Thanks Tom,
In my original KIP-170 I mentioned that the method
public Map<String, Integer> topicsPartitionCount();
was just a starting point for a general purpose ClusterState
as it happened to be exactly the info we needed for our policy
implementation :-)
it definitely doesn't feel general purpose enough.
what about
interface ClusterState {
public TopicState topicState(String topicName);
public Set<String> topics();
}
I think that the implementation of ClusterState that the server will pass
to the policy.validate method
would just lazily tap into MetadataCache. No need for big upfront
allocations.
ciao,
Edo
--------------------------------------------------
Edoardo Comar
IBM Message Hub
IBM UK Ltd, Hursley Park, SO21 2JN
From: Tom Bentley <t.j.bent...@gmail.com>
To: dev@kafka.apache.org
Date: 26/09/2017 17:39
Subject: Re: [DISCUSS] KIP-201: Rationalising Policy interfaces
Hi Edoardo,
what about a single method in ClusterState
>
> interface ClusterState {
> public Map<String,TopicState> topicsState();
>
> }
>
> which could return a read-only snapshot of the cluster metadata ?
>
Sure that would work too. A concern with that is that we end up allocating
a potentially rather large amount for the Map and the collections present
in the TopicStates in order to provide the snapshot. The caller might only
be interested in one item from the TopicState for one topic in the map.
Accessing this information via methods means the caller only pays for what
they use.
Cheers,
Tom
Unless stated otherwise above:
IBM United Kingdom Limited - Registered in England and Wales with number
741598.
Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU
