class Arvados::V1::JobsController < ApplicationController
accept_attribute_as_json :script_parameters, Hash
- accept_attribute_as_json :resource_limits, Hash
+ accept_attribute_as_json :runtime_constraints, Hash
accept_attribute_as_json :tasks_summary, Hash
skip_before_filter :find_object_by_uuid, :only => :queue
+ skip_before_filter :render_404_if_no_object, :only => :queue
- def index
- want_ancestor = @where[:script_version_descends_from]
- if want_ancestor
- # Check for missing commit_ancestor rows, and create them if
- # possible.
- @objects.
- dup.
- includes(:commit_ancestors). # I wish Rails would let me
- # specify here which
- # commit_ancestors I am
- # interested in.
- each do |o|
- if o.commit_ancestors.
- select { |ca| ca.ancestor == want_ancestor }.
- empty? and !o.script_version.nil?
- begin
- o.commit_ancestors << CommitAncestor.find_or_create_by_descendant_and_ancestor(o.script_version, want_ancestor)
- rescue
+ def create
+ [:repository, :script, :script_version, :script_parameters].each do |r|
+ if !resource_attrs[r]
+ return render json: {
+ :error => "#{r} attribute must be specified"
+ }, status: :unprocessable_entity
+ end
+ end
+
+ # We used to ask for the minimum_, exclude_, and no_reuse params
+ # in the job resource. Now we advertise them as flags that alter
+ # the behavior of the create action.
+ [:minimum_script_version, :exclude_script_versions].each do |attr|
+ if resource_attrs.has_key? attr
+ params[attr] = resource_attrs.delete attr
+ end
+ end
+ if resource_attrs.has_key? :no_reuse
+ params[:find_or_create] = !resource_attrs.delete(:no_reuse)
+ end
+
+ if params[:find_or_create]
+ r = Commit.find_commit_range(current_user,
+ resource_attrs[:repository],
+ params[:minimum_script_version],
+ resource_attrs[:script_version],
+ params[:exclude_script_versions])
+ # Search for jobs whose script_version is in the list of commits
+ # returned by find_commit_range
+ @object = nil
+ incomplete_job = nil
+ Job.readable_by(current_user).where(script: resource_attrs[:script],
+ script_version: r).
+ each do |j|
+ if j.nondeterministic != true and
+ ((j.success == true and j.output != nil) or j.running == true) and
+ j.script_parameters == resource_attrs[:script_parameters]
+ if j.running
+ # We'll use this if we don't find a job that has completed
+ incomplete_job ||= j
+ else
+ # Record the first job in the list
+ if !@object
+ @object = j
+ end
+ # Ensure that all candidate jobs actually did produce the same output
+ if @object.output != j.output
+ @object = nil
+ break
+ end
end
end
- o.commit_ancestors.
- select { |ca| ca.ancestor == want_ancestor }.
- select(&:is).
- first
+ @object ||= incomplete_job
+ if @object
+ return show
+ end
end
- # Now it is safe to do an .includes().where() because we are no
- # longer interested in jobs that have other ancestors but not
- # want_ancestor.
- @objects = @objects.
- includes(:commit_ancestors).
- where('commit_ancestors.ancestor = ? and commit_ancestors.is = ?',
- want_ancestor, true)
end
+
super
end
+ def cancel
+ reload_object_before_update
+ @object.update_attributes! cancelled_at: Time.now
+ show
+ end
+
class LogStreamer
- def initialize(job)
+ Q_UPDATE_INTERVAL = 12
+ def initialize(job, opts={})
@job = job
+ @opts = opts
end
def each
if @job.finished_at
yield "#{@job.uuid} finished at #{@job.finished_at}\n"
return
end
+ while not @job.started_at
+ # send a summary (job queue + available nodes) to the client
+ # every few seconds while waiting for the job to start
+ last_ack_at ||= Time.now - Q_UPDATE_INTERVAL - 1
+ if Time.now - last_ack_at >= Q_UPDATE_INTERVAL
+ nodes_in_state = {idle: 0, alloc: 0}
+ ActiveRecord::Base.uncached do
+ Node.where('hostname is not ?', nil).collect do |n|
+ if n.info[:slurm_state]
+ nodes_in_state[n.info[:slurm_state]] ||= 0
+ nodes_in_state[n.info[:slurm_state]] += 1
+ end
+ end
+ end
+ job_queue = Job.queue
+ n_queued_before_me = 0
+ job_queue.each do |j|
+ break if j.uuid == @job.uuid
+ n_queued_before_me += 1
+ end
+ yield "#{Time.now}" \
+ " job #{@job.uuid}" \
+ " queue_position #{n_queued_before_me}" \
+ " queue_size #{job_queue.size}" \
+ " nodes_idle #{nodes_in_state[:idle]}" \
+ " nodes_alloc #{nodes_in_state[:alloc]}\n"
+ last_ack_at = Time.now
+ end
+ sleep 3
+ ActiveRecord::Base.uncached do
+ @job.reload
+ end
+ end
@redis = Redis.new(:timeout => 0)
+ if @redis.exists @job.uuid
+ # A log buffer exists. Start by showing the last few KB.
+ @redis.
+ getrange(@job.uuid, 0 - [@opts[:buffer_size], 1].max, -1).
+ sub(/^[^\n]*\n?/, '').
+ split("\n").
+ each do |line|
+ yield "#{line}\n"
+ end
+ end
+ # TODO: avoid missing log entries between getrange() above and
+ # subscribe() below.
@redis.subscribe(@job.uuid) do |event|
event.message do |channel, msg|
if msg == "end"
end
end
+ def self._log_tail_follow_requires_parameters
+ {
+ buffer_size: {type: 'integer', required: false, default: 2**13}
+ }
+ end
def log_tail_follow
if !@object.andand.uuid
return render_not_found
end
- self.response.headers['Last-Modified'] = Time.now.ctime.to_s
- self.response_body = LogStreamer.new @object
+ if client_accepts_plain_text_stream
+ self.response.headers['Last-Modified'] = Time.now.ctime.to_s
+ self.response_body = LogStreamer.new @object, {
+ buffer_size: (params[:buffer_size].to_i rescue 2**13)
+ }
+ else
+ render json: {
+ href: url_for(uuid: @object.uuid),
+ comment: ('To retrieve the log stream as plain text, ' +
+ 'use a request header like "Accept: text/plain"')
+ }
+ end
end
def queue
load_where_param
@where.merge!({
started_at: nil,
- is_locked_by: nil,
- cancelled_at: nil
+ is_locked_by_uuid: nil,
+ cancelled_at: nil,
+ success: nil
})
- params[:order] ||= 'priority desc, created_at'
+ params[:order] ||= ['priority desc', 'created_at']
find_objects_for_index
index
end