This page is meant as a template for writing a KIP. To create a KIP choose Tools->Copy on this page and modify with your content and replace the heading with the next KIP number and a description of your issue. Replace anything in italics with your own description.

Status

Current state:   [One of "Under Discussion", "Accepted", "Rejected"]

Discussion thread: here [Change the link from the KIP proposal email archive to your own email thread]

JIRA: here [Change the link from KAFKA-1 to your own ticket]

Please keep the discussion on the mailing list rather than commenting on the wiki (wiki discussions get unwieldy fast).

Motivation

While upgrading a Connect cluster we have to replace workers. The leader today waits for sometime before deciding to re-assign tasks which were running on the worker which was replaced. We would like to have a way for a worker to to relinquish tasks and thereby signal the leader that the tasks can be safely re-assigned.

Public Interfaces

Briefly list any new interfaces that will be introduced as part of this proposal or any existing interfaces that will be removed or changed. The purpose of this section is to concisely call out the public contract that will come along with this feature.

A public interface is any change to the following:

• Binary log format

• The network protocol and api behavior

• Any class in the public packages under clientsConfiguration, especially client configuration

  • org/apache/kafka/common/serialization

  • org/apache/kafka/common

  • org/apache/kafka/common/errors

  • org/apache/kafka/clients/producer

  • org/apache/kafka/clients/consumer (eventually, once stable)

• Monitoring

• Command line tools and arguments

• Anything else that will likely break existing users in some way when they upgrade

Proposed Changes

During the discussion for KIP-980, which proposed the creation of connectors in STOPPED state, there was a suggestion to also allow setting the initial offset for a connector in the connector creation API. The proposal was deemed valid (point no.4) but was deferred to a future KIP. This KIP proposes to implement that change.

This feature will be helpful to users who want to create a connector and start from a specific offset.

To do this today, user will have to

1. first create a connector in STOPPED state and check the response for errors
2. alter the offsets for said connector and check the response for errors
3. finally set the connector state to RUNNING.

Public Interfaces

"POST /connectors" REST API endpoint

A new optional field "initial_offsets"  will be added to the request body format for the POST /connectors endpoint. The value for this field has to adhere to format specified in KIP-875 else a 400 Bad Request response will be returned.

If the field is omitted in the request body, the current behaviour of creating connectors will be preserved. An example request body would look like:

CreateConnectorRequest

As mentioned in KIP-875: First-class offsets support in Kafka Connect#Altering/resettingoffsets(response) , the response message (`initial_offsets_response`) for successfully setting the initial_offsets will depend on whether the connector has implemented alterOffsets()  method.

If the connector has implemented alterOffsets  and everything has succeeded, the HTTP status will be 200 and the create-connector response body will be:

response (definite success)

If the connector has not implemented alterOffsets  but everything else has succeeded, the HTTP status will be 200 and the create-connector response body will be:
response (possible success)

This method will be supported in both the distributed mode as well as the standalone mode.

KIP-980 had introduced an option to pass JSON config to bin/connect-standalone.sh. This new field will be supported there as well.

Proposed Changes

When Connect runtime receives a request with initial_offsets to create a connector via POST /connectors, it will perform the following steps

1. Validate connector configs
2. Validate the offset using the same checks performed while altering connector offsets (PATCH /$connector/offsets ) as specified in KIP-875 (We will skip validations which check if the connector exists or is in STOPPED state)
3. Wipe all existing offsets for the connector (i.e we do not want to merge existing offsets and "initial_offsets")
4. Set the offset of the connector to the value of “initial_offsets”

5. Write connector config (with whatever initial state is specified in the request, or the default if none is specified)
   
   

If Connect runtime encounters an error in any of these steps, it will cleanup (if required) and return an error response mentioning the step at which the failure was encountered.

If step no. 4  fails, the existing offsets of the connector would've already been wiped in step no. 3.

If step no. 5  fails, the initial_offset set in step 4 will be wiped as part of cleanup.

In either failure case, the offset of the connector, set prior to invoking this API,  will be unrecoverable.

Minor details

Validate offset before connector creation

There is a slight chance of a race condition if two connector creation requests with same name , with and without initial_offsets , are received. The request without offset might create the connector while the other one validates and writes the offset. We still would like to validate the offset and config first - so that there is lesser chance of a cleanup while creating the connector and setting it's offset.

Remanent offsets from connectors with same name

There could be cases when a previous instance of connector with same name has offsets left in the offsets topic. The offsets sent along with the create connector request will be written after wiping out the previous offsets. This is done to ensure a clean slate for the connector, thereby avoiding any surprises laterDescribe the new thing you want to do in appropriate detail. This may be fairly extensive and have large subsections of its own. Or it may be a few sentences. Use judgement based on the scope of the change.

Compatibility, Deprecation, and Migration Plan

• What impact (if any) will there be on existing users?
• If we are changing behavior how will we phase out the older behavior?
• If we need special migration tools, describe them here.
• When will we remove the existing behavior?

Test Plan

Describe in few sentences how the KIP will be tested. We are mostly interested in system tests (since unit-tests are specific to implementation details). How will we know that the implementation works as expected? How will we know nothing broke?

Rejected Alternatives

Since “initial_offsets” is an optional field and when unspecified will default to current behaviour, we do not foresee a compatibility issue while downgrading the Connect cluster. This new feature will reuse existing logic for storing offsets. Hence the connector will be indistinguishable from one which started with no initial offset specified and in due course committed an offset to the store.

Test Plan

The following scenarios will be tested using Unit, Integration or system tests

• Create connector with initial_offsets in standalone mode
• Create connector with initial_state and initial_offsets in standalone mode
• Create connector with initial_offsets in distributed mode
• Create connector with initial_state and initial_offsets in distributed mode
• Failure while creating connector with initial_offsets will clean up and return error response
• Standalone mode cli accepts initial_offsets in JSON config file

Rejected Alternatives

Restrict initial_offsets usage to cases where initial_state == STOPPED

We could enforce intial_offset usage only to cases when a connector is created in STOPPED state. This would’ve reduced one point of failure while performing the action. However,  we feel that this will impact the usability of this new API and will not offer any major benefit from the steps which a user has to run currently , namely, creating a connector in STOPPED state , altering the offset and then setting the state to RUNNINGIf there are alternative ways of accomplishing the same thing, what were they? The purpose of this section is to motivate why the design is the way it is and not some other way.