Claude
5db56e70a0
Six consumers on two partitions under a consumer group, a share group with the default share.acquire.mode and a share group with record_limit; the three acknowledgement types and a record that is released until Kafka stops redelivering it; a new share group starting at latest; the ordering a share group gives up; and Spring Kafka's share containers in implicit and manual acknowledgement modes. Tests run against a real broker (scripts/broker-up.sh) and write output/. Co-Authored-By: Claude Sonnet 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_01Lm655hdgDo5xsSLNomgdsy
share-groups
Companion code for Queues for Kafka: Share Groups with Spring for Apache Kafka 4 on ankurm.com.
What a Kafka share group (KIP-932) does differently from a consumer group, measured against a real broker: how many consumers can share one partition, what the three acknowledgement types do to a record, why a new share group sees nothing, and what ordering you give up. The same ideas are then run through Spring Kafka's share containers.
Versions
| Artifact | Version |
|---|---|
| Spring Boot | 4.1.1 (GA; 4.2.0-M2 is a milestone) |
| Spring Kafka | 4.1.1, managed by the Boot BOM (share consumer support is already in this GA release) |
| kafka-clients | 4.2.1, managed by the Boot BOM |
| Kafka broker | 4.3.1, started by scripts/broker-up.sh (share.version is finalized at level 1 by default; no feature flag was set) |
| JDK | 25 (LTS) |
Quickstart
scripts/broker-up.sh # downloads Kafka 4.3.1 (checksum-verified), starts KRaft on localhost:9092
scripts/run-all.sh # runs every test and rewrites output/*.txt
scripts/broker-down.sh
The tests do not start a broker themselves. Use -Dbroker=host:port if yours is elsewhere. Each test
creates its own uniquely named topic and group, so reruns never see earlier records.
What each test is
| Test | Role |
|---|---|
WorkSpreadTest |
6 consumers, a 2-partition topic, 60 records at 50 ms each: consumer group, share group with the default share.acquire.mode, and share group with record_limit |
AckTypesTest |
ACCEPT, RELEASE and REJECT through the raw KafkaShareConsumer, including a record that is released every time |
OffsetResetTest |
A new share group starts at the latest offset; share.auto.offset.reset=earliest is a group config set through the Admin API |
OrderingTest |
One key, one partition, three consumers of different speed: arrival order is not produce order |
SpringShareListenerTest |
ShareKafkaListenerContainerFactory in implicit and manual acknowledgement modes with @KafkaListener |
ShareConfig is the hand-written configuration: Spring Boot 4.1.1 has no auto-configuration for share
consumers.
Captured output (output/)
| File | What it captures |
|---|---|
01-work-spread.txt |
Which of six consumers received records under each of the three setups, with indicative timings |
02-ack-types.txt |
Every delivery of four records: accepted, released once, rejected, released until Kafka stops redelivering |
03-share-group-starts-at-latest.txt |
3 records produced first: 0 received by default, 3 received after switching the group to earliest |
04-no-ordering-guarantee.txt |
The order 40 same-key records were processed in, and the number of out-of-order neighbours |
05-spring-share-listeners.txt |
Listener threads under implicit mode, and per-record release, reject and acknowledge under manual mode |
Timings and the exact split of records between consumers differ from run to run. The shape of each result is stable; the milliseconds are not.
No docs/ chapter directory in this repository -- the intermediate and reference-depth material is in
accordion sections inside the WordPress post.