You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Over time, we should toss the job input from the database. Reasons:
Size - People will send over large pickled python objects because they didn't know about all the globals they set. This may sound like premature optimization, but we'll want to be ready for it...
Privacy - Might as well tear down the input to the job since it shouldn't be needed any more. If someone wants to re-run they should re-submit.
The text was updated successfully, but these errors were encountered:
We'll need to leave the rest of the job metadata in place, so you can trawl through your history and figure out how you used 120 hours of compute three months ago. Looking through Job and SubmittedJob, I would say that we should age off:
Environment
Stdin
Result
Stderr
Stdout
I'm debating Cmd as well. Because we're really running arbitrary processes, it's possible that some user (or some SDK) could pass input by command-line parameters instead of stdin. Maybe we could truncate it?
That makes sense, since we're still keeping the metering data. Some shortening of the Command makes sense, but maybe we don't bother reaping that until we notice a problem.
Stdin should definitely be reaped, as that's our interface for sending pickled python over.
Over time, we should toss the job input from the database. Reasons:
The text was updated successfully, but these errors were encountered: