Skip to main content

ActiveMQ Artemis

{
  "url": "tcp://artemis:61616",
  "username": null,
  "password": null,
  "default_address": null,
  "default_routing_type": "multicast",
  "durable": true,
  "duplicate_detection": false,
  "consumer_window_size": 1048576,
  "reconnect_attempts": -1,
  "call_timeout_ms": 10000,
  "tls": { "enabled": false, "client_certs": [], "trusted_certs": [], "trust_all": false, "hostname_verification": true }
}

url is required: an Artemis core url, a list of brokers ((tcp://a:61616,tcp://b:61616)?ha=true), with its netty parameters; tls adds the TLS ones (sslEnabled, the key stores).

  • durable: messages written to disk by the broker, kept across a restart (the durable option per call).
  • duplicate_detection gives each message the source and the id of its event as duplicate id (_AMQ_DUPL_ID): the broker drops an event published twice.
  • consumer_window_size: what a consumer buffers ahead, in bytes (0 none, -1 no limit).
  • The client reconnects by itself reconnect_attempts times (-1: forever); while it does, a call waits call_timeout_ms at most.

Addresses and queues​

The extension speaks the core protocol, not JMS: it keeps Artemis' own model.

  • What is sent to a multicast address goes to every queue bound to it. A subscription without a group gets a temporary queue of its own (every message); with a group, the members share a durable queue named after the group, which keeps what is published while none of them is there. That is what a JMS shared durable subscription is: a JMS application subscribing under the same name shares it too.
  • What is sent to an anycast address goes to its queue, named after it, which its consumers share. A group on it is refused.
  • routing_type says which one an address is (queue instead of address means anycast); without it, the connection's default_routing_type. address::queue reads, or sends to, that very queue.
  • Addresses and queues are made as the broker's address settings allow (auto-create-*); otherwise the broker's answer is a 502.

Publishing​

Each send waits for the broker, so every result says what happened to its message (an address that does not exist, a security refusal), and gives its JMS message id. Options: priority (0 to 9), ttl (millis), delay or deliver_at (epoch millis, a scheduled delivery), durable, key: the group id of the message, whose messages go to one consumer at a time, in order.

Consuming​

  • selector: an Artemis filter, which the cloudEvents_* properties make useful (cloudEvents_type = 'order.created' AND priority > 4).
  • prefetch (100). at-most-once acks as the message leaves the subscription, at-least-once once it is handed over (on a multicast address, with a group: a temporary queue goes with its subscription).
  • A consumer that closes gives back what it did not ack as never delivered: only a consumer that fails counts its messages as delivered (artemisdeliverycount, and max-delivery-attempts before the dead letters). What never reached a client does not get closer to the dead letters.
  • A slow subscriber holds its own delivery thread, then the consumer window stops the broker: never a thread of a shared pool.

JMS applications read what the gateway publishes, and the other way around: the attributes go in cloudEvents_* properties (see CloudEvents), a JMS text message is read as a text/plain event.