tscheduler

tscheduler

Command line tasks/jobs scheduler with from seconds to years precision and the jobs control

Disclaimer

This is an early ALPHA and will probably stay as it is, you've been warned.

If you're looking for a Cron compatible implementation to use as a library in your applications, then please take a look at robfig/cron or any of gron forks, I've used a lot of ideas from those.

I don't think that tscheduler is mature enough to be used as a separate library package, so will not split it into the library from the application code for now.

Motivation

Mostly for the Go practice, but also I don't like the classic Crontab format as had to constantly recall the fields position. Some things should just go. Also built in job control is IMO better than trying to script locks and timeouts.

The snippet of config file below is self-explanatory.

logging:
  # Development key puts the logger in development mode, which changes the behavior of DPanicLevel and takes stacktraces more liberally.
  development: true
  # level: info
  level: debug
  encoding: console
  # encoding: json
  disable_caller: false
  disable_stacktrace: false
  disable_color: false
  # output_paths: ["stdout", "/tmp/1.log"]
  output_paths: ["stdout"]
  error_output_paths: ["stderr"]
# Starts Prometheus metrics service on http://127.0.0.1:9002/metrics
metrics:
  enabled: true
  address: 127.0.0.1:9002
# Starts management service on http://127.0.0.1:9001/
# Required for 'status', 'shutdown', 'pause', 'resume' commands
management:
  enabled: true
  address: 127.0.0.1:9001
  # Wait timeout for scheduler to stop, seconds
  scheduler_stop_timeout: 5
  # After scheduler is stopped, wait for this seconds for jobs to finish
  jobs_termination_timeout: 5
jobs:
  - id: 'Test Job #1'
    parallel: false
    # parallel: true
    # timeout: 3s
    command: 
      - bash
      - -c 
      - 'ls -la /tmp/1out.log'
    stdout: "/tmp/1out.log"
    stderr: /tmp/1err.log
    schedule:
      - year: 2021-2022,2030
        month: 1, 2, 4-5, 3
        day: 1
        weekday: "*"
        hour: 1
        minute: 2, 3
        second: 3
        location: UTC
      - month: "*"
        day: "*"
        weekday: "*"
        hour: "*"
        minute: "*"
        second: 00
        location: UTC
      - month: "*"
        day: "*"
        weekday: "0,2,3,4,5,6"
        hour: "*"
        minute: "*"
        second: "/3"
        # If location is missing, location: "Local", i.e. local timezone
  - id: 'Test Job #2'
    command: 
      - /bin/sleep
      - 10
    parallel: false
    timeout: 10s
    schedule:
      - month: "*"
        # if weekday is missed, default - any day
        day: "*"
        hour: "*"
        minute: "*"
        # 30, 32, ...,58
        second: "30/2"

Features

• Multiple schedules per job to run the same command on different days at different times. This is not Cron expressions compatible implementation, I've struggled a lot trying to recall the position of specific time in its spec, so tscheduler was written to use the fully specified time keywords in its config instead. It still uses some Crontab constructs to specify schedule for convenience:

  • "*" - all time spec values, i.e. for minutes: 00 - 59.

  • "start/step", e.g. */4 for 0, 4, ..., 56. Or 40/6 for minute to specify 40, 46, 52, 58, 00 - BUT BEWARE, we end on 00, NOT 58min + 6min = 4min.

  • "min-max" ranges, e.g. 40-50 means 40, 41, 42, ... , 50.

  • comma separated values, e.g. 4, 5, 6-10, 30/2.

  NOT SUPPORTED:

  • "#", "?", "W"
  • "L" (e.g. the last day of the month). Workaround is possible with multiple schedules, though beware of Feb 29.
  • symbolic definitions of weekday (Mon, Fri) and month (Jan, Sep) - only numeric is used.

• Year and seconds precision by default.

  If the year is ommited, will try to find out the next job run for the every year from the current to 10 years in the future.

  If seconds are ommited, by default it specifies 00.

• Location (Timezone) per schedule. A job can run with the specified timezone per schedule.

  Note that we don't handle summer/winter daylight shifts, it is possible for job to skip its run or run twice during the shift. Recommended to use UTC. By default the location is Local, i.e. the one of the machine.

• Job control:

  • Stdout and stderr redirection of the running command to files. File rotation is not supported.
  • Simultaneous run of job is disabled by default, change it with "parallel" key.
  • Job timeout after which the job will be killed.

• Instrumentation:

  • HTTP based management interface to pause, resume, or gracefully shutdown scheduler with configurable timeout to wait for jobs to finish.
  • Prometheus metrics.
  • Scheduler logging based on uber/zapp library, with the ability to switch to structured (json) logging. Logs rotation is not supported. Colored severity keywords, internal information like function caller and line are also configurable. Jobs run status and next scheduled time is reported.

Installation

Fetch the prebuilt binary from the Releases page. Only Linux is avaliable for now.

Installation from the source:

go get https://github.com/Tarick/tscheduler

Configuration

See examples/config.yaml.

The path to the config file could be passed via --config option or looked up in order in:

• current dir
• $HOME/.config/tscheduler/
• /etc/tscheduler/