Re: Heads up on HBASE-28210

2023-11-25 Thread Duo Zhang
After HBASE-28210, we will lose some performance for large procedures
which have a lot of sub procedures, like SCP.

So I filed HBASE-28212 to get some performance back for procedures
which do not support rollback, like SCP.

The PR is ready, please take a look if you have interest.

https://github.com/apache/hbase/pull/5538

Thanks.

张铎(Duo Zhang)  于2023年11月17日周五 23:45写道:
>
> Found this problem when implementing HBASE-28199.
>
> The reason why the code change in HBASE-28199 can expose the problem
> is that, we add the crash code between adding the procedure to root
> procedure state and updating store, so we can reproduce the problem
> that later procedure can be persistent earlier reliably, and then
> cause trouble while loading procedure.
>
> It is not very easy to reproduce in the normal world but if it
> happens, it will be a very big problem and can only be recovered
> manually.
>
> I think this is a critical problem which we should fix ASAP.
>
> Thanks.


Heads up on HBASE-28210

2023-11-17 Thread Duo Zhang
Found this problem when implementing HBASE-28199.

The reason why the code change in HBASE-28199 can expose the problem
is that, we add the crash code between adding the procedure to root
procedure state and updating store, so we can reproduce the problem
that later procedure can be persistent earlier reliably, and then
cause trouble while loading procedure.

It is not very easy to reproduce in the normal world but if it
happens, it will be a very big problem and can only be recovered
manually.

I think this is a critical problem which we should fix ASAP.

Thanks.