Thanks,  🙂
JooHyukKim (vince)

On Fri, Jul 7, 2023 at 11:59 AM Joo Hyuk Kim <beansk...@gmail.com> wrote:

> > When Pulsar undergoes a graceful shutdown, it unloads all topics, but
> there is no limit on the rate of unloading
>
> Is it possible that following code might already cover what we need here?
>
>
> https://github.com/apache/pulsar/blob/7636e8989f4d3fc24fce69a356d54e1c550945ed/pulsar-broker/src/main/java/org/apache/pulsar/broker/admin/impl/BrokersBase.java#L557
>
> On Fri, Jul 7, 2023 at 11:06 AM Dave Fisher <wave4d...@comcast.net> wrote:
>
>> Hi -
>>
>> Interesting. I’m curious to know what Heesung, Lari, and others think.
>>
>> Best,
>> Dave
>>
>> Sent from my iPhone
>>
>> > On Jul 6, 2023, at 7:01 PM, labuladong <labulad...@foxmail.com> wrote:
>> >
>> > Dear Apache Pulsar community,
>> >
>> >
>> > When Pulsar undergoes a graceful shutdown, it unloads all topics, but
>> there is no limit on the rate of unloading. As a result, during a rolling
>> upgrade, the broker cluster might generate a large volume of unload
>> requests in a short period, which could affect the service of the broker.
>> Implementing a limit on the unload rate would make the upgrade process
>> smoother.
>> >
>> >
>> > This issue has been previously raised in a pull request and a solution
>> was proposed by adding an unload interface with concurrency control:&nbsp;
>> >
>> >
>> > https://github.com/apache/pulsar/pull/14114
>> >
>> >
>> > However, the current broker graceful shutdown doesn't use this
>> interface. Hence, my proposal is pretty straightforward:
>> >
>> >
>> > 1. Add a dynamic config to the Pulsar broker to control the concurrency
>> of unloading.
>> >
>> >
>> > 2. Modify the following code to use the function with the concurrency
>> parameter for unloading.
>> >
>> >
>> >
>> https://github.com/apache/pulsar/blob/7636e8989f4d3fc24fce69a356d54e1c550945ed/pulsar-broker/src/main/java/org/apache/pulsar/broker/service/BrokerService.java#L790-L791
>> >
>> >
>> > I look forward to your thoughts on this.
>> >
>> >
>> >
>> > Best regards, donglai
>>
>>

Reply via email to