Re: Issue 05 was (lease expiration, stable storage (was "questions/comments on draft-ietf-nfsv4-rfc3010bis-02-04.txt")

New Message Reply About this list Date view Thread view Subject view Author view Attachment view

From: Mike Eisler (mike@eisler.com)
Date: 08/21/02-07:16:21 PM Z


Message-ID: <3D642D55.8400C1D4@eisler.com>
Date: Wed, 21 Aug 2002 17:16:21 -0700
From: Mike Eisler <mike@eisler.com>
Subject: Re: Issue 05 was (lease expiration, stable storage (was  "questions/comments on draft-ietf-nfsv4-rfc3010bis-02-04.txt")

mike.kupfer@sun.com wrote:

> Modulo some wordsmithing and one technical concern, I'm happy with
> your text.
> 
>     mre>    For the two
>     mre>    aforementioned edge conditions, the harshest a server can
>     mre>    be, and still support a grace period for reclaims,
>     mre>    requires that the server record in stable storage
>     mre>    information for each client containing the client's id
>     mre>    string, a boolean that indicates if the client's lease
>     mre>    expired or if there was administrative intervention (see
>     mre>    the section, Server Revocation of Locks) to revoke a
>     mre>    record lock, share reservation, or delegation, and a
>     mre>    timestamp of the last time the client held locking
>     mre>    state.
> 
> If I understand your proposal, it seems like the server would have to
> update the timestamp (and therefore, stable storage) fairly often.
> I'm wondering if a less expensive approach is possible.  The Spritely
> NFS papers talk about recording the "epoch" of the client's activity,
> where each server reboot starts a new epoch.  This would presumably
> reduce how often the server has to update stable storage.

Yet another example as to why we have to be careful about implementation
suggestions. I agree that a more optimal approach is possible.

I wrote:

   For the second edge condition, after the server reboots
   for a second time, the record that the client had an
   unexpired lock established before the server's previous
   incarnation means that the server must reject a reclaim
   from client A with the error NFS4ERR_NO_GRACE.

I think if the server simply recorded the fact that client established
a lock after a server incarnation, it would not be necessary for
the server to keep re-recording this information. That seems to be
equivalent to what you wrote regarding Spritely NFS epochs.

If we've reached consensus, I will send Spencer the corrected edits
today.

> 
>     mre>    The third lock revocation event can occur as a result of
>     mre>    administrative intervention within the lease period.
>     mre>    While this is considered a rare event, it is possible that
>     mre>    the server's administrator has decided to release or
>     mre>    revoke a particular lock held by the client.  As a result
>     mre>    of revocation, the client will receive an error of
>     mre>    NFS4ERR_ADMIN_REVOKED and the error is received within the
>     mre>    lease period for the lock.
> 
> I'm not sure that a new ADMIN_REVOKED error is necessary, but it does
> seem cleaner, so I don't have a problem with it.
> 
> cheers,
> mike


New Message Reply About this list Date view Thread view Subject view Author view Attachment view

This archive was generated by hypermail 2.1.2 : 03/04/05-01:50:16 AM Z CST