Closed jesse-gallagher closed 5 years ago
...except the last idea is bad, since it's only ever going to pass in an equal value at the minimum anyway.
It may be "safe enough" to explicitly invalidate the cache on known DB-modifying methods. That won't account for changes from Notes or other servers, but the odds of those changes happening in the same tick are low.
The cache invalidation mostly fixes it, but an scp up can still crap out after a good while for some reason.
This appears to be unrelated to https://github.com/IKSG/nsf-file-server/issues/28 - even when moving all NSF operations to a single thread, it craps out eventually.
The current shape of it isn't due to cache - disabling that entirely still leads to trouble.
The trouble is actually with empty files, which deserves its own issue.
Disabling the cache code or debugging with a breakpoint per-file still works, so it must be a timing thing. It could be related to Domino only supporting "ticks", which will be a tough problem. Maybe the cache should invalidate itself on
>=
instead of just>
.