[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Nmh-workers] folder +./ # Blocks forever where ./ currently mounted
Re: [Nmh-workers] folder +./ # Blocks forever where ./ currently mounted by autofs (NFS4).
Fri, 05 Aug 2016 21:23:39 -0400
>> I don't know the specifics of your NFS filesystem, but does locking
>> actually work on it?
>Sorry, don't know. I might research a way to test that.
If you're capable, a simple test program would do it. One that checks
read locks and write locks. From what you posted, read locks work, write
locks do not (at least to the older server). It looks like you're running
the right things; it may just be a bug somewhere.
>This shell is on the same host as before. I automount ~/Mail to a
>remote NFS filesystem, but that remote host is physically in same
>datacenter. All 3 hosts have similar firewall setups, that I think (I'm
>not an expert) these are good/secure for NFS.
It does occur to me that a firewall can interfere with RPC services, since
you don't know what port the services will end up using; I see "nlockmgr"
running on port 45923 and 43311 on your server, so maybe checking with
tcpdump to see if your client is sending stuff to it and if it is actually
getting received by the server would be useful.
This does bring to light some nmh behavior that is sub-optimal. We use
write-locks in folder.c to read the sequence file (that's what the
second argument to folder_read() ends up doing). But unless we're packing
the folder that isn't necessary; it occurs to me that we should probably just
get a read lock when we're not calling folder_pack(). This might make
the "folder" command work for you, but plenty of other nmh commands lock
the sequence file so it would just be a band-aid on top of your real problem.