Valentin,
I understand your point and it makes sense to me. To summarize discussion
I'll do the following. Leave timeout, use LT for spammed stack traces and
file separate ticket for serialization issue in calculations as suggested
Semyon.
On Wed, Nov 16, 2016 at 1:11 AM, Valentin Kulichenko <
Dmitry,
I absolutely agree with this and I'm not saying that timeout solves any of
these problems. However, it provides a bit of safety for such cases (client
continues working instead of hanging forever) and also allows not too wait
for the service too long which can be also useful sometimes
Valentin, as I understand, this situation is abnormal, and if user
thread hangs on getting service proxy, it's worth to check logs for
failure instead of making workaround.
On 15.11.2016 19:47, Valentin Kulichenko wrote:
I would still add a timeout there. In my view, it makes sense to have
I would still add a timeout there. In my view, it makes sense to have such
option. Currently user thread can block indefinitely or loop in while(true)
forever.
-Val
On Tue, Nov 15, 2016 at 7:10 AM, Dmitriy Karachentsev <
dkarachent...@gridgain.com> wrote:
> Perfect, thanks!
>
> On Tue, Nov 15,
Perfect, thanks!
On Tue, Nov 15, 2016 at 5:56 PM, Vladimir Ozerov
wrote:
> To avoid log pollution we usually use LT class (alias for GridLogThrottle).
> Please check if it can help you.
>
> On Tue, Nov 15, 2016 at 5:48 PM, Dmitriy Karachentsev <
>
To avoid log pollution we usually use LT class (alias for GridLogThrottle).
Please check if it can help you.
On Tue, Nov 15, 2016 at 5:48 PM, Dmitriy Karachentsev <
dkarachent...@gridgain.com> wrote:
> Vladimir, thanks for your reply!
>
> What you suggest definitely makes sense and it looks more
Vladimir, thanks for your reply!
What you suggest definitely makes sense and it looks more reasonable
solution.
But there remains other thing. The second issue, that this solution solves,
is prevent log pollution with GridServiceNotFoundException. The reason why
it happens is because
Also we implemented the same thing for platforms some time ago. In short,
job result processing was implemented as follows (pseudocode):
// Execute.
Object res;
try {
res = job.run();
}
catch (Exception e) {
res = e
}
// Serialize result.
try {
SERIALIZE(res);
}
catch (Exception e)
Dmitriy,
Shouldn't we report serialization problem properly instead? We already had
a problem when node hanged during job execution in case it was impossible
to deserialize the job on receiver side. It was resolved properly - we
caught exception on receiver side and reported it back sender side.
Hi Igniters!
I'd like to modify our public API and add IgniteServices.serviceProxy()
method with timeout argument as part of the task
https://issues.apache.org/jira/browse/IGNITE-3862
In short, without timeout, in case of serialization error, or so, service
acquirement may hang and infinitely
10 matches
Mail list logo