gdreade
Dabbler
- Joined
- Mar 11, 2015
- Messages
- 34
Here's an entry for the Really Weird Sh*t category.
I had a situation with 9.3-STABLE where I had two nested filesystems: /mnt/pool1/first and /mnt/pool1/first/second. I set up an AFP share for /mnt/pool1/first as volume "First". On the client end (Yosemite), I was able to access both "/Volumes/First" and "/Volumes/First/second" as one mount. (I was expecting to have to set up an AFP export and a corresponding mount for both, but *shrug*.)
Then I proceeded to run tests.
I found out that when I was generating a large directory hierarchy on *second, the application would error out in rather unexpected ways: It would complain that some things that should be directories were regular files and vice versa. (But not all of them; just some.) Exiting the application and looking at the filesystem on the client (via the shell) showed the weirdness as well; it wasn't just the application.
So I look on the FreeNAS box using the shell. Everything is as expected: Things that should be directories are directories, and things that should be files are files. At least that much is ok (ie: no underlying data corruption).
Restarting the AFP server or rebooting the client makes things look sane again, but before long similar problems are observed (not necessarily with the same files). After playing and flailing a while, I changed things around such that the filesystems were *not* hierarchical: /mnt/pool1/first and /mnt/pool1/second, using two exports on the AFP server side and two mounts on the client side. Now things seem to be sane.
I'm wondering if there was an inconsistency with FSIDs (ie: cache effects and duplicate inode numbers on different filesystems messing things up), but I don't have any data to support it either way.
Theories are solicited.
#define ASIDE
Some people comment that AFP is being deprecated. Others report that SMB on OS-X sucks. Apple's been making it harder to do NFS mounts. It would sure be nice if there was *some* enterprise network filesystem for OS-X that both worked now and looked like it would be supported going forward. And that cloud crap doesn't count. Iterop seems to be taking a nose-dive ...
#undef ASIDE
I had a situation with 9.3-STABLE where I had two nested filesystems: /mnt/pool1/first and /mnt/pool1/first/second. I set up an AFP share for /mnt/pool1/first as volume "First". On the client end (Yosemite), I was able to access both "/Volumes/First" and "/Volumes/First/second" as one mount. (I was expecting to have to set up an AFP export and a corresponding mount for both, but *shrug*.)
Then I proceeded to run tests.
I found out that when I was generating a large directory hierarchy on *second, the application would error out in rather unexpected ways: It would complain that some things that should be directories were regular files and vice versa. (But not all of them; just some.) Exiting the application and looking at the filesystem on the client (via the shell) showed the weirdness as well; it wasn't just the application.
So I look on the FreeNAS box using the shell. Everything is as expected: Things that should be directories are directories, and things that should be files are files. At least that much is ok (ie: no underlying data corruption).
Restarting the AFP server or rebooting the client makes things look sane again, but before long similar problems are observed (not necessarily with the same files). After playing and flailing a while, I changed things around such that the filesystems were *not* hierarchical: /mnt/pool1/first and /mnt/pool1/second, using two exports on the AFP server side and two mounts on the client side. Now things seem to be sane.
I'm wondering if there was an inconsistency with FSIDs (ie: cache effects and duplicate inode numbers on different filesystems messing things up), but I don't have any data to support it either way.
Theories are solicited.
#define ASIDE
Some people comment that AFP is being deprecated. Others report that SMB on OS-X sucks. Apple's been making it harder to do NFS mounts. It would sure be nice if there was *some* enterprise network filesystem for OS-X that both worked now and looked like it would be supported going forward. And that cloud crap doesn't count. Iterop seems to be taking a nose-dive ...
#undef ASIDE