This would be roughly the same as today's approach:
* The limit per partition is X/s
* You are given N partitions, so the max quota is X*N
Which would be easy to convert in the total limit allowed is Y and you
have N topic, therefore the per-topic quota is Y/N
The problem is that there will be
Yes, this is the drawback if do static partitioning.
Thanks,
Penghui
Kaushik Ghosh 于2021年3月3日周三 上午9:21写道：
> Wouldn't such a static partitioning approach have the drawback that in a
> pathological case, all the namespaces associated with a certain
> namespace-bundle may be inactive (and not use
Wouldn't such a static partitioning approach have the drawback that in a
pathological case, all the namespaces associated with a certain
namespace-bundle may be inactive (and not use their quota) while other
namespaces are over-active and being restricted?
Thanks,
Kaushik
On Tue, Mar 2, 2021 at 5
The approach is sharing the quotas between brokers through an internal
topic, for example, if the rate limit is 100msgs/s and the current rate is
50 msgs/s
. If share quotas between brokers, we still need to achieve the policy to
assign the remaining quotas to multiple brokers.
How about assigning
https://github.com/apache/pulsar/wiki/PIP-82%3A-Tenant-and-namespace-level-rate-limiting
=
* **Status**: Proposal
* **Authors**: Bharani Chadalavada, Kaushik Ghosh, Ravi Vaidyanathan,
Matteo Merli
* **Pull Request**:
* **Mailing List discussion**:
* **Release**:
## Motivation
Curre
