Discussing this with Oleg Nenashev on IRC right now, on ci.jenkins.io we have:
% ls -lah /var/lib/jenkins/support/winsw/ | wc -l 22536 %
The instance is only provisioning Windows agents via the Azure VM Agents plugin, so it appears that none of the old logs are being cleaned up after the agents are deleted.
My first guess is that SmartLogFetcher is not that smart actually: https://github.com/jenkinsci/support-core-plugin/blob/48faf43ac7d7e55d1f9115df66bc216dbaa0c232/src/main/java/com/cloudbees/jenkins/support/impl/SlaveLogs.java#L104 18:05 <@rtyler> Tyler huh, interesting 18:06 I definitely haven't generated 22k support bundles 18:07 well, intentionally 18:08 <+oleg-nenashev> Oleg Nenashev "support/winsw/ " is a cache. So there will be one file for every log entry which *seems* to be a WinSW log 18:09 <+oleg-nenashev> Oleg Nenashev rtyler: So you will get an entry for any *.out or *.log entry in the root of any agent (Windows or Linux): https://github.com/jenkinsci/support-core-plugin/blob/2889a28ae41ab7379127c413ae112e9243c778ba/src/main/java/com/cloudbees/jenkins/support/impl/WinswLogfileFilter.java#L13-L20
Apparently it just fails to clean the cache properly for offline agents. Not 100% sure
PR sent