Hi all,
Since there are no more suggestions for more than 72 hours, I have updated
the PIP https://github.com/apache/pulsar/issues/13761 according to the
discussion.
And then will send a vote thread soon.
Aloys Zhang 于2022年2月22日周二 10:38写道：
> >Should this be "positions"? We are going to split on
>Should this be "positions"? We are going to split one bundle into
multi-bundles,
in most cases, bundle number will be position number + 1, right?
Sure, it should be “positions” or “positionList”
PengHui Li 于2022年2月21日周一 20:50写道：
> > And the (anti-)affinity way needs more discussion or maybe we
> And the (anti-)affinity way needs more discussion or maybe we can
introduce
a new PIP for it.
+1
Thanks,
Penghui
On Mon, Feb 21, 2022 at 6:47 PM Haiting Jiang
wrote:
> > 2. calculate the position to split this bundle(also need a new API)
> Should this be "positions"? We are going to split on
> 2. calculate the position to split this bundle(also need a new API)
Should this be "positions"? We are going to split one bundle into
multi-bundles,
in most cases, bundle number will be position number + 1, right?
> And the (anti-)affinity way needs more discussion or maybe we can introduce
Hi penghui and haiting,
I try to figure out how the (anti-)affinity works.
> if I understand correctly, it looks like if we have a partitioned topic
with 10
> partitions under a namespace with 16 bundles, if applies the
anti-affinity policy,
> partition-0 map to bundle 0, partition-1 map to bundl
Hi, penghui
> The new API does not necessarily have to query by topic one by one,
we have listed all the "topic -> position" of a bundle?
I see. After we got all the positions of the topics we want to split in a
bundle, it's quite easy for us to decide how to it.
Haiting Jiang 于2022年2月20日周日 12
> Do you have an example for affinity? I don't fully understand how this is
> used
> in practice.
IMO, this affinity serves the purpose of isolating an abnormal topic to some
spare
brokers. These brokers host these kind of topics only. Here are some cases :
1. A topic may have unexpected short
Hi Haiting,
> I think this approach have more potential with abnormal topic isolation.
If we can introduce
some kind of bundle isolation strategy, (like broker-bundle affinity and
anti-affinity mechanism), we can easily isolate some unexpected traffic to
some empty brokers.
IMO, this would improve
Hi Aloys,
> Do you mean that
1. First, add a new API, maybe `getHashPositioin`, to get the hash
position in a bundle
2. Then use this position to split the overloaded bundle
If so, when we split a bundle with multi partitions of a topic, we need to
call the `getHashPositioin` multi times to get
Hi Aloys,
+1 for this great PIP.
> The Admin CLI `bin/pulsar-admin namespaces split-bundle -b ${bundle_range}`
> will add a new parameter "--topic" or "-t" for `outstanding topic` name.
Do we have limitation on this "topic" parameter. Can this be a partitioned
topic?
If so, will this new algori
Hi Penghui,
> I am considering if we can add a boundary param for split bundle API,
> The boundary must be between the start and the end of the bundle.
> looks like the followings:
>
> ```java
> void splitNamespaceBundle(String namespace, String bundle, boolean
> unloadSplitBundles,
>
Hi Aloys,
Thanks for the great proposal.
I am considering if we can add a boundary param for split bundle API,
The boundary must be between the start and the end of the bundle.
looks like the followings:
```java
void splitNamespaceBundle(String namespace, String bundle, boolean
unloadSplitBundle
Hi Pulsar Community,
This is a PIP discussion on how to support split partitions belonging to
specified topics in a bundle.
The issue can be found: https://github.com/apache/pulsar/issues/13761
I copy the content here for convenience, any suggestions are welcome and
appreciated.
## Motivation
13 matches
Mail list logo
