[OpenAFS-devel] Re: idle dead timeout processing in clients

[Simon Wilkinson]

On 30 Nov 2011, at 18:25, Andrew Deason wrote:

> The 1.4 unix client will get an RX_CALL_TIMEOUT error on an idle call,
> will it not? Why would we ever use RX_CALL_DEAD ?

The idle dead code isn't in any shipping versions of 1.4. Current 1.4 =
clients won't get RX_CALL_TIMEOUT, or RX_CALL_DEAD.

> Meaning... a call against a hanging fileserver will block forever? =
That
> does not sound very acceptable; it sounds like a 1.4 regression.

It's not a regression against _shipping_ 1.4. I suspect that we'll need =
to reconsider the code that's on the 1.4 branch before we ship 1.4.16. =
Basically, the current client-side idle dead code is unusable, because =
it causes connection timeouts where a file server is overloaded and, =
more critically, when a fileserver has to break a callback against a =
client that has disappeared.=20

S.