avoid introducing such new configurations unless we
have to.
Best
Yun Tang
From: Yuan Mei
Sent: Monday, July 11, 2022 11:53
To: dev
Cc: ro...@apache.org
Subject: Re: Re: Re: [DISCUSS ] Make state.backend.incremental as true by
default
Hey Lihe Ma,
1. I do not
Hey Lihe Ma,
1. I do not think making different state backends with different default
setups (configs) is a bad idea. I have some similar ideas as Zakelly,
Hangxiang and you mentioned above.
Think about this: even though the incremental checkpoint for HashMap
Statebackend is released in 1.16, it m
The suggestion of Zakelly (Using noDefaultValue to make different
StateBackend have its own default value) makes sense to me.
I haven't gotten a better idea about it.
Maybe @ro...@apache.org who is the owner of incremental
checkpoint support of HashMapStateBackend could share more ideas about it.
Hey @Lihe Ma,
Thanks and I appreciate your feedback!
*What do I think of introducing a new state backend?*
I am hesitant to follow this way of introducing a new state backend to
implement `HashMapStateBackend with incremental checkpoint enabled`,
because:
- It causes more confusion for users: why
Strongly +1
Best,
LuNing Wang
Zakelly Lan 于2022年6月17日周五 00:15写道：
> Thanks for bringing this up.
> I'm +1 on enabling the incremental checkpoint by default on RocksDB. But I
> also agree with Yuan about not enabling this on newly implemented
> incremental checkpoint for hashmap statebackend.
>
Thanks for bringing this up.
I'm +1 on enabling the incremental checkpoint by default on RocksDB. But I
also agree with Yuan about not enabling this on newly implemented
incremental checkpoint for hashmap statebackend.
I am wondering can we make it behave differently for different state
backends w
