Script server design

Wiki > The Backend System > Nicos > Script server design

What it should do

• Accept multiple clients
• Serve up a list of available commands for the specified instrument over some protocol. The command should contain command parameters and a description, for example:
  • Name: set_ei
  • Parameters: incident energy (double)
  • Description: Positions the choppers to allow the specified incident energy through
• The commands are instrument specific scripts written by the instrument scientists and/or controls team, and are stored on the server
• New commands can be added relatively quickly without requiring changes to the client
• Accept and queue up jobs from clients and add them to the job queue
• Provide feedback on the status of the job queue
• Provide feedback on the status of the current job (% done, current line etc.)
• Iterate through the job queue
• Execute the jobs, one at a time from the queue
• Allow the job queue to be edited (e.g. reordered, jobs skipped or repeated)
• Allow the current job to be paused or aborted
• When a job is paused the next action is to resume it or abort it. It is not possible to run a different job then return to the paused job
• If the job is paused and the user manually changes the beamline then that change is keep when the job is resumed
• If a job is aborted it can be run again, but it will start from the beginning; however, it can be edited before being re-run
• Allow individual jobs to be edited (excluding the current job?)
• Provide a description of what existing jobs consist of (to allow editing)
• Provide an estimate of how long a job will take
• Provide the elapsed time for the current job
• Run a black-box script as a job (see submitted scripts section below)
• Provide script validation, e.g. allow a dry run of a script or job
• Provide robust error handling and reporting

Submitted scripts

Different types of script:

• Script server aware – created by a script generator
  • Loops unwound so each iteration is used to produce a job
  • Each job has tags for providing feedback, such as progress information, to the clients
• Script server aware – written by hand
  • Includes tags for the script server to provide extra functionality for the script server; such as “pause points” and progress information
• Black-box scripts – written by hand
  • Contains no information for the script server
  • Script server runs it as a single job but cannot pause it or report progress
  • Can be aborted

Architectural design

Decisions already made:

• Scripts, commands and jobs will be written and run in CPython

A simple diagram of the design:

Commands:

• The instrument specific commands are loaded from a certain area on the NDX
• The command sources contain documentation describing the command and the parameters
• The list of commands and their information are served

Jobs:

• The Job Queue contains the jobs to be run
• The jobs are Python scripts/snippets
• The Job Manager is responsible for managing the queue and the running of the job
• For each job the Manager spawns a new [#point1 (1)] Python process to run the job
• If the job is configured correctly the Job Manager can query the status of the job

This means there is no state shared between jobs – do we want this?