Re: [PATCH] SUNRPC: Fail over more quickly on connect errors

[Chuck Lever <chuck.lever@oracle.com>]

On 04/16/2010 06:29 PM, Trond Myklebust wrote:
> On Fri, 2010-04-16 at 16:47 -0400, Trond Myklebust wrote:
>> We should not allow soft tasks to wait for longer than the major timeout
>> period when waiting for a reconnect to occur.
>>
>> Signed-off-by: Trond Myklebust<Trond.Myklebust@netapp.com>
>> ---
>>   net/sunrpc/xprt.c |    2 +-
>>   1 files changed, 1 insertions(+), 1 deletions(-)
>>
>> diff --git a/net/sunrpc/xprt.c b/net/sunrpc/xprt.c
>> index c71d835..01449a3 100644
>> --- a/net/sunrpc/xprt.c
>> +++ b/net/sunrpc/xprt.c
>> @@ -710,7 +710,7 @@ void xprt_connect(struct rpc_task *task)
>>   		if (task->tk_rqstp)
>>   			task->tk_rqstp->rq_bytes_sent = 0;
>>
>> -		task->tk_timeout = xprt->connect_timeout;
>> +		task->tk_timeout = min(req->rq_timeout, xprt->connect_timeout);
>                                           ^^^ task->tk_rqstp->rq_timeout
>
> Apologies. I though I had tested that...
>
>> 		rpc_sleep_on(&xprt->pending, task, xprt_connect_status);
>>
>>   		if (test_bit(XPRT_CLOSING,&xprt->state))

I tested this series of patches with soft mounts, and RPC requests now 
fail, after the timeout period, if the client can't reconnect.

I also observed appropriate exponential back-off behavior as the client 
attempts to reconnect.  I would suggest one more patch to reduce the 
reestablish timeout maximum to 30 seconds.

For the series:

Reviewed-by: Chuck Lever <chuck.lever@oracle.com>

   and/or

Tested-by: Chuck Lever <chuck.lever@oracle.com>

-- 
chuck[dot]lever[at]oracle[dot]com