7 %w{TERM INT}.each do |sig|
10 $stderr.puts "Received #{signame} signal"
15 if ENV["CRUNCH_DISPATCH_LOCKFILE"]
16 lockfilename = ENV.delete "CRUNCH_DISPATCH_LOCKFILE"
17 lockfile = File.open(lockfilename, File::RDWR|File::CREAT, 0644)
18 unless lockfile.flock File::LOCK_EX|File::LOCK_NB
19 abort "Lock unavailable on #{lockfilename} - exit"
23 ENV["RAILS_ENV"] = ARGV[0] || ENV["RAILS_ENV"] || "development"
25 require File.dirname(__FILE__) + '/../config/boot'
26 require File.dirname(__FILE__) + '/../config/environment'
30 LOG_BUFFER_SIZE = 2**20
33 include ApplicationHelper
36 return act_as_system_user
41 @todo_pipelines = PipelineInstance.queue
42 @pipe_auth_tokens ||= { }
46 @@slurm_version ||= Gem::Version.new(`sinfo --version`.match(/\b[\d\.]+\b/)[0])
47 if Gem::Version.new('2.3') <= @@slurm_version
48 `sinfo --noheader -o '%n:%t'`.strip
50 # Expand rows with hostname ranges (like "foo[1-3,5,9-12]:idle")
51 # into multiple rows with one hostname each.
52 `sinfo --noheader -o '%N:%t'`.split("\n").collect do |line|
53 tokens = line.split ":"
54 if (re = tokens[0].match /^(.*?)\[([-,\d]+)\]$/)
55 re[2].split(",").collect do |range|
56 range = range.split("-").collect(&:to_i)
57 (range[0]..range[-1]).collect do |n|
58 [re[1] + n.to_s, tokens[1..-1]].join ":"
68 def update_node_status
69 if Server::Application.config.crunch_job_wrapper.to_s.match /^slurm/
70 @nodes_in_state = {idle: 0, alloc: 0, down: 0}
76 re = line.match /(\S+?):+(idle|alloc|down)/
79 # sinfo tells us about a node N times if it is shared by N partitions
80 next if node_seen[re[1]]
81 node_seen[re[1]] = true
83 # count nodes in each state
84 @nodes_in_state[re[2].to_sym] += 1
86 # update our database (and cache) when a node's state changes
87 if @node_state[re[1]] != re[2]
88 @node_state[re[1]] = re[2]
89 node = Node.where('hostname=?', re[1]).first
91 $stderr.puts "dispatch: update #{re[1]} state to #{re[2]}"
92 node.info[:slurm_state] = re[2]
95 $stderr.puts "dispatch: sinfo reports '#{re[1]}' is not down, but no node has that name"
109 if job.runtime_constraints['min_nodes']
110 min_nodes = begin job.runtime_constraints['min_nodes'].to_i rescue 1 end
115 next if @nodes_in_state[:idle] < min_nodes
119 next if @running[job.uuid]
123 case Server::Application.config.crunch_job_wrapper
126 when :slurm_immediate
127 cmd_args = ["salloc",
132 "--job-name=#{job.uuid}",
133 "--nodes=#{min_nodes}"]
135 raise "Unknown crunch_job_wrapper: #{Server::Application.config.crunch_job_wrapper}"
138 if Server::Application.config.crunch_job_user
139 cmd_args.unshift("sudo", "-E", "-u",
140 Server::Application.config.crunch_job_user,
141 "PERLLIB=#{ENV['PERLLIB']}")
144 job_auth = ApiClientAuthorization.
145 new(user: User.where('uuid=?', job.modified_by_user_uuid).first,
149 crunch_job_bin = (ENV['CRUNCH_JOB_BIN'] || `which arv-crunch-job`.strip)
150 if crunch_job_bin == ''
151 raise "No CRUNCH_JOB_BIN env var, and crunch-job not in path."
154 cmd_args << crunch_job_bin
155 cmd_args << '--job-api-token'
156 cmd_args << job_auth.api_token
160 commit = Commit.where(sha1: job.script_version).first
162 cmd_args << '--git-dir'
163 if File.exists?(File.
164 join(Rails.configuration.git_repositories_dir,
165 commit.repository_name + '.git'))
167 join(Rails.configuration.git_repositories_dir,
168 commit.repository_name + '.git')
171 join(Rails.configuration.git_repositories_dir,
172 commit.repository_name, '.git')
176 $stderr.puts "dispatch: #{cmd_args.join ' '}"
179 i, o, e, t = Open3.popen3(*cmd_args)
181 $stderr.puts "dispatch: popen3: #{$!}"
187 $stderr.puts "dispatch: job #{job.uuid}"
188 start_banner = "dispatch: child #{t.pid} start #{Time.now.ctime.to_s}"
189 $stderr.puts start_banner
190 $redis.set job.uuid, start_banner + "\n"
191 $redis.publish job.uuid, start_banner
192 $redis.publish job.owner_uuid, start_banner
194 @running[job.uuid] = {
210 # no-op -- let crunch-job take care of locking.
215 # no-op -- let crunch-job take care of locking.
220 @running.each do |job_uuid, j|
223 # Throw away child stdout
225 j[:stdout].read_nonblock(2**20)
226 rescue Errno::EAGAIN, EOFError
229 # Read whatever is available from child stderr
232 stderr_buf = j[:stderr].read_nonblock(2**20)
233 rescue Errno::EAGAIN, EOFError
237 j[:stderr_buf] << stderr_buf
238 if j[:stderr_buf].index "\n"
239 lines = j[:stderr_buf].lines("\n").to_a
240 if j[:stderr_buf][-1] == "\n"
243 j[:stderr_buf] = lines.pop
246 $stderr.print "#{job_uuid} ! " unless line.index(job_uuid)
248 pub_msg = "#{Time.now.ctime.to_s} #{line.strip}"
249 $redis.publish job.owner_uuid, pub_msg
250 $redis.publish job_uuid, pub_msg
251 $redis.append job_uuid, pub_msg + "\n"
252 if LOG_BUFFER_SIZE < $redis.strlen(job_uuid)
255 .getrange(job_uuid, (LOG_BUFFER_SIZE >> 1), -1)
265 return if 0 == @running.size
271 pid_done = waitpid(-1, Process::WNOHANG | Process::WUNTRACED)
273 j_done = @running.values.
274 select { |j| j[:wait_thr].pid == pid_done }.
277 rescue SystemCallError
278 # I have @running processes but system reports I have no
279 # children. This is likely to happen repeatedly if it happens at
280 # all; I will log this no more than once per child process I
282 if 0 < @running.select { |uuid,j| j[:warned_waitpid_error].nil? }.size
283 children = @running.values.collect { |j| j[:wait_thr].pid }.join ' '
284 $stderr.puts "dispatch: IPC bug: waitpid() error (#{$!}), but I have children #{children}"
286 @running.each do |uuid,j| j[:warned_waitpid_error] = true end
289 @running.each do |uuid, j|
290 if j[:wait_thr].status == false
291 pid_done = j[:wait_thr].pid
299 job_done = j_done[:job]
300 $stderr.puts "dispatch: child #{pid_done} exit"
301 $stderr.puts "dispatch: job #{job_done.uuid} end"
303 # Ensure every last drop of stdout and stderr is consumed
305 if j_done[:stderr_buf] and j_done[:stderr_buf] != ''
306 $stderr.puts j_done[:stderr_buf] + "\n"
310 j_done[:wait_thr].value
312 jobrecord = Job.find_by_uuid(job_done.uuid)
313 jobrecord.running = false
314 jobrecord.finished_at ||= Time.now
315 # Don't set 'jobrecord.success = false' because if the job failed to run due to an
316 # issue with crunch-job or slurm, we want the job to stay in the queue.
319 # Invalidate the per-job auth token
320 j_done[:job_auth].update_attributes expires_at: Time.now
322 $redis.publish job_done.uuid, "end"
324 @running.delete job_done.uuid
328 expire_tokens = @pipe_auth_tokens.dup
329 @todo_pipelines.each do |p|
330 if @pipe_auth_tokens[p.uuid].nil?
331 pipe_auth = ApiClientAuthorization.
332 new(user: User.where('uuid=?', p.modified_by_user_uuid).first,
335 @pipe_auth_tokens[p.uuid] = pipe_auth
337 pipe_auth = @pipe_auth_tokens[p.uuid]
338 puts `export ARVADOS_API_TOKEN=#{pipe_auth.api_token} && arv-run-pipeline-instance --run-here --no-wait --instance #{p.uuid}`
339 expire_tokens.delete p.uuid
342 expire_tokens.each do |k, v|
343 v.update_attributes expires_at: Time.now
344 @pipe_auth_tokens.delete k
351 $stderr.puts "dispatch: ready"
352 while !$signal[:term] or @running.size > 0
355 @running.each do |uuid, j|
356 if !j[:started] and j[:sent_int] < 2
358 Process.kill 'INT', j[:wait_thr].pid
360 # No such pid = race condition + desired result is
367 refresh_todo unless did_recently(:refresh_todo, 1.0)
369 unless @todo.empty? or did_recently(:start_jobs, 1.0) or $signal[:term]
372 unless (@todo_pipelines.empty? and @pipe_auth_tokens.empty?) or did_recently(:update_pipelines, 5.0)
377 select(@running.values.collect { |j| [j[:stdout], j[:stderr]] }.flatten,
384 def did_recently(thing, min_interval)
386 if !@did_recently[thing] or @did_recently[thing] < Time.now - min_interval
387 @did_recently[thing] = Time.now
395 # This is how crunch-job child procs know where the "refresh" trigger file is
396 ENV["CRUNCH_REFRESH_TRIGGER"] = Rails.configuration.crunch_refresh_trigger