3 class WhRunPipelineInstance
6 if RUBY_VERSION < '1.9.3' then
8 #{$0.gsub(/^\.\//,'')} requires Ruby version 1.9.3 or higher.
18 require 'google/api_client'
22 #{$0}: fatal: #{l.message}
23 Some runtime dependencies may be missing.
24 Try: gem install arvados pp google-api-client json trollop
28 def debuglog(message, verbosity=1)
29 $stderr.puts "#{File.split($0).last} #{$$}: #{message}" if $debuglevel >= verbosity
32 # Parse command line options (the kind that control the behavior of
33 # this program, that is, not the pipeline component parameters).
35 p = Trollop::Parser.new do
40 arv-run-pipeline-instance --template TEMPLATE_UUID [options] [--] [parameters]
41 arv-run-pipeline-instance --instance INSTANCE_UUID [options] [--] [parameters]
44 param_name=param_value
45 param_name param_value
46 Set (or override) the default value for every
47 pipeline component parameter with the given
50 component_name::param_name=param_value
51 component_name::param_name param_value
52 --component_name::param_name=param_value
53 --component_name::param_name param_value
54 Set the value of a parameter for a single
60 "Do not start any new jobs or wait for existing jobs to finish. Just find out whether jobs are finished, queued, or running for each component.",
64 "Store plain text status in given file.",
67 :default => '/dev/stdout')
69 "Store json-formatted pipeline in given file.",
72 :default => '/dev/null')
74 "Do not wait for jobs to finish. Just look up status, submit new jobs if needed, and exit.",
78 "Do not reuse existing jobs to satisfy pipeline components. Submit a new job for every component.",
82 "Print extra debugging information on stderr.",
85 "Set debug verbosity level.",
89 "UUID of pipeline template, or path to local pipeline template file.",
93 "UUID of pipeline instance.",
97 "Submit the pipeline instance to the server, and exit. Let the Crunch dispatch service satisfy the components by finding/running jobs.",
100 opt(:run_pipeline_here,
101 "Manage the pipeline instance in-process. Submit jobs to Crunch as needed. Do not exit until the pipeline finishes (or fails).",
105 "Run jobs in the local terminal session instead of submitting them to Crunch. Implies --run-pipeline-here. Note: this results in a significantly different job execution environment, and some Crunch features are not supported. It can be necessary to modify a pipeline in order to make it run this way.",
109 "Synonym for --run-jobs-here.",
113 "Description for the pipeline instance.",
117 "UUID of the project for the pipeline instance.",
122 $options = Trollop::with_standard_exception_handling p do
125 $debuglevel = $options[:debug_level] || ($options[:debug] && 1) || 0
127 $options[:run_jobs_here] ||= $options[:run_here] # old flag name
128 $options[:run_pipeline_here] ||= $options[:run_jobs_here] # B requires A
130 if $options[:instance]
131 if $options[:template] or $options[:submit]
132 abort "#{$0}: syntax error: --instance cannot be combined with --template or --submit."
134 elsif not $options[:template]
135 $stderr.puts "error: you must supply a --template or --instance."
140 if $options[:run_pipeline_here] == $options[:submit]
141 abort "#{$0}: error: you must supply --run-pipeline-here, --run-jobs-here, or --submit."
144 # Set up the API client.
146 $arv = Arvados.new api_version: 'v1'
147 $client = $arv.client
148 $arvados = $arv.arvados_api
150 class PipelineInstance
152 result = $client.execute(:api_method => $arvados.pipeline_instances.get,
156 :authenticated => false,
158 authorization: 'OAuth2 '+$arv.config['ARVADOS_API_TOKEN']
160 j = JSON.parse result.body, :symbolize_names => true
161 unless j.is_a? Hash and j[:uuid]
162 debuglog "Failed to get pipeline_instance: #{j[:errors] rescue nil}", 0
165 debuglog "Retrieved pipeline_instance #{j[:uuid]}"
169 def self.create(attributes)
170 result = $client.execute(:api_method => $arvados.pipeline_instances.create,
172 :pipeline_instance => attributes
174 :authenticated => false,
176 authorization: 'OAuth2 '+$arv.config['ARVADOS_API_TOKEN']
178 j = JSON.parse result.body, :symbolize_names => true
179 unless j.is_a? Hash and j[:uuid]
180 abort "\n#{Time.now} -- pipeline_template #{@template[:uuid]}\nFailed to create pipeline_instance: #{j[:errors] rescue nil} #{j.inspect}"
182 debuglog "Created pipeline instance: #{j[:uuid]}"
186 result = $client.execute(:api_method => $arvados.pipeline_instances.update,
191 :pipeline_instance => @attributes_to_update
193 :authenticated => false,
195 authorization: 'OAuth2 '+$arv.config['ARVADOS_API_TOKEN']
197 j = JSON.parse result.body, :symbolize_names => true
198 unless j.is_a? Hash and j[:uuid]
199 debuglog "Failed to save pipeline_instance: #{j[:errors] rescue nil}", 0
202 @attributes_to_update = {}
207 @attributes_to_update[x] = y
215 $arv.log.create log: {
216 event_type: 'stderr',
217 object_uuid: self[:uuid],
218 owner_uuid: self[:owner_uuid],
219 properties: {"text" => msg},
225 @attributes_to_update = {}
233 result = $client.execute(:api_method => $arvados.jobs.get,
237 :authenticated => false,
239 authorization: 'OAuth2 '+$arv.config['ARVADOS_API_TOKEN']
241 @cache[uuid] = JSON.parse result.body, :symbolize_names => true
243 def self.where(conditions)
244 result = $client.execute(:api_method => $arvados.jobs.list,
247 :where => conditions.to_json
249 :authenticated => false,
251 authorization: 'OAuth2 '+$arv.config['ARVADOS_API_TOKEN']
253 list = JSON.parse result.body, :symbolize_names => true
254 if list and list[:items].is_a? Array
260 def self.create(pipeline, component, job, create_params)
263 body = {job: no_nil_values(job)}.merge(no_nil_values(create_params))
265 result = $client.execute(:api_method => $arvados.jobs.create,
266 :body_object => body,
267 :authenticated => false,
269 authorization: 'OAuth2 '+$arv.config['ARVADOS_API_TOKEN']
271 j = JSON.parse result.body, :symbolize_names => true
272 if j.is_a? Hash and j[:uuid]
275 debuglog "create job: #{j[:errors] rescue nil} with attributes #{body}", 0
278 j[:errors].each do |err|
279 msg += "Error creating job for component #{component}: #{err}\n"
281 msg += "Job submission was: #{body.to_json}"
283 pipeline.log_stderr(msg)
290 def self.no_nil_values(hash)
291 hash.reject { |key, value| value.nil? }
295 class WhRunPipelineInstance
296 attr_reader :instance
298 def initialize(_options)
302 def fetch_template(template)
303 if template.match /[^-0-9a-z]/
304 # Doesn't look like a uuid -- use it as a filename.
305 @template = JSON.parse File.read(template), :symbolize_names => true
307 result = $client.execute(:api_method => $arvados.pipeline_templates.get,
311 :authenticated => false,
313 authorization: 'OAuth2 '+$arv.config['ARVADOS_API_TOKEN']
315 @template = JSON.parse result.body, :symbolize_names => true
317 abort "#{$0}: fatal: failed to retrieve pipeline template #{template} #{@template[:errors].inspect rescue nil}"
323 def fetch_instance(instance_uuid)
324 @instance = PipelineInstance.find(instance_uuid)
325 @template = @instance
329 def apply_parameters(params_args)
330 params_args.shift if params_args[0] == '--'
332 while !params_args.empty?
333 if (re = params_args[0].match /^(--)?([^-].*?)=(.+)/)
334 params[re[2]] = re[3]
336 elsif params_args.size > 1
337 param = params_args.shift.sub /^--/, ''
338 params[param] = params_args.shift
340 abort "\n#{Time.now} -- pipeline_template #{@template[:uuid]}\nSyntax error: I do not know what to do with arg \"#{params_args[0]}\""
344 if not @template[:components].is_a?(Hash)
345 abort "\n#{Time.now} -- pipeline_template #{@template[:uuid]}\nSyntax error: Template missing \"components\" hash"
347 @components = @template[:components].dup
349 bad_components = @components.each_pair.select do |cname, cspec|
350 not cspec.is_a?(Hash)
352 if bad_components.any?
353 abort "\n#{Time.now} -- pipeline_template #{@template[:uuid]}\nSyntax error: Components not specified with hashes: #{bad_components.map(&:first).join(', ')}"
356 bad_components = @components.each_pair.select do |cname, cspec|
357 not cspec[:script_parameters].is_a?(Hash)
359 if bad_components.any?
360 abort "\n#{Time.now} -- pipeline_template #{@template[:uuid]}\nSyntax error: Components missing \"script_parameters\" hashes: #{bad_components.map(&:first).join(', ')}"
364 @components.each do |componentname, component|
365 component[:script_parameters].each do |parametername, parameter|
366 parameter = { :value => parameter } unless parameter.is_a? Hash
368 (params["#{componentname}::#{parametername}"] ||
370 (parameter[:output_of].nil? &&
371 (params[parametername.to_s] ||
372 parameter[:default])) ||
375 ![false,'false',0,'0'].index parameter[:required]
376 if parameter[:output_of]
377 if not @components[parameter[:output_of].intern]
378 errors << [componentname, parametername, "output_of refers to nonexistent component '#{parameter[:output_of]}'"]
382 errors << [componentname, parametername, "required parameter is missing"]
384 debuglog "parameter #{componentname}::#{parametername} == #{value}"
386 component[:script_parameters][parametername] =
387 parameter.dup.merge(value: value)
391 abort "\n#{Time.now} -- pipeline_template #{@template[:uuid]}\nErrors:\n#{errors.collect { |c,p,e| "#{c}::#{p} - #{e}\n" }.join ""}"
393 debuglog "options=" + @options.pretty_inspect
399 @instance[:properties][:run_options] ||= {}
400 if @options[:no_reuse]
401 # override properties of existing instance
402 @instance[:properties][:run_options][:enable_job_reuse] = false
404 # Default to "enable reuse" if not specified. (This code path
405 # can go away when old clients go away.)
406 if @instance[:properties][:run_options][:enable_job_reuse].nil?
407 @instance[:properties][:run_options][:enable_job_reuse] = true
411 description = $options[:description] ||
412 ("Created at #{Time.now.localtime}" + (@template[:name].andand.size.andand>0 ? " using the pipeline template *#{@template[:name]}*" : ""))
414 components: @components,
417 enable_job_reuse: !@options[:no_reuse]
420 pipeline_template_uuid: @template[:uuid],
421 description: description,
422 state: ($options[:submit] ? 'RunningOnServer' : 'RunningOnClient')
424 if @options[:project_uuid]
425 instance_body[:owner_uuid] = @options[:project_uuid]
427 @instance = PipelineInstance.create(instance_body)
436 if @instance[:started_at].nil?
437 @instance[:started_at] = Time.now
440 job_creation_failed = 0
443 @components.each do |cname, c|
445 owner_uuid = @instance[:owner_uuid]
446 # Is the job satisfying this component already known to be
447 # finished? (Already meaning "before we query API server about
448 # the job's current state")
449 c_already_finished = (c[:job] &&
451 ["Complete", "Failed", "Cancelled"].include?(c[:job][:state]))
453 c[:script_parameters].select { |pname, p| p.is_a? Hash and p[:output_of]}.empty?
454 # No job yet associated with this component and is component inputs
455 # are fully specified (any output_of script_parameters are resolved
457 my_submit_id = "instance #{@instance[:uuid]} rand #{rand(2**64).to_s(36)}"
458 job = JobCache.create(@instance, cname, {
459 :script => c[:script],
460 :script_parameters => Hash[c[:script_parameters].map do |key, spec|
463 :script_version => c[:script_version],
464 :repository => c[:repository],
465 :nondeterministic => c[:nondeterministic],
466 :runtime_constraints => c[:runtime_constraints],
467 :owner_uuid => owner_uuid,
468 :is_locked_by_uuid => (@options[:run_jobs_here] ? owner_uuid : nil),
469 :submit_id => my_submit_id,
470 :state => (if @options[:run_jobs_here] then "Running" else "Queued" end)
472 # This is the right place to put these attributes when
473 # dealing with new API servers.
474 :minimum_script_version => c[:minimum_script_version],
475 :exclude_script_versions => c[:exclude_minimum_script_versions],
476 :find_or_create => (@instance[:properties][:run_options].andand[:enable_job_reuse] &&
477 !c[:nondeterministic]),
478 :filters => c[:filters]
481 debuglog "component #{cname} new job #{job[:uuid]}"
483 c[:run_in_process] = (@options[:run_jobs_here] and
484 job[:submit_id] == my_submit_id)
486 debuglog "component #{cname} new job failed", 0
487 job_creation_failed += 1
491 if c[:job] and c[:run_in_process] and not ["Complete", "Failed", "Cancelled"].include? c[:job][:state]
495 Open3.popen3("arv-crunch-job", "--force-unlock",
496 "--job", c[:job][:uuid]) do |stdin, stdout, stderr, wait_thr|
497 debuglog "arv-crunch-job pid #{wait_thr.pid} started", 0
500 rready, wready, = IO.select([stdout, stderr], [])
503 buf = rready[0].read_nonblock(2**20)
507 (rready[0] == stdout ? $stdout : $stderr).write(buf)
511 debuglog "arv-crunch-job pid #{wait_thr.pid} exit #{wait_thr.value.to_i}", 0
513 if not $arv.job.get(uuid: c[:job][:uuid])[:finished_at]
514 raise Exception.new("arv-crunch-job did not set finished_at.")
516 rescue Exception => e
517 debuglog "Interrupted (#{e}). Failing job.", 0
518 $arv.job.update(uuid: c[:job][:uuid],
525 if c[:job] and c[:job][:uuid]
526 if ["Running", "Queued"].include?(c[:job][:state])
527 # Job is running (or may be soon) so update copy of job record
528 c[:job] = JobCache.get(c[:job][:uuid])
531 if c[:job][:state] == "Complete"
532 # Populate script_parameters of other components waiting for
534 @components.each do |c2name, c2|
535 c2[:script_parameters].each do |pname, p|
536 if p.is_a? Hash and p[:output_of] == cname.to_s
537 debuglog "parameter #{c2name}::#{pname} == #{c[:job][:output]}"
538 c2[:script_parameters][pname] = {value: c[:job][:output]}
543 unless c_already_finished
544 # This is my first time discovering that the job
545 # succeeded. (At the top of this loop, I was still
546 # waiting for it to finish.)
548 if @instance[:name].andand.length.andand > 0
549 pipeline_name = @instance[:name]
550 elsif @template.andand[:name].andand.length.andand > 0
551 pipeline_name = @template[:name]
553 pipeline_name = @instance[:uuid]
555 if c[:output_name] != false
556 # Create a collection located in the same project as the pipeline with the contents of the output.
557 portable_data_hash = c[:job][:output]
558 collections = $arv.collection.list(limit: 1,
559 filters: [['portable_data_hash', '=', portable_data_hash]],
560 select: ["portable_data_hash", "manifest_text"]
563 name = c[:output_name] || "Output #{portable_data_hash[0..7]} of #{cname} of #{pipeline_name}"
565 # check if there is a name collision.
566 name_collisions = $arv.collection.list(filters: [["owner_uuid", "=", owner_uuid],
567 ["name", "=", name]])[:items]
569 newcollection_actual = nil
570 if name_collisions.any? and name_collisions.first[:portable_data_hash] == portable_data_hash
571 # There is already a collection with the same name and the
572 # same contents, so just point to that.
573 newcollection_actual = name_collisions.first
576 if newcollection_actual.nil?
577 # Did not find a collection with the same name (or the
578 # collection has a different portable data hash) so create
579 # a new collection with ensure_unique_name: true.
581 owner_uuid: owner_uuid,
583 portable_data_hash: collections.first[:portable_data_hash],
584 manifest_text: collections.first[:manifest_text]
586 debuglog "Creating collection #{newcollection}", 0
587 newcollection_actual = $arv.collection.create collection: newcollection, ensure_unique_name: true
590 c[:output_uuid] = newcollection_actual[:uuid]
592 debuglog "Could not find a collection with portable data hash #{portable_data_hash}", 0
596 elsif ["Queued", "Running"].include? c[:job][:state]
597 # Job is running or queued to run, so indicate that pipeline
598 # should continue to run
600 elsif c[:job][:state] == "Cancelled"
601 debuglog "component #{cname} job #{c[:job][:uuid]} cancelled."
603 elsif c[:job][:state] == "Failed"
608 @instance[:components] = @components
611 if @options[:no_wait]
615 # If job creation fails, just give up on this pipeline instance.
616 if job_creation_failed > 0
624 debuglog "interrupt", 0
631 c_in_state = @components.values.group_by { |c|
632 c[:job] and c[:job][:state]
634 succeeded = c_in_state["Complete"].andand.count || 0
635 failed = (c_in_state["Failed"].andand.count || 0) + (c_in_state["Cancelled"].andand.count || 0)
636 ended = succeeded + failed
638 success = (succeeded == @components.length)
640 # A job create call failed. Just give up.
641 if job_creation_failed > 0
642 debuglog "job creation failed - giving up on this pipeline instance", 0
649 @instance[:state] = 'Complete'
651 @instance[:state] = 'Paused'
654 if ended == @components.length or failed > 0
655 @instance[:state] = success ? 'Complete' : 'Failed'
659 if @instance[:finished_at].nil? and ['Complete', 'Failed'].include? @instance[:state]
660 @instance[:finished_at] = Time.now
663 debuglog "pipeline instance state is #{@instance[:state]}"
665 # set components_summary
666 components_summary = {"todo" => @components.length - ended, "done" => succeeded, "failed" => failed}
667 @instance[:components_summary] = components_summary
673 if @instance and @instance[:state] == 'RunningOnClient'
674 @instance[:state] = 'Paused'
688 if @options[:status_json] != '/dev/null'
689 File.open(@options[:status_json], 'w') do |f|
690 f.puts @components.pretty_inspect
694 if @options[:status_text] != '/dev/null'
695 File.open(@options[:status_text], 'w') do |f|
697 f.puts "#{Time.now} -- pipeline_instance #{@instance[:uuid]}"
698 namewidth = @components.collect { |cname, c| cname.size }.max
699 @components.each do |cname, c|
700 jstatus = if !c[:job]
702 else case c[:job][:state]
704 "#{c[:job][:tasks_summary].inspect}"
708 "cancelled #{c[:job][:cancelled_at]}"
710 "failed #{c[:job][:finished_at]}"
712 "queued #{c[:job][:created_at]}"
715 f.puts "#{cname.to_s.ljust namewidth} #{c[:job] ? c[:job][:uuid] : '-'.ljust(27)} #{jstatus}"
723 if ["New", "Ready", "RunningOnClient",
724 "RunningOnServer"].include?(@instance[:state])
725 @instance[:state] = "Failed"
726 @instance[:finished_at] = Time.now
729 @instance.log_stderr(msg)
735 runner = WhRunPipelineInstance.new($options)
737 if $options[:template]
738 runner.fetch_template($options[:template])
740 runner.fetch_instance($options[:instance])
742 runner.apply_parameters(p.leftovers)
743 runner.setup_instance
746 puts runner.instance[:uuid]
750 rescue Exception => e