There’s a mesh with 5-8 members. One member [running on linux] renames a dir. In 30 sec another member [windows7] deletes the new dir and its content. The latter machine is unattended (nobody have deleted the dir manually).
I have normal verbose and audit logs, but don’t see any obvious reason. The logs are available on request.
I ask around before opening a bug report… at least that’s the plan.
I can’t reproduce it intentionally. I can send you both verbose and audit, you may see something in the latter, but I only have log from one node which runs on a real OS and not on a game console (=windows).
Tried again, didn’t happen, but possibly related to networking delays, order of updates or to which node updates which chunk from where.
Or to something I have screwed up, but I really don’t see what.
All I can do is watching audit for deletes closely… it was accidental that I even noticed that.
We can’t usually fix bugs that cannot be reproduced. Also log from one side is close to of having no value, especially given the other side is at fault. If you can repro, open an issue, until then, we leave it in the X-Files folder.
Out of curiosity, did you tweak folder settings during the scan/transfer?
But syncthing deleting stuff by itself is rather scary. Unfortunately I can only speak for myself and I cannot force others to debug.
I was thinking along the line that what could cause ST to remove a freshly created dir and its content [whether it’s possible in a mesh with large delays, unordered traffic, …], but if you don’t have any ideas I cannot really help it much.
Sounds interesting. Have you managed to reproduce this so that we could look into this? The modifiedBy is a bit of a wishy washy value tho, so it could be just misreporting.
Nope, since we don’t usually modify the share actively on multiple nodes while adding new nodes which in syncing the multi-gigabyte folders. However since it’s not easy to see the global state (at least not without writing a realtime dashboard based on the audit log, which I start to become inclined to) I cannot tell what have happened at that moment. I don’t have any idea how to reproduce it on purpose.
Maybe having nodes with long delays, longer circular paths and rapid add/modify/delete cycles? Would require plenty of efforts to put together in a lab env.
Would be nice to have audit to contain the path, or at least the origin of a change.