Quickstart

sis command

Sisyphus is started by running the sis command in it’s folder. The main mode of this tool is sis manager or short sis m it will parses the config directive and will submit the required job to the cluster. The manager will periodically check which jobs have finished and submits all jobs that became runnable to the cluster as long as it is running. If you stop the manager (using Ctrl-C) no further jobs are submitted, but jobs submitted to the cluster will continue.

Cheatsheet

Manager commands

What to do

Command

load the default graph and run the manager

./sis m -r

load the default graph and prompt for options

./sis m

load a sub-graph function

./sis m config.sub_module.sub_function

Handling jobs and tasks

What to do

Command

run a (already created) task directly*

./sis worker <path/to/job.hash> <task_name>

create a non-existing job manually

  1. ./sis c

  2. jobs = tk.find_job("<query>")
    this will list all found jobs
  3. j = jobs[<desired_index>]

  4. tk.setup_job_directory(j)

restart a crashed job (by deleting the job)

rm -r <path/to/job.hash>

restart a crashed task (keeping the job)

  1. rm <path/to/job.hash>/error.<task_name>.*

  2. rm <path/to/job.hash>/log.<task_name>.*

restart an interrupted task

rm <path/to/job.hash>/log.<task_name>.*

restart/continue after a retry_error

rm <path/to/job.hash>/submit_log.<task_name>

*(WARNING: this might run the job with different ENVIRONMENT variables)

Quick console commands

What to do

Command

import jobs from another work directory
(symlink or copy)
  1. ./sis c

  2. tk.import_work_directory(<path/to/other/work>)

delete jobs that are no longer part of
the graph (WARNING: make sure your default
graph contains all subgraphs)
  1. ./sis c

  2. tk.cleaner.cleanup_unused()