Skip to content

Feature: Allow unordered processing of messages without a key in 'KEY' ordering mode #183

@rjoly-qlik

Description

@rjoly-qlik

In our use case, we have a vast majority of messages that are not order-dependent but we have a small pocket of messages that are. Our intention was to use the NULL key for the messages that do not have an order requirement and use a proper key value on the few occasions where the order matters. Rough guess is that 95% of our messages will use the NULL key. In order to ensure that the keyed messages are processed in order, we have no choice but to use the KEY ordering method but this has the very adverse effect of essentially serializing the remaining 95% which negates the benefits of the PC. This is because PC considers the NULL key as an actual key value and will process every message that share that NULL key in the order they appear in the partition.

It would be nice if the PC offered an option (nullKeysUnordered?) that would make it process all un-keyed messages in parallel and only order the messages that do have a key.

The issue was discussed with @astubbs in this Slack thread.

Metadata

Metadata

Assignees

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions