X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/402461a12dbff7d0ec00c8446b925b021aabc2ed..aee63d7cbb2f8e39b417baebc145889d6290315e:/sdk/cli/bin/arv-run-pipeline-instance diff --git a/sdk/cli/bin/arv-run-pipeline-instance b/sdk/cli/bin/arv-run-pipeline-instance index 9f1663d1a3..472c20bd73 100755 --- a/sdk/cli/bin/arv-run-pipeline-instance +++ b/sdk/cli/bin/arv-run-pipeline-instance @@ -42,6 +42,8 @@ # [--status-json path] Print JSON status report to a file or # fifo. Default: /dev/null # +# [--description] Description for the pipeline instance. +# # == Parameters # # [param_name=param_value] @@ -59,8 +61,6 @@ class WhRunPipelineInstance end -$application_version = 1.0 - if RUBY_VERSION < '1.9.3' then abort <<-EOS #{$0.gsub(/^\.\//,'')} requires Ruby version 1.9.3 or higher. @@ -74,6 +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' @@ -84,7 +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 @@ -108,21 +109,6 @@ if $arvados_api_host.match /local/ suppress_warnings { OpenSSL::SSL::VERIFY_PEER = OpenSSL::SSL::VERIFY_NONE } end -class Google::APIClient - def discovery_document(api, version) - api = api.to_s - return @discovery_documents["#{api}:#{version}"] ||= - begin - response = self.execute!( - :http_method => :get, - :uri => self.discovery_uri(api, version), - :authenticated => false - ) - response.body.class == String ? JSON.parse(response.body) : response.body - end - end -end - # Parse command line options (the kind that control the behavior of # this program, that is, not the pipeline component parameters). @@ -167,13 +153,25 @@ p = Trollop::Parser.new do :short => :none, :type => :string) opt(:submit, - "Do not try to satisfy any components. Just create a pipeline instance and output its UUID.", + "Submit the pipeline instance to the server, and exit. Let the Crunch dispatch service satisfy the components by finding/running jobs.", + :short => :none, + :type => :boolean) + opt(:run_pipeline_here, + "Manage the pipeline instance in-process. Submit jobs to Crunch as needed. Do not exit until the pipeline finishes (or fails).", + :short => :none, + :type => :boolean) + opt(:run_jobs_here, + "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.", :short => :none, :type => :boolean) opt(:run_here, - "Manage the pipeline in process.", + "Synonym for --run-jobs-here.", :short => :none, :type => :boolean) + opt(:description, + "Description for the pipeline instance.", + :short => :none, + :type => :string) stop_on [:'--'] end $options = Trollop::with_standard_exception_handling p do @@ -181,16 +179,21 @@ $options = Trollop::with_standard_exception_handling p do end $debuglevel = $options[:debug_level] || ($options[:debug] && 1) || 0 +$options[:run_jobs_here] ||= $options[:run_here] # old flag name +$options[:run_pipeline_here] ||= $options[:run_jobs_here] # B requires A + if $options[:instance] if $options[:template] or $options[:submit] abort "#{$0}: syntax error: --instance cannot be combined with --template or --submit." end elsif not $options[:template] - abort "#{$0}: syntax error: you must supply a --template or --instance." + puts "error: you must supply a --template or --instance." + p.educate + abort end -if $options[:run_here] == $options[:submit] - abort "#{$0}: syntax error: you must supply either --run-here or --submit." +if $options[:run_pipeline_here] == $options[:submit] + abort "#{$0}: error: you must supply --run-pipeline-here, --run-jobs-here, or --submit." end # Suppress SSL certificate checks if ARVADOS_API_HOST_INSECURE @@ -211,12 +214,9 @@ end # Set up the API client. -$client ||= Google::APIClient. - new(:host => $arvados_api_host, - :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' +$client = $arv.client +$arvados = $arv.arvados_api class PipelineInstance def self.find(uuid) @@ -224,10 +224,10 @@ 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 @@ -239,14 +239,16 @@ class PipelineInstance end def self.create(attributes) result = $client.execute(:api_method => $arvados.pipeline_instances.create, - :body => { - :api_token => ENV['ARVADOS_API_TOKEN'], + :body_object => { :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}" + abort "\n#{Time.now} -- pipeline_template #{@template[:uuid]}\nFailed to create pipeline_instance: #{j[:errors] rescue nil} #{j.inspect}" end debuglog "Created pipeline instance: #{j[:uuid]}" self.new(j) @@ -256,11 +258,13 @@ class PipelineInstance :parameters => { :uuid => @pi[:uuid] }, - :body => { - :api_token => ENV['ARVADOS_API_TOKEN'], - :pipeline_instance => @attributes_to_update.to_json + :body_object => { + :pipeline_instance => @attributes_to_update }, - :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 @@ -277,6 +281,16 @@ class PipelineInstance def [](x) @pi[x] end + + def log_stderr(msg) + $arv.log.create log: { + event_type: 'stderr', + object_uuid: self[:uuid], + owner_uuid: self[:owner_uuid], + properties: {"text" => msg}, + } + end + protected def initialize(j) @attributes_to_update = {} @@ -289,20 +303,24 @@ 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] @@ -310,22 +328,39 @@ class JobCache [] end end - def self.create(attributes) + def self.create(pipeline, component, job, create_params) @cache ||= {} + + body = {job: no_nil_values(job)}.merge(no_nil_values(create_params)) + result = $client.execute(:api_method => $arvados.jobs.create, - :parameters => { - :api_token => ENV['ARVADOS_API_TOKEN'], - :job => attributes.to_json - }, - :authenticated => false) + :body_object => body, + :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} with attribute #{attributes}", 0 + debuglog "create job: #{j[:errors] rescue nil} with attributes #{body}", 0 + + msg = "" + j[:errors].each do |err| + msg += "Error creating job for component #{component}: #{err}\n" + end + msg += "Job submission was: #{body.to_json}" + + pipeline.log_stderr(msg) nil end end + + protected + + def self.no_nil_values(hash) + hash.reject { |key, value| value.nil? } + end end class WhRunPipelineInstance @@ -339,17 +374,15 @@ class WhRunPipelineInstance if template.match /[^-0-9a-z]/ # Doesn't look like a uuid -- use it as a filename. @template = JSON.parse File.read(template), :symbolize_names => true - if !@template[:components] - abort ("#{$0}: Template loaded from #{template} " + - "does not have a \"components\" key") - end 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}" @@ -375,12 +408,29 @@ class WhRunPipelineInstance param = params_args.shift.sub /^--/, '' params[param] = params_args.shift else - abort "Syntax error: I do not know what to do with arg \"#{params_args[0]}\"" + abort "\n#{Time.now} -- pipeline_template #{@template[:uuid]}\nSyntax error: I do not know what to do with arg \"#{params_args[0]}\"" end end + if not @template[:components].is_a?(Hash) + abort "\n#{Time.now} -- pipeline_template #{@template[:uuid]}\nSyntax error: Template missing \"components\" hash" + end @components = @template[:components].dup + bad_components = @components.each_pair.select do |cname, cspec| + not cspec.is_a?(Hash) + end + if bad_components.any? + abort "\n#{Time.now} -- pipeline_template #{@template[:uuid]}\nSyntax error: Components not specified with hashes: #{bad_components.map(&:first).join(', ')}" + end + + bad_components = @components.each_pair.select do |cname, cspec| + not cspec[:script_parameters].is_a?(Hash) + end + if bad_components.any? + abort "\n#{Time.now} -- pipeline_template #{@template[:uuid]}\nSyntax error: Components missing \"script_parameters\" hashes: #{bad_components.map(&:first).join(', ')}" + end + errors = [] @components.each do |componentname, component| component[:script_parameters].each do |parametername, parameter| @@ -404,44 +454,131 @@ class WhRunPipelineInstance end end if !errors.empty? - abort "Errors:\n#{errors.collect { |c,p,e| "#{c}::#{p} - #{e}\n" }.join ""}" + abort "\n#{Time.now} -- pipeline_template #{@template[:uuid]}\nErrors:\n#{errors.collect { |c,p,e| "#{c}::#{p} - #{e}\n" }.join ""}" end debuglog "options=" + @options.pretty_inspect self end def setup_instance - @instance ||= PipelineInstance. - create(:components => @components, - :pipeline_template_uuid => @template[:uuid], - :active => true) + if @instance + @instance[:properties][:run_options] ||= {} + if @options[:no_reuse] + # override properties of existing instance + @instance[:properties][:run_options][:enable_job_reuse] = false + else + # Default to "enable reuse" if not specified. (This code path + # can go away when old clients go away.) + if @instance[:properties][:run_options][:enable_job_reuse].nil? + @instance[:properties][:run_options][:enable_job_reuse] = true + end + end + else + description = $options[:description] + description = ("Created at #{Time.now.localtime}" + (@template[:name].andand.size.andand>0 ? " using the pipeline template *#{@template[:name]}*" : "")) if !description + @instance = PipelineInstance. + create(components: @components, + properties: { + run_options: { + enable_job_reuse: !@options[:no_reuse] + } + }, + pipeline_template_uuid: @template[:uuid], + description: description, + state: ($options[:submit] ? 'RunningOnServer' : 'RunningOnClient')) + end self end def run moretodo = true + interrupted = false + + if @instance[:started_at].nil? + @instance[:started_at] = Time.now + end + + job_creation_failed = 0 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 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], - :minimum_script_version => c[:minimum_script_version], - :exclude_script_versions => c[:exclude_minimum_script_versions], - :nondeterministic => c[:nondeterministic], - :no_reuse => @options[:no_reuse]}) + my_submit_id = "instance #{@instance[:uuid]} rand #{rand(2**64).to_s(36)}" + job = JobCache.create(@instance, cname, { + :script => c[:script], + :script_parameters => c[:script_parameters], + :script_version => c[:script_version], + :repository => c[:repository], + :nondeterministic => c[:nondeterministic], + :runtime_constraints => c[:runtime_constraints], + :owner_uuid => owner_uuid, + :is_locked_by_uuid => (@options[:run_jobs_here] ? owner_uuid : nil), + :submit_id => my_submit_id, + }, { + # 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 => (@instance[:properties][:run_options].andand[:enable_job_reuse] && + !c[:nondeterministic]), + :filters => c[:filters] + }) if job debuglog "component #{cname} new job #{job[:uuid]}" c[:job] = job + c[:run_in_process] = (@options[:run_jobs_here] and + job[:submit_id] == my_submit_id) else - debuglog "component #{cname} new job failed" + debuglog "component #{cname} new job failed", 0 + job_creation_failed += 1 + end + end + + if c[:job] and c[:run_in_process] and c[:job][:success].nil? + report_status + begin + require 'open3' + Open3.popen3("arv-crunch-job", "--force-unlock", + "--job", c[:job][:uuid]) do |stdin, stdout, stderr, wait_thr| + debuglog "arv-crunch-job pid #{wait_thr.pid} started", 0 + stdin.close + while true + rready, wready, = IO.select([stdout, stderr], []) + break if !rready[0] + begin + buf = rready[0].read_nonblock(2**20) + rescue EOFError + break + end + (rready[0] == stdout ? $stdout : $stderr).write(buf) + end + stdout.close + stderr.close + debuglog "arv-crunch-job pid #{wait_thr.pid} exit #{wait_thr.value.to_i}", 0 + end + if not $arv.job.get(uuid: c[:job][:uuid])[:finished_at] + raise Exception.new("arv-crunch-job did not set finished_at.") + end + rescue Exception => e + debuglog "Interrupted (#{e}). Failing job.", 0 + $arv.job.update(uuid: c[:job][:uuid], + job: { + finished_at: Time.now, + running: false, + success: false + }) end end @@ -449,7 +586,7 @@ class WhRunPipelineInstance 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]) + c[:job] = JobCache.get(c[:job][:uuid]) end if c[:job][:success] @@ -464,6 +601,59 @@ 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 @instance[:name].andand.length.andand > 0 + pipeline_name = @instance[:name] + elsif @template.andand[:name].andand.length.andand > 0 + pipeline_name = @template[:name] + else + pipeline_name = @instance[:uuid] + end + if c[:output_name] != false + # Create a collection located in the same project as the pipeline with the contents of the output. + portable_data_hash = c[:job][:output] + collections = $arv.collection.list(limit: 1, + filters: [['portable_data_hash', '=', portable_data_hash]], + select: ["portable_data_hash", "manifest_text"] + )[:items] + if collections.any? + name = c[:output_name] || "Output #{portable_data_hash[0..7]} of #{cname} of #{pipeline_name}" + + # check if there is a name collision. + name_collisions = $arv.collection.list(filters: [["owner_uuid", "=", owner_uuid], + ["name", "=", name]])[:items] + + newcollection_actual = nil + if name_collisions.any? and name_collisions.first[:portable_data_hash] == portable_data_hash + # There is already a collection with the same name and the + # same contents, so just point to that. + newcollection_actual = name_collisions.first + end + + if newcollection_actual.nil? + # Did not find a collection with the same name (or the + # collection has a different portable data hash) so create + # a new collection with ensure_unique_name: true. + newcollection = { + owner_uuid: owner_uuid, + name: name, + portable_data_hash: collections.first[:portable_data_hash], + manifest_text: collections.first[:manifest_text] + } + debuglog "Creating collection #{newcollection}", 0 + newcollection_actual = $arv.collection.create collection: newcollection, ensure_unique_name: true + end + + c[:output_uuid] = newcollection_actual[:uuid] + else + debuglog "Could not find a collection with portable data hash #{portable_data_hash}", 0 + end + end + end elsif c[:job][:running] || (!c[:job][:started_at] && !c[:job][:cancelled_at]) # Job is still running @@ -474,19 +664,24 @@ class WhRunPipelineInstance end end @instance[:components] = @components - @instance[:active] = moretodo report_status if @options[:no_wait] moretodo = false end + # If job creation fails, just give up on this pipeline instance. + if job_creation_failed > 0 + moretodo = false + end + if moretodo begin sleep 10 rescue Interrupt debuglog "interrupt", 0 - abort + interrupted = true + break end end end @@ -496,28 +691,54 @@ class WhRunPipelineInstance failed = 0 @components.each do |cname, c| if c[:job] - if c[:job][:finished_at] + if c[:job][:finished_at] or c[:job][:cancelled_at] or (c[:job][:running] == false and c[:job][:success] == false) ended += 1 if c[:job][:success] == true succeeded += 1 - elsif c[:job][:success] == false + elsif c[:job][:success] == false or c[:job][:cancelled_at] failed += 1 end end end end - - if ended == @components.length or failed > 0 - @instance[:active] = false - @instance[:success] = (succeeded == @components.length) + + success = (succeeded == @components.length) + + # A job create call failed. Just give up. + if job_creation_failed > 0 + debuglog "job creation failed - giving up on this pipeline instance", 0 + success = false + failed += 1 + end + + 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 + if @instance[:finished_at].nil? and ['Complete', 'Failed'].include? @instance[:state] + @instance[:finished_at] = Time.now + end + + debuglog "pipeline instance state is #{@instance[:state]}" + + # 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 @@ -555,6 +776,8 @@ class WhRunPipelineInstance "failed #{c[:job][:finished_at]}" elsif c[:job][:started_at] "started #{c[:job][:started_at]}" + elsif c[:job][:is_locked_by_uuid] + "starting #{c[:job][:started_at]}" else "queued #{c[:job][:created_at]}" end @@ -563,6 +786,19 @@ class WhRunPipelineInstance end end end + + def abort(msg) + if @instance + if ["New", "Ready", "RunningOnClient", + "RunningOnServer"].include?(@instance[:state]) + @instance[:state] = "Failed" + @instance[:finished_at] = Time.now + @instance.save + end + @instance.log_stderr(msg) + end + Kernel::abort(msg) + end end runner = WhRunPipelineInstance.new($options)