On 2025/07/13 07:01:55 Yubiao Feng wrote:
> I just know what you suggested now:
> - Use `WRITE_BUFFER_WATER_MARK` instead of the new configuration
> - The event that changes channel state to inwritable will be the same as
> `!channel.isWritable() && exceed limitation`
>
> I consider it is a great solution, thanks, I will change the implementation
> later.
The solution should be based on Netty's standard way of controlling the write
buffer size from the beginning. Why would you want to postpone it?
> I reviewed the feature `ServerCnxThrottleTracker`, which was introduced in
> release `3.2.0`. I think we can use the tool class in a separate PR, the
> first PR can be cherry-picked into branch-3.0, not only branch-3.3 and
> branch-4.0. And the second PR can be cherry-picked into branch-3.3 and
> branch-4.0.
Since PRs target the master branch, ServerCnxThrottleTracker should be used in
the PR.
> BTW, the feature `ServerCnxThrottleTracker` has a bug that the
> rate limiter and the mechanism named max pending publishing bytes will
> affect each other. I will fix the issue in the second PR as well.
It's not a mistake. The throttling should be effective if any of the conditions
for throttling is present.
-Lari
