Scheduling Processes

[ybnd]

Problem statement

Resource-intensive Processes such as Discovery reindexing on large databases or importing many Items at once can have a significant impact on performance.

In practice, we mitigate this by starting when we expect lower traffic (e.g. overnight).

Approach

We'd like to introduce a feature to schedule Processes, so administrators can do this sort of thing ahead of time, from the Angular UI.

This would also be useful for periodic work.

• We already use Spring's @Scheduled for a few things, but we lack a bit of insight into what's running when compared to e.g. actual cron scripts
• The UI and separate log files we already have for Processes could help a lot here

Considerations

For this to work correctly we'll have to improve how we handle Processes on multi-instance setups:

• Processes should be tracked in the Java VM to support multiple Tomcat instances DSpace#10318
• Problem with multi-instance deployment and processes DSpace#10828
• Repository degradation and backend lock caused by concurrent metadata processes in DSpace 8 HA environment DSpace#11814

---

REST issue: DSpace/DSpace#11816

[mwoodiupui]

I'm curious as to why we would not use the OS' job scheduler which is already well developed and serves admirably. What do we gain by duplicating it? Especially since we lose visibility of (some) background tasks and scatter the background-work schedule among multiple schedulers. We also lose the ability to run these tasks in separate processes at lower priority.

Where DSpace has acquired timed background execution of its own I have tended to disable it and add cron jobs instead. When wearing my sysadmin. hat I very much dislike having to hunt all over the system for various little private job schedulers when trying to quiesce the system for maintenance.

[mwoodiupui]

I failed to note that OS job schedulers also have single-shot facilities such as batch and at for running heavy ad-hoc tasks, many of which may be configured to be sensitive to system load when deciding whether to start another task at a particular time.