-
Notifications
You must be signed in to change notification settings - Fork 504
Open
Milestone
Description
Some apps may have large (~1GB) input files that are used by multiple jobs in a batch, and by multiple batches.
There are two aspects in managing such files:
- We don't want to repeatedly download it to a client. Once it's on a client, it should remain there until we're sure that no future jobs will use it.
- We need to eventually delete it from both client and server.
Currently we can make the file sticky. That does 1) but not 2).
We have tables job_file and batch_file_assoc but aren't really using them.
Proposal:
- A file in a user sandbox can be 'pinned'. If a file is pinned it can't be changed or deleted while it's associated with a non-retired batch.
- If a job submission uses a pinned file, it creates a job_file record and a batch association. On the client, the file is sticky.
- Scheduler: list of sticky files is in request message. If any is not longer pinned, tell the client to delete it.
- Job submitters get automated emails reminding them to unpin files. Maybe auto-unpin after some period.
MagicQuantumMechanic
Metadata
Metadata
Assignees
Type
Projects
Status
Backlog
Status
Backlog