {
+                               authorization: 'OAuth2 '+ENV['ARVADOS_API_TOKEN']
+                             })
     j = JSON.parse result.body, :symbolize_names => true
     unless j.is_a? Hash and j[:uuid]
       debuglog "Failed to save pipeline_instance: #{j[:errors] rescue nil}", 0
@@ -298,20 +295,24 @@ class JobCache
     @cache ||= {}
     result = $client.execute(:api_method => $arvados.jobs.get,
                              :parameters => {
-                               :api_token => ENV['ARVADOS_API_TOKEN'],
                                :uuid => uuid
                              },
-                             :authenticated => false)
+                             :authenticated => false,
+                             :headers => {
+                               authorization: 'OAuth2 '+ENV['ARVADOS_API_TOKEN']
+                             })
     @cache[uuid] = JSON.parse result.body, :symbolize_names => true
   end
   def self.where(conditions)
     result = $client.execute(:api_method => $arvados.jobs.list,
                              :parameters => {
-                               :api_token => ENV['ARVADOS_API_TOKEN'],
                                :limit => 10000,
                                :where => conditions.to_json
                              },
-                             :authenticated => false)
+                             :authenticated => false,
+                             :headers => {
+                               authorization: 'OAuth2 '+ENV['ARVADOS_API_TOKEN']
+                             })
     list = JSON.parse result.body, :symbolize_names => true
     if list and list[:items].is_a? Array
       list[:items]
@@ -319,19 +320,21 @@ class JobCache
       []
     end
   end
-  def self.create(attributes)
+  def self.create(job, create_params)
     @cache ||= {}
     result = $client.execute(:api_method => $arvados.jobs.create,
-                             :parameters => {
-                               :api_token => ENV['ARVADOS_API_TOKEN'],
-                               :job => attributes.to_json
-                             },
-                             :authenticated => false)
+                             :body => {
+                               :job => job.to_json
+                             }.merge(create_params),
+                             :authenticated => false,
+                             :headers => {
+                               authorization: 'OAuth2 '+ENV['ARVADOS_API_TOKEN']
+                             })
     j = JSON.parse result.body, :symbolize_names => true
     if j.is_a? Hash and j[:uuid]
       @cache[j[:uuid]] = j
     else
-      debuglog "create job: #{j[:errors] rescue nil}", 0
+      debuglog "create job: #{j[:errors] rescue nil} with attributes #{job}", 0
       nil
     end
   end
@@ -355,10 +358,12 @@ class WhRunPipelineInstance
     else
       result = $client.execute(:api_method => $arvados.pipeline_templates.get,
                                :parameters => {
-                                 :api_token => ENV['ARVADOS_API_TOKEN'],
                                  :uuid => template
                                },
-                               :authenticated => false)
+                               :authenticated => false,
+                               :headers => {
+                                 authorization: 'OAuth2 '+ENV['ARVADOS_API_TOKEN']
+                               })
       @template = JSON.parse result.body, :symbolize_names => true
       if !@template[:uuid]
         abort "#{$0}: fatal: failed to retrieve pipeline template #{template} #{@template[:errors].inspect rescue nil}"
@@ -420,97 +425,77 @@ class WhRunPipelineInstance
   end
 
   def setup_instance
-    @instance ||= PipelineInstance.
-      create(:components => @components,
+    if $options[:submit]
+      @instance ||= PipelineInstance.
+        create(:components => @components,
+              :pipeline_template_uuid => @template[:uuid],
+              :state => 'New')
+    else
+      @instance ||= PipelineInstance.
+        create(:components => @components,
              :pipeline_template_uuid => @template[:uuid],
-             :active => true)
+             :state => 'RunningOnClient')
+    end
     self
   end
 
   def run
     moretodo = true
+    interrupted = false
+
     while moretodo
       moretodo = false
       @components.each do |cname, c|
         job = nil
+        owner_uuid = @instance[:owner_uuid]
+        # Is the job satisfying this component already known to be
+        # finished? (Already meaning "before we query API server about
+        # the job's current state")
+        c_already_finished = (c[:job] &&
+                              c[:job][:uuid] &&
+                              !c[:job][:success].nil?)
         if !c[:job] and
-            c[:script_parameters].select { |pname, p| p.is_a? Hash }.empty?
-          # Job is fully specified (all parameter values are present) but
-          # no particular job has been found.
-
-          debuglog "component #{cname} ready to satisfy."
-
-          c.delete :wait
-          second_place_job = nil # satisfies component, but not finished yet
-
-          (@options[:no_reuse] ? [] : JobCache.
-           where(script: c[:script],
-                 script_parameters: c[:script_parameters],
-                 script_version_descends_from: c[:script_version])
-           ).each do |candidate_job|
-            candidate_params_downcase = Hash[candidate_job[:script_parameters].
-                                             map { |k,v| [k.downcase,v] }]
-            c_params_downcase = Hash[c[:script_parameters].
-                                     map { |k,v| [k.downcase,v] }]
-
-            debuglog "component #{cname} considering job #{candidate_job[:uuid]} version #{candidate_job[:script_version]} parameters #{candidate_params_downcase.inspect}", 3
-
-            unless candidate_params_downcase == c_params_downcase
-              next
-            end
-
-            if c[:script_version] !=
-                candidate_job[:script_version][0,c[:script_version].length]
-              debuglog "component #{cname} would be satisfied by job #{candidate_job[:uuid]} if script_version matched.", 2
-              next
-            end
-
-            unless candidate_job[:success] || candidate_job[:running] ||
-                (!candidate_job[:started_at] && !candidate_job[:cancelled_at])
-              debuglog "component #{cname} would be satisfied by job #{candidate_job[:uuid]} if it were running or successful.", 2
-              next
-            end
-
-            if candidate_job[:success]
-              unless @options[:no_reuse_finished]
-                job = candidate_job
-                $stderr.puts "using #{job[:uuid]} (finished at #{job[:finished_at]}) for component #{cname}"
-                c[:job] = job
-              end
-            else
-              second_place_job ||= candidate_job
-            end
-            break
-          end
-          if not c[:job] and second_place_job
-            job = second_place_job
-            $stderr.puts "using #{job[:uuid]} (running since #{job[:started_at]}) for component #{cname}"
+            c[:script_parameters].select { |pname, p| p.is_a? Hash and p[:output_of]}.empty?
+          # No job yet associated with this component and is component inputs
+          # are fully specified (any output_of script_parameters are resolved
+          # to real value)
+          job = JobCache.create({
+            :script => c[:script],
+            :script_parameters => c[:script_parameters],
+            :script_version => c[:script_version],
+            :repository => c[:repository],
+            :nondeterministic => c[:nondeterministic],
+            :output_is_persistent => c[:output_is_persistent] || false,
+            :owner_uuid => owner_uuid,
+            # TODO: Delete the following three attributes when
+            # supporting pre-20140418 API servers is no longer
+            # important. New API servers take these as flags that
+            # control behavior of create, rather than job attributes.
+            :minimum_script_version => c[:minimum_script_version],
+            :exclude_script_versions => c[:exclude_minimum_script_versions],
+            :no_reuse => @options[:no_reuse] || c[:nondeterministic],
+          }, {
+            # This is the right place to put these attributes when
+            # dealing with new API servers.
+            :minimum_script_version => c[:minimum_script_version],
+            :exclude_script_versions => c[:exclude_minimum_script_versions],
+            :find_or_create => !(@options[:no_reuse] || c[:nondeterministic]),
+          })
+          if job
+            debuglog "component #{cname} new job #{job[:uuid]}"
             c[:job] = job
+          else
+            debuglog "component #{cname} new job failed"
           end
-          if not c[:job]
-            debuglog "component #{cname} not satisfied by any existing job."
-            if !@options[:dry_run]
-              debuglog "component #{cname} new job."
-              job = JobCache.create(:script => c[:script],
-                                    :script_parameters => c[:script_parameters],
-                                    :runtime_constraints => c[:runtime_constraints] || {},
-                                    :script_version => c[:script_version] || 'master')
-              if job
-                debuglog "component #{cname} new job #{job[:uuid]}"
-                c[:job] = job
-              else
-                debuglog "component #{cname} new job failed"
-              end
-            end
-          end
-        else
-          c[:wait] = true
         end
+
         if c[:job] and c[:job][:uuid]
           if (c[:job][:running] or
               not (c[:job][:finished_at] or c[:job][:cancelled_at]))
+            # Job is running so update copy of job record
             c[:job] = JobCache.get(c[:job][:uuid])
           end
+
           if c[:job][:success]
             # Populate script_parameters of other components waiting for
             # this job
@@ -523,8 +508,47 @@ class WhRunPipelineInstance
                 end
               end
             end
+            unless c_already_finished
+              # This is my first time discovering that the job
+              # succeeded. (At the top of this loop, I was still
+              # waiting for it to finish.)
+              if c[:output_is_persistent]
+                # I need to make sure a resources/wants link is in
+                # place to protect the output from garbage
+                # collection. (Normally Crunch does this for me, but
+                # here I might be reusing the output of someone else's
+                # job and I need to make sure it's understood that the
+                # output is valuable to me, too.)
+                wanted = c[:job][:output]
+                debuglog "checking for existing persistence link for #{wanted}"
+                @my_user_uuid ||= $arv.user.current[:uuid]
+                links = $arv.link.list(limit: 1,
+                                       filters:
+                                       [%w(link_class = resources),
+                                        %w(name = wants),
+                                        %w(tail_uuid =) + [@my_user_uuid],
+                                        %w(head_uuid =) + [wanted]
+                                       ])[:items]
+                if links.any?
+                  debuglog "link already exists, uuid #{links.first[:uuid]}"
+                else
+                  newlink = $arv.link.create link: \
+                  {
+                    link_class: 'resources',
+                    name: 'wants',
+                    tail_kind: 'arvados#user',
+                    tail_uuid: @my_user_uuid,
+                    head_kind: 'arvados#collection',
+                    head_uuid: wanted,
+                    owner_uuid: owner_uuid
+                  }
+                  debuglog "added link, uuid #{newlink[:uuid]}"
+                end
+              end
+            end
           elsif c[:job][:running] ||
               (!c[:job][:started_at] && !c[:job][:cancelled_at])
+            # Job is still running
             moretodo = true
           elsif c[:job][:cancelled_at]
             debuglog "component #{cname} job #{c[:job][:uuid]} cancelled."
@@ -532,7 +556,6 @@ class WhRunPipelineInstance
         end
       end
       @instance[:components] = @components
-      @instance[:active] = moretodo
       report_status
 
       if @options[:no_wait]
@@ -544,7 +567,8 @@ class WhRunPipelineInstance
           sleep 10
         rescue Interrupt
           debuglog "interrupt", 0
-          abort
+          interrupted = true
+          break
         end
       end
     end
@@ -564,18 +588,31 @@ class WhRunPipelineInstance
         end
       end
     end
-    
-    if ended == @components.length or failed > 0
-      @instance[:active] = false
-      @instance[:success] = (succeeded == @components.length)
+
+    success = (succeeded == @components.length)
+
+    if interrupted
+     if success
+        @instance[:state] = 'Complete'
+     else
+        @instance[:state] = 'Paused'
+      end
+    else
+      if ended == @components.length or failed > 0
+        @instance[:state] = success ? 'Complete' : 'Failed'
+      end
     end
 
+    # set components_summary
+    components_summary = {"todo" => @components.length - ended, "done" => succeeded, "failed" => failed}
+    @instance[:components_summary] = components_summary
+
     @instance.save
   end
 
   def cleanup
-    if @instance
-      @instance[:active] = false
+    if @instance and @instance[:state] == 'RunningOnClient'
+      @instance[:state] = 'Paused'
       @instance.save
     end
   end