Re: [leafnode-list] locking, again

[Matthias Andree <ma@xxxxxxxxxxxxxxxxxxxxxxxxxxxx>]

krasel@xxxxxxxxxxxxxxxxxxxxxxxxxxxx (Cornelius Krasel) writes:

> [I can't really argue about NFS because I am only parrotting what other
>  people say about it - I have no own experiences with it.]

User-space NFSd only does v2 (tcp, udp) and cannot lock.  Kernel-space
NFSd does not interoperate with non-Linux clients, does not do v3, it
can do locking, it breaks clients if the server reboots.  Kernel-space
NFSd from 2.2.18+ and 2.4 will be interoperable with non-Linux clients
(I have it working with Solaris currently on a patched 2.2.17), it will
have locking, so effectively, these are the minimum requirements for
PROPER NFS:

kernel 2.2.18 or 2.4 (neither of these is yet released).
nfs-utils 0.2.1 with proper boot script.

Apart from that, distributors tend to mess things up severely, SuSE 6.4
does not work properly out-of-the-box. SuSE 6.2 does not either, but it
is to be excused since kernel-space NFSd was not mature at the time SuSE
6.2 came out. I'll shortly have first-hand experience with SuSE 7.0 as
well. Update has been delivered, but not yet installed.

> As far as I can see, there is no process unlinking anything that doesn't
> belong to it. (At least in theory; it will be true in practice as well
> when writeactive() is moved into the main fetchnews process as suggested
> by several people.) Several processes are unlinking lockfiles which do
> not belong to anybody any more (because the process originally owning
> the lockfile crashed). I believe this is a legitimate approach.

The entire locking will need to be encapsulated. I have not yet found
the time to do that, I will provide a lock library and incorporate that
into 2.0b5_ma5.

-- 
Matthias Andree

-- 
leafnode-list@xxxxxxxxxxxxxxxxxxxxxxxxxxxx -- mailing list for leafnode
To unsubscribe, send mail with "unsubscribe" in the subject to the list