feat(consumer): scaffold KIP-932 share consumer surface#838
Draft
j7nw4r wants to merge 13 commits into
Draft
Conversation
Mirrors the KIP's Java surface as a parallel Rust API gated behind a new `kip-932` cargo feature, with stubbed FFI returning a new `KafkaError::Unsupported` variant until librdkafka exposes the public C API (tracking confluentinc/librdkafka#5441).
Clarifies the working mode: the plan ships in a draft PR on the fork and evolves in-tree until librdkafka exposes the public KIP-932 C API.
Introduces the gated `consumer::share` module that will host the KIP-932 share consumer surface. Submodules (`acknowledge`, `config`, `consumer`, `context`, `records`) are declared but empty; subsequent commits fill them in. The `kip-932` cargo feature is off by default and added to the docs.rs feature set so the surface renders on docs.rs once populated.
Introduces the value types that the share consumer surface will consume: * `AcknowledgeType` (Accept, Release, Reject) with `Display` / `FromStr` round trip and case-insensitive parsing. * `AcknowledgementMode`, `AutoOffsetReset`, `IsolationLevel` for the KIP-932 enum-valued configuration properties. * `ShareConsumerConfig` builder layered on top of `ClientConfig`, emitting canonical librdkafka property keys (`share.acknowledgement.mode`, `group.share.delivery.attempt.limit`, etc.) via `into_client_config`. * `AcknowledgementCommitResult` for surfacing async acknowledgement outcomes through the share consumer context. Defaults follow KIP-932 (implicit acknowledgement, latest offset reset, read_uncommitted isolation, 5 delivery attempts, 30s record lock).
Surfaces a non-retriable error for API methods that exist in rust-rdkafka but are not yet backed by the underlying librdkafka build. Used by the KIP-932 share consumer scaffolding until librdkafka exposes the public share consumer C API. `KafkaError` is already `#[non_exhaustive]`, so callers matching on it already needed a fallback arm; no existing match becomes incomplete.
Introduces the central pieces of the KIP-932 surface: * `ShareConsumerContext` mirrors `ConsumerContext` with a single `acknowledgement_commit` callback for async ack results. * `ShareConsumer` trait spells out the runtime methods: subscribe / unsubscribe / subscription, poll, acknowledge / acknowledge_offset, commit_sync / commit_async, wakeup, close. * `BaseShareConsumer` is the stub backing the trait. Construction parses `group.id` from `ClientConfig` (failure surfaces as the usual `ClientCreation` error). Every runtime method returns `KafkaError::Unsupported`; `wakeup` and `close` are intentionally safe no-ops so application shutdown paths still work. * `ShareConsumerRecords` / `ShareRecord` are borrowed wrappers ready for librdkafka's eventual fetch surface; the stub never yields a non-empty batch. * All new types are re-exported from `crate::consumer::share` and, under the `kip-932` feature, from `crate::consumer`.
Locks in the public shape so the eventual FFI swap is mechanical: * `ShareConsumerConfig` emits the canonical librdkafka property keys (`share.acknowledgement.mode`, `group.share.*`, ...). * `BaseShareConsumer` constructs via both `from_config` and `from_config_and_context`. * Every runtime method returns `KafkaError::Unsupported` with a reason that names KIP-932. * `close` and `wakeup` are safe no-ops so application shutdown paths remain valid. * `AcknowledgeType` survives `to_string` / `FromStr` round-trips. The test does not require a broker; the share consumer is a stub until librdkafka exposes the public share consumer C API.
* `lint` runs clippy and the share-specific tests with `--features kip-932` so the gated surface is lint-clean and exercised on every PR. * `check` matrix gains an Ubuntu entry that builds with the `kip-932` feature alone, so the feature flag is verified across the default toolchain.
* README and crate-level rustdoc list the new `kip-932` feature in the Features bullets, with the librdkafka tracking link. * changelog.md gains an Unreleased entry covering both the share consumer scaffolding and the new `KafkaError::Unsupported` variant. * Tighten the share module rustdoc so it builds clean under `-D rustdoc::redundant_explicit_links`.
clippy::needless_lifetimes on Rust 1.85+ flags impl<'a> ShareRecord<'a> because the impl block doesn't use 'a beyond the type parameter. Switch to impl ShareRecord<'_>. Caught by the upstream CI lint job (Rust 1.85).
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What this lands
Adds the public Rust surface for KIP-932 behind a new
kip-932cargo feature, off by default. Types compile and unit tests lock in the shape. Every runtime method returns a newKafkaError::Unsupporteduntil librdkafka exposes the public share consumer C API; tracking at confluentinc/librdkafka#5441.What KIP-932 changes
Kafka consumer groups normally assign each partition to one consumer at a time. Parallelism caps at the partition count, and a slow consumer blocks its partition. Share groups (this KIP) let multiple consumers cooperatively pull from the same partition. The broker hands out per-record locks with a timeout and tracks delivery counts. The consumer explicitly accepts each record (success), releases it (transient failure, broker redelivers), or rejects it (poison message, broker archives it).
Net effect: queue-style consumption on top of the existing log. At-least-once, no strict per-partition ordering, but you can scale consumers past the partition count without re-sharding.
Status
Draft, not ready to merge. The plan lives at
PLAN.md(first commit) and will iterate in-tree as the upstream API settles. Once a public C surface lands, swapping the stubs for real FFI is mechanical.Reviewer note
KafkaErrorgains a newUnsupported(&'static str)variant. It's#[non_exhaustive], so wildcard arms still work.