izgzhen / nightly

Feather-light scheduler, monitor and storage for your research tasks

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Nightly

CI

Introduction

Nightly is a system for scheduling, monitoring and collecting outputs from configuration-defined command-line tasks. It is developed to support my daily research work since I can't find an existing solution that is lightweight (depends on nothing more than python, mysql and ssh) and customizable (written in less than a few hundreds lines of Python in total). I hope it might be useful to you as well.

Features:

  • Define compute/storage resources declaratively
  • Define job declaratively (similar to CI)
  • Run job periodically, or as a daemon, or just once
    • For "just once", you can also run an ad-hoc command line job without defining it
  • Automatically transfer job log and output to storage
  • Monitor the progress via web panel
  • Receive notification in job success/fail/...

Data objects:

  • Job
  • Resource

Components:

  • Log DB: database containing job log (status) entries
  • Scheduler: master node, launch jobs and monitor process status
  • Runner: launch job on runner node through SSH
  • Storage: persisting job output and selected files
  • Panel: web panel daemon
  • Notif: notification (only email for now)

Dependencies

poetry shell # get poetry: https://python-poetry.org/
poetry install
# symblink the virtual env created by poetry to `.venv` under project root as well

Config

There should be two files inside config folder, which is ignored by git. See example_config for examples and comments.

Users can use different config folders in different environment (production or development).

See https://app.sendgrid.com/guide/integrate/langs/python for email notification configuration keys sendgrid_api_key, notif_sender_email and notif_receiver_email.

Runbook

Initialize or update database:

make upgrade-db

Truncate and backup log table:

make truncate-all-log

Run the master daemon for scheduling jobs (on a server within tmux):

make run # it will tee the log to main.log as well

Run web panel daemon:

make panel # debug mode
export BASIC_AUTH_USERNAME=...; export BASIC_AUTH_PASSWORD=...; make panel-prod

Currently, panel server and master daemon must be on the same node.

Launch an ad-hoc job from command line:

scripts/launch-nightly sleep 10 # launch an ad-hoc one on the master node
scripts/launch-nightly --job ... # launch a defined job. Note that it must has "once" schedule

Demo

Homepage:

Output page:

Email notification:

About

Feather-light scheduler, monitor and storage for your research tasks

License:Apache License 2.0


Languages

Language:Python 85.0%Language:HTML 7.7%Language:JavaScript 3.6%Language:Makefile 2.5%Language:Shell 1.1%Language:CSS 0.1%