@MoritzMuehlenhoff found out this morning that mw1152 had /tmp full of already transcoded files. It seems like we don't clean them up upon job execution, or something like that.
I could set up a cronjob to clean them but I think we should just take care of removing them once the files are copied to swift.
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Resolved | Joe | T113447 HHVM timeouts mean videoscaling can't clean locally transcoded files from the filesystem | |||
| Resolved | Joe | T113284 Frequent job timeouts on HHVM video scalers |
Are these from completed jobs or jobs that timed out during the shell-out, halting HHVM PHP execution before it got a chance to clean up its files?
I notice a lot of the webms have 2-pass log files alongside them, so definitely looks like jobs got shut down while in progress, leaving no chance for the PHP code to clean up the file. The code looks like it correctly purges when done, as long as it has a chance to run.