NSQ-compatible messaging for Kubernetes

Durable queues.Share nothing.

NSQ-compatible messaging with one durable PVC per Kubernetes broker and no central coordinator.

Install with Helm Explore architecture

Independent broker nodes, each connected to its own storage disk — Independent brokers. Independent storage. One explicit failure boundary.

ACK boundary: Local fsync by default
Delivery: At least once
Topology: Share-nothing brokers
Protocol: NSQ V2 core

Every broker owns its queue.

Kubernetes handles placement and discovery. RustQueue keeps message ownership local, visible, and operationally direct.

Publish NSQ producer Existing client

Proxy or Gateway

Broker A Topic owner RWO PVC

Broker B Topic owner RWO PVC

Broker C Topic owner RWO PVC

Consumers discover every broker that owns a Topic, then connect directly over the NSQ protocol.

Read the Kubernetes runbook

A broker module persisting message capsules to its local storage platter

Choose the acknowledgement boundary

Durability is a setting, not a footnote.

The default waits for local persistence. Two explicit relaxed modes are available when the crash-loss window is an accepted tradeoff.

Response: After local segment fsync
Visibility: Only committed data is delivered

The production default and strongest local acknowledgement contract.

Keep the clients. Change the core.

Use official Go and Python NSQ clients across publish, consume, lookup, TLS, AUTH, and compression paths.

IDENTIFY AUTH SUB PUB MPUB DPUB RDY FIN REQ TOUCH NOP CLS

TLS and mTLS

Encrypted client paths with optional external AUTH.

Snappy and Deflate

Protocol compression without changing the client contract.

Lookup and fan-out

Discovery returns the real broker owners for each Topic.

Kodo profile

A default-off gateway and discovery compatibility boundary.

Performance only matters when the durability boundary is named.

Review the NSQ comparison contract

Strict and relaxed profiles are reported separately.

Consumer runs require complete unique delivery.

Unexpected duplicates invalidate a benchmark result.

Start on Kubernetes

One Helm command.

Point the chart at your image and an SSD-backed RWO StorageClass. RustQueue handles the rest of the topology.

shell

helm upgrade --install rustqueue deploy/helm/rustqueue \
  --namespace rustqueue --create-namespace \
  --set queue.image=registry.example/rustqueue:0.8.4 \
  --set queue.storageClassName=ssd-rwo