@docker_bin = ENV['CRUNCH_JOB_DOCKER_BIN']
@docker_run_args = ENV['CRUNCH_JOB_DOCKER_RUN_ARGS']
+ @cgroup_root = ENV['CRUNCH_CGROUP_ROOT']
@arvados_internal = Rails.configuration.git_internal_dir
if not File.exists? @arvados_internal
'--job', job.uuid,
'--git-dir', @arvados_internal]
+ if @cgroup_root
+ cmd_args += ['--cgroup-root', @cgroup_root]
+ end
+
if @docker_bin
cmd_args += ['--docker-bin', @docker_bin]
end
log_throttle_bytes_so_far: 0,
log_throttle_lines_so_far: 0,
log_throttle_bytes_skipped: 0,
+ log_throttle_partial_line_last_at: Time.new(0),
+ log_throttle_first_partial_line: true,
}
i.close
@todo_job_retries.delete(job.uuid)
message = false
linesize = line.size
if running_job[:log_throttle_is_open]
- running_job[:log_throttle_lines_so_far] += 1
- running_job[:log_throttle_bytes_so_far] += linesize
- running_job[:bytes_logged] += linesize
+ partial_line = false
+ skip_counts = false
+ matches = line.match(/^\S+ \S+ \d+ \d+ stderr (.*)/)
+ if matches and matches[1] and matches[1].start_with?('[...]') and matches[1].end_with?('[...]')
+ partial_line = true
+ if Time.now > running_job[:log_throttle_partial_line_last_at] + Rails.configuration.crunch_log_partial_line_throttle_period
+ running_job[:log_throttle_partial_line_last_at] = Time.now
+ else
+ skip_counts = true
+ end
+ end
+
+ if !skip_counts
+ running_job[:log_throttle_lines_so_far] += 1
+ running_job[:log_throttle_bytes_so_far] += linesize
+ running_job[:bytes_logged] += linesize
+ end
if (running_job[:bytes_logged] >
Rails.configuration.crunch_limit_log_bytes_per_job)
elsif (running_job[:log_throttle_bytes_so_far] >
Rails.configuration.crunch_log_throttle_bytes)
remaining_time = running_job[:log_throttle_reset_time] - Time.now
- message = "Exceeded rate #{Rails.configuration.crunch_log_throttle_bytes} bytes per #{Rails.configuration.crunch_log_throttle_period} seconds (crunch_log_throttle_bytes). Logging will be silenced for the next #{remaining_time.round} seconds.\n"
+ message = "Exceeded rate #{Rails.configuration.crunch_log_throttle_bytes} bytes per #{Rails.configuration.crunch_log_throttle_period} seconds (crunch_log_throttle_bytes). Logging will be silenced for the next #{remaining_time.round} seconds."
running_job[:log_throttle_is_open] = false
elsif (running_job[:log_throttle_lines_so_far] >
Rails.configuration.crunch_log_throttle_lines)
remaining_time = running_job[:log_throttle_reset_time] - Time.now
- message = "Exceeded rate #{Rails.configuration.crunch_log_throttle_lines} lines per #{Rails.configuration.crunch_log_throttle_period} seconds (crunch_log_throttle_lines), logging will be silenced for the next #{remaining_time.round} seconds.\n"
+ message = "Exceeded rate #{Rails.configuration.crunch_log_throttle_lines} lines per #{Rails.configuration.crunch_log_throttle_period} seconds (crunch_log_throttle_lines), logging will be silenced for the next #{remaining_time.round} seconds."
running_job[:log_throttle_is_open] = false
+
+ elsif partial_line and running_job[:log_throttle_first_partial_line]
+ running_job[:log_throttle_first_partial_line] = false
+ message = "Rate-limiting partial segments of long lines to one every #{Rails.configuration.crunch_log_partial_line_throttle_period} seconds."
end
end
if message
# Yes, write to logs, but use our "rate exceeded" message
# instead of the log message that exceeded the limit.
+ message += " A complete log is still being written to Keep, and will be available when the job finishes.\n"
line.replace message
true
+ elsif partial_line
+ false
else
running_job[:log_throttle_is_open]
end
j[:log_throttle_lines_so_far] = 0
j[:log_throttle_bytes_skipped] = 0
j[:log_throttle_is_open] = true
+ j[:log_throttle_partial_line_last_at] = Time.new(0)
+ j[:log_throttle_first_partial_line] = true
end
j[:buf].each do |stream, streambuf|
jobrecord = Job.find_by_uuid(job_done.uuid)
if exit_status == EXIT_RETRY_UNLOCKED or (exit_tempfail and @job_retry_counts.include? jobrecord.uuid)
- # The job failed because all of the nodes allocated to it
- # failed. Only this crunch-dispatch process can retry the job:
+ # Only this crunch-dispatch process can retry the job:
# it's already locked, and there's no way to put it back in the
# Queued state. Put it in our internal todo list unless the job
# has failed this way excessively.