Hi Jun, Thanks for your comments. JR5: This is a very interesting one. I did have some nervousness about changing client ID without RPC version bumps, and you've brought up an excellent example.
I have removed the change to the client ID behavior from this KIP. While I still think it would be beneficial to stop sending the client ID strings repeatedly, it needs to be done properly. That was not the main purpose of this KIP, so I think it's best to pull back on this part. JR6: Done. JR7: I wonder why the authentication needs to have completed before the client instance ID would be cached. The client instance ID is generated by the client and the client should use the same value for all of its requests, and the broker will check that it has. JR8: As mentioned in my response to JR5, I've removed the client ID changes. JR9: Done. JR10: I've clarified that the checking only applies to requests that use the v2 header. JR11: I've added v1 for GetTelemetrySubscriptions and PushTelemetry in this KIP. JR12: Done. Thanks, Andrew On 2026/05/05 00:04:13 Jun Rao via dev wrote: > Hi, Andrew, > > Thanks for the updated KP. A few more comments. > > JR5. It seems that this design could break the clientId based quota if a > new client communicates with an old broker. The KIP proposes sending > ClientId only on the first request, then null for all subsequent requests, > without bumping up the request version. So, the old server will receive the > first request and ignore ClientId. The client doesn't know the server > ignores the clientId and sends a null clientId to an old broker for > subsequent requests. The broker will enforce quotas against a null > clientId, which is incorrect. > > JR6. Motivation: It would be useful to motivate the benefit of including > clientInstanceID on every request. For example, one immediate benefit is > that it can be included in request logging for troubleshooting. > > JR7. "The initial client instance ID for each connection will be cached by > the broker for checking" > It would be useful to clarify that the caching occurs only after > authentication completes. Some requests can be issued before authentication > completes. > > JR8. "This KIP also proposes a second change to the Kafka protocol. It > proposes sending the client ID only on the initial request on each > connection, and then sending a null client ID for all subsequent requests." > Similar to the above, it would be useful to clarify that the clientId is > included on the first request after authentication completes. > > JR9. GetTelemetrySubscriptionsRequest > "about": "Unique id for this client instance, must be set to 0 on the > first request." > Could we change the description above since clientInstanceId is no longer 0 > on first request? > > JR10. "Once a client has specified a client instance ID in the request > header of its first request, any subsequent requests which are missing the > client instance ID or which specify a different value for the client > instance ID will be rejected with error code INVALID_REQUEST." > Hmm, does this work for a new AdminClient sending an OffsetDeleteRequest > that still uses a v1 request header? > > JR11. Since ClientInstanceId exists in the header, should we remove > clientInstanceId from both GetTelemetrySubscriptions and PushTelemetry? > > JR12. Could we add the description in RequestHeader to explain the > difference between ClientId and ClientInstanceId? Historically, clientId is > meant for identifying an application, not an instance of an application. > > Jun > > On Thu, Apr 30, 2026 at 3:00 AM Andrew Schofield <[email protected]> > wrote: > > > Hi Apoorv, > > Thanks for your suggestion. > > > > AM1: That's a great idea. Done. > > > > Thanks, > > Andrew > > > > On 2026/04/27 12:24:12 Apoorv Mittal wrote: > > > Hi Andrew, > > > Thanks for the KIP. I have one question: > > > > > > AM1: With KIP-1082, the consumer now generates its own member ID at > > > construction time. With KIP-1313, the client will generate its own client > > > instance ID as a Uuid at construction time. Both are client-generated > > > UUIDs, immediately available at startup, and remain constant for the > > > process's lifetime. For a consumer (or share consumer), these two IDs > > serve > > > overlapping identification purposes at the same scope: identifying this > > > particular client incarnation. The ClientInstanceIds interface in Streams > > > already maintains a separate client instance ID per consumer per thread, > > so > > > there is no cardinality mismatch even in multi-threaded Streams > > > applications. Has unifying these been considered? The consumer could > > > generate one UUID at startup and use it both as its member ID in > > heartbeat > > > requests and as the client instance ID in request headers and telemetry. > > > > > > Regards, > > > Apoorv Mittal > > > > > > > > > On Fri, Apr 24, 2026 at 5:48 PM Jun Rao via dev <[email protected]> > > > wrote: > > > > > > > Hi, Andrew, > > > > > > > > Thanks for the reply. > > > > > > > > JR1. Caching the clientID from the first request in > > ChannelMetadataRegistry > > > > sounds good. We can then use it to fill the request header in > > subsequent > > > > requests. > > > > > > > > JR2.1 Ok. It's fine to include clientInstanceID in the request header. > > It > > > > would be useful to document that it's only set on the first request and > > > > cached in ChannelMetadataRegistry. > > > > > > > > Jun > > > > > > > > On Fri, Apr 24, 2026 at 6:28 AM Andrew Schofield < > > [email protected]> > > > > wrote: > > > > > > > > > Hi Jun, > > > > > Thanks for your comments. > > > > > > > > > > JR1: I was thinking of attaching it to some channel-based context, > > such > > > > as > > > > > the ChannelMetadataRegistry. I've not dug into the code in detail at > > this > > > > > point. However, as mentioned in the KIP, removing the client ID from > > all > > > > > except the initial request is perhaps overreach in this KIP. I wonder > > > > > whether actually this would be better handled in a major release. > > Let me > > > > > know your thoughts here. > > > > > > > > > > JR2.1: I did consider put it in the ApiVersionsRequest, but I do > > want it > > > > > in every request to help with problem diagnosis. Also, ApiVersions is > > > > not a > > > > > mandatory part of the protocol, so we can't rely on it being the > > first > > > > > request. > > > > > > > > > > JR2.2: Similar to JR1, I was thinking of hanging it off some > > > > channel-based > > > > > context. > > > > > > > > > > JR3: Indeed. I've added share consumer. > > > > > > > > > > JR4: Thinking about migration more, it seems to me that it is best > > not to > > > > > introduce new versions of the PushTelemetry and > > GetTelemetrySubscriptions > > > > > RPCs. For an AK 4.4 Java client, we know that the client will > > generate a > > > > > client instance ID before its initial connection. For any other > > client, > > > > we > > > > > do not know that. Consequently, even in the broker supports a newer > > > > version > > > > > of PushTelemetry which omits ClientInstanceId from the request > > schema, we > > > > > would still need to support the broker-side ClientInstanceId > > > > initialization > > > > > in which the response contains the ID created by the broker. I think > > that > > > > > leaving the slightly redundant ClientInstanceId in the KIP-714 RPCs > > is > > > > > best. I've changed the text accordingly. Let me know what you think. > > > > > > > > > > Thanks, > > > > > Andrew > > > > > > > > > > On 2026/04/23 19:04:37 Jun Rao via dev wrote: > > > > > > Hi, Andrew, > > > > > > > > > > > > Thanks for the KIP. A few comments. > > > > > > > > > > > > JR1. Currently, you can configure a client quota based on > > > > > > clientId. Currently, the broker extracts the clientId from the > > request > > > > > > header. If clientId is not included in every request, can we > > document > > > > > where > > > > > > the server will obtain it? > > > > > > > > > > > > JR2. client instance ID: > > > > > > JR2.1 Is it only sent on the first request in a connection? If so, > > have > > > > > we > > > > > > considered making it part of the ApiVersionRequest instead of the > > > > request > > > > > > header? > > > > > > JR2.2 It would be useful to include the client instance ID in > > request > > > > > > logging on the server side. Could we document how to make that > > > > available > > > > > > for each request the server processes? > > > > > > > > > > > > JR3. "The client instance ID will be calculated during the > > constructor > > > > of > > > > > > the Producer , Consumer and Admin". > > > > > > What about ShareConsumer? > > > > > > > > > > > > JR4. GetTelemetrySubscriptions: Could we document the changes in > > the > > > > > > response too? > > > > > > > > > > > > Jun > > > > > > > > > > > > > > > > > > On Fri, Apr 3, 2026 at 9:18 AM Andrew Schofield < > > > > > [email protected]> > > > > > > wrote: > > > > > > > > > > > > > Hi, > > > > > > > I would like to start the discussion on KIP-1313. This adds a > > unique > > > > > > > client instance ID to the request header of all Kafka protocol > > > > > requests to > > > > > > > give a unique identifier which can be used to correlate the > > requests > > > > > from > > > > > > > each client for the purposes of problem determination. > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > https://urldefense.com/v3/__https://cwiki.apache.org/confluence/display/KAFKA/KIP-1313*3A*Client*instance*ID*in*all*request*headers__;JSsrKysrKys!!Ayb5sqE7!txpES9zmXlCh8usguNsiJLRnSXvMtgWuM6R1rx8VxYYyi-sThQTmaOrq3ZfhArG25CjhHKuaLaJ63z_dzny8$ > > > > > > > > > > > > > > Thanks, > > > > > > > Andrew > > > > > > > > > > > > > > > > > > > > > > > > > > > >
