Bug: logback timeout for import logs too long (30 min)

[jamoore]

Internal reference: ​https://trello.com/c/P4wOlFcd/86-close-sifted-logs
See also: #12012

Each imported fileset opens a new log file of the form:

$ManagedRepo/$User/$yearmonth/$day/$timestamp.log

These files are only cleaned up on the *next* import after 30 minutes. This can run the server out of file handles.

In newer version of logback, this value is configurable.

[jamoore]

This still planned for 5.0.1

[cblackburn]

Is that a question? If so I'd say it's not straightforward given the logback issue. If it's a statement then I'd say the same but it will involve us hacking our own version of the sifting appender at the very least.

[jamoore]

Guess we need to summarize the state of things with your latest PRs. If there's a chance of files hanging around for 30 minutes and if there's a way we can prevent that without a week of work, it'd probably be a good thing. If not, let's push to 5.0.2 and hope for a logback release.

[cblackburn]

Although logback has been updated this ticket is far from resolved due to existing limitations in SiftingAppenders. I'm not what the best way to move forward is at the moment but certainly pushing to 5.0.2 for now.

[jamoore]

Does this need to be pushed again?

[jamoore]

Short summary from chat: yes.

[cblackburn]

Bumped to 5.1 as per Trello card.

[cblackburn]

The limitations on SiftingAppenders are still there, ie ​http://jira.qos.ch/browse/LOGBACK-940?page=com.atlassian.jira.plugin.system.issuetabpanels:changehistory-tabpanel so pushing this ticket to 5.x until there is some movement on that issue or some other solution presents itself. Given this is in production use and there haven't been any bug reports due to open log files I'm not convinced it should be listed as critical either so downgrading a little.

[mtbcarroll]

There may be some interaction with this issue if we start to zip the log files.