Thanks for the KIP. Overall this does sound useful.
About Almog's comments
(A1) -- I don't think that `topics.internal.require.explicit.naming`
would be a good name, as we use `topic.` prefix for actual topic
configs. Thus, KS would pickup `internal.require.explicit.naming` and
try to apply is as topic config what would either crash or log a
spurious WARN what would be annoying?
(A2) -- We would need to check the code if such a strategy would work as
expected? If users pass in a name for a previously un-named topic, we
might get an cascading index shift which might be undesired (and could
"break" processor level metrics)?
But I am also not sure if we actually need a migration path? If might be
ok if this new feature only work for new deployments? (Or use can re-set
their application.)
MJS-1: I have another question though about the risk such a config
implies? Given that configs are often managed from "outside", one could
easily break an exiting application which does not have this config
enable, by enable the config. Of course, we already have similar config
which are equally dangerous; however, most people don't like that one
need to pass in the config into `StreamsBuilder.build(configs)` method
anyway, so maybe we could make a first step to get rid of this via this KIP?
Thus, I am wondering if a config is actually the right way to go? Should
we instead make it feature of `StreamsBuilder` that one can enable
programmatically? Not 100% how we would do this, but maybe we could use
a builder pattern (to allow us to add similar thing in the future), and
deprecate the current constructor of `StreamsBuilder`?
// new way to build a `StreamsBuilder`
StreamsBuilder builder = StreamsBuilder.build();
// if we don't extend the scope of this KIP, we might also need:
StreamsBuilder builder = StreamsBuilder.build(Properties);
Or we do extend this KIP and deprecate existing config (like topology
optimization) if favor of the new builder pattern:
// enable the new feature
StreamsBuilder builder = StreamsBuilder.requireExplicitNaming().build();
// or
StreamsBuilder builder = StreamsBuilder.disableNameGeneration().build();
Thoughts? I am not 100% sure if this is a good idea or not, but thought
it cannot hurt to throw it out.
-Matthias
On 11/18/24 12:57 PM, Almog Gavra wrote:
Hi Sebastien,
Thanks for the KIP! In general, I'm a fan of giving users the tools they
need to protect their organization so I'm supportive of this proposal. A
few nits and comments:
A1. [nit] consider 'topics.internal.require.explicit.naming' so that (a) we
can group anything else we introduce for "topics.internal" with the same
prefix and (b) it's not a double negative (don't disallow is the default,
instead of don't require).
A2. I think we can improve on the implementation by making it typesafe
instead of checking whether the topic matches some pattern. I think a
migration path for users that want to turn this flag on, but already have
some auto-generated names, is to manually specify the auto-generated names
for preexisting topics. This would enforce future topics naming, but not
penalize them for having used auto generated names in the past. This makes
the implementation a little more challenging, but I think it's worthwhile.
Cheers,
Almog
On Fri, Nov 15, 2024 at 12:20 AM Sebastien Viale <
sebastien.vi...@michelin.com> wrote:
Hi Everyone,
I would like to start a discussion on KIP-1111: Enforcing Explicit Naming
for Kafka Streams Internal Topics<
https://cwiki.apache.org/confluence/display/KAFKA/KIP-1111%3A+Enforcing+Explicit+Naming+for+Kafka+Streams+Internal+Topics
This proposal aims to add a configuration that prevents a Kafka Streams
application from starting if any of its internal topics have auto-generated
names.
Regards,
Sébastien
