Daniel Gryniewicz

Hmm... As far as I can tell, expiration_time never did anything, even when it was first added years ago. In addition, names are never removed from the cache, either. I'm...

Hmm... It appears that there's no way to define CEPH_PNFS, which means that the pNFS for CephFS hasn't been tested for years, and may not work. You could add the...

This is how the FSAL API works. Default implementations of all API functions are provided in default_methods.c (usually just returning not supported), and a FSAL can override those with a...

In general, you can't have path information. It doesn't exist. The LOOKUP OP has information about the current file/directory being looked up, but that's it. Other OPs only have a...

FWIW, this kind of auditing is hard on any kind of POSIX-like system, because you don't generally have paths. File I/O is generally associated with Inodes, and the "path" to...

The default sizes for the hw_marks are large enough to hold the entire directory, to that isn't a problem. Readdir with gluster should be fast, and was last time I...

I'm not sure what's wrong, then. I haven't tested gluster personally in a long time (maybe gluster 5?) so I'm not sure what's changed. It used to return a large...

What I've done in the past is enable lttng and look at timestamps on the traces to see where time is being spent. That should point us to where the...

137 entries shouldn't cause a problem, that's small enough. Processing multiple entries in a callaback shouldn't make a difference. function call latency isn't that high, and 137 is a tiny...

Not if we want to be correct. The issue is that the client doesn't have to send a SETATTR at all, but we have to set the attributes that were...