Customer 'stale file handle' problem
Zendesk: https://gitlab.zendesk.com/agent/tickets/103632
Customer has multiple web and SSH application nodes that have git-data on NFS mounts. Under some scenario, the repository becomes unavailable on some nodes and then operations like git clone
return an error ./packed-refs: stale file handle
.
In the particular case we saw on a call, two of six nodes reported this error on the same repository. We then went on one of the nodes, navigated to the repository, and ran ls
in the directory. git clone
then worked fine.
- What was it about navigating to the repo and/or
ls
in that directory that 'refreshed' and allowed things to work? Why wouldn't agit clone
cause the same refresh that anls
did? - What causes the stale handle?
- Is some process keeping
packed-refs
open?
Edited by Stan Hu