I agree this behavior is strange and causes NPEs, saw it myself in tests.
Probably this is required for correct re-initialization of injected
resources after node restart.
Looks like clearing can be removed if node starts successfully after
stopping. Could you write a test to check ?
Another possible fix - remove resource injection for build-in SPIs. Do we
really need it at all ?
пт, 25 окт. 2019 г. в 17:58, Дмитрий Сорокин <
[hidden email]>:
> Hi folks!
>
> At the moment every resources injected by GridResourceProcessor on node
> start, also is clearing on node stop, but before the stopping of
> executors. That behavior on node stopping sometimes causes NPEs at calls
> which performed in threads of StripedExecutor, for example, see [1].
>
> Does anybody remember the motivation of cleaning injected resources?
>
> Also, can anybody to tell what is bad things can happen if we won't clean
> (nullify) injected logger resources?
>
> [1]
https://issues.apache.org/jira/browse/IGNITE-1606>
> —
> Dmitriy Sorokin
>
--
Best regards,
Alexei Scherbakov