[OpenAFS] Re: OpenAFS callbacks

Simon Wilkinson sxw@inf.ed.ac.uk
Mon, 11 Apr 2011 21:21:59 +0100

On 11 Apr 2011, at 21:11, Derrick Brashear wrote:

>> And also, to be clear: I was just wondering about the technical =
>> I'm pretty sure I wouldn't even use this functionality if we had it, =
>> I haven't heard a lot of clamoring for it (the search indexing =
project I
>> consider to be separate).
> users of beegle/famd would love us long time.

The adminstrators of their fileservers might not, however.=20

One of the big considerations with turning on any kind of pinning, be it =
for this use case, or for disconnected, is how you avoid creating =
callback storms at the fileserver. The problem here is when (number of =
pinned files) * (number of clients) > (number of callbacks), and you end =
up effectively thrashing the callback list. There needs to be some way =
for a client to tell that it has encountered that situation, and to =
degrade gracefully. However, degrading is going to mean refreshing fewer =
callbacks, or cycling them slower, which means that the client is much =
more likely to miss changes, or to not be up to date when going offline.

Again, extended callbacks helps hugely with this, because we can tell if =
a fileserver has dropped our callback because it's run out of internal =
state, or if there has actually been a change.