Meetup summary, 7 October 2026

[justin-laserdata]

Meetup summary, 7 October 2026

The first Apache Iggy community meetup ran on 7 October 2026. Justin Mclean (@justinmclean) hosted and Kranti Parisa (@kparisa) co-hosted. Piotr Gankiewicz (@spetz) gave a talk on clustering in 0.9.0, followed by questions from the room and the chat.

Piotr opened with the project's path from the first commit to Incubator entry, graduation as a top-level project two months ago, and the 0.9.0 release a month later. That release retired the old server binary in favour of the new one and brought the SDKs up to the clustered server, with over half of its contributors being first-timers.

Clustering runs from the single-server binary and is enabled in the cluster section of the config file, with a node list. The consensus layer is Viewstamped Replication Revisited, chosen over Raft for its deterministic leader election, which also makes testing easier. Each partition is replicated as its own group. All writes go through a single primary today. A multi-leader design, where each node leads a set of partitions and backs up the rest, is in progress and could land in the coming weeks.

Testing uses a deterministic simulation binary that controls the seed and injects delays, crashes, packet loss and network splits, so any failure can be replayed from its seed. Benchmarks on bench.iggy.apache.org show the test machines' network as the cap.

Questions and answers

The first item was a request rather than a question: an official cluster container for local development. Kranti suggested a three-node cluster that starts with no config editing. Piotr agreed it is easy to do.

Asked what production users want most, Piotr said cluster reliability and correctness come first, then tiered storage, which already has an RFC. Kranti added Kafka API compatibility, so existing Kafka code can be pointed at Iggy for A/B tests, and noted the Kafka proxy work by Krishna Vishal and Raveendra Yerraguntla (ryerraguntla).

Danil (@mmxa) asked where a newer contributor can help besides connectors. Kranti said a roadmap document is coming, and the core is changing too fast for newcomers to take on big engine features for a few months. He pointed to connectors, documentation, tests and trying to break clustering. Piotr suggested checking the other SDKs for feature parity with Rust, such as missing transports, builders and async iterators, and running benchmarks against them. Justin added profiling SDKs. Kranti and Hubert Gruszecki (@hubcio) both encouraged code reviews, which anyone can do without permissions and which are the fastest path to committer.

On dead letter queues, Piotr said they are done on the application side today by rerouting to a dedicated topic. A server-side feature that marks messages as unprocessable would need an internal key-value registry, and an RFC is welcome.

On delivery guarantees to consumers, Grzegorz Koszyk said to disable auto-commit and store offsets manually when durability matters, since auto-commit marks a message delivered as soon as it is sent. Piotr noted auto-commit saves a network call. Justin said he added examples covering this to the docs last week.

Follow-ups from the call

A cluster Docker Compose or image for local development was taken on board. Kranti said a roadmap document will be published soon. Speakers are wanted for the next meetup. Questions can go to Discord, the mailing list or GitHub.