Before you start
A cluster needs a subnet of a VPC. It’s reached only from inside the VPC: there’s no public address.
Its first user’s password goes to a new Secret in the project, written as you, so creating
a cluster also needs secrets.secrets.create and secrets.versions.create in the project.
Create
- Open Hosted Queues and choose Create cluster.
- Pick Kafka or RabbitMQ, a version, and Single node or Replicated.
- Pick a size, storage and a subnet, and whether to serve the plaintext port.
- Review, then choose Create cluster.
ezgh hosted-queues clusters create events --engine kafka --engine-version 4.4 --mode replicated \
--instance-type gp-1.medium --storage-size-gib 100 --subnet sbn_2m9x0a7qpe4k --waitezgh hosted-queues clusters create jobs --engine rabbitmq --engine-version 4.3 --mode single \
--instance-type gs-1.small --storage-size-gib 20 --subnet sbn_2m9x0a7qpe4k --waitA create takes minutes, and --wait waits until the cluster is available. The cluster exists from
the start, creating, with its ID (hqc_…). Its endpoint is null until its names resolve.
The cluster comes with one user, app, with full access to its topics or vhosts, whose password is in a
Secret. See Users and permissions.
Settings
| Setting | |
|---|---|
engine |
kafka or rabbitmq. Fixed. |
engineVersion |
ezgh hosted-queues clusters engine-versions lists them. Upgrades move to a newer one. |
mode |
single or replicated. A single-node cluster can grow to replicated; a replicated one stays replicated. |
instanceType |
The size of each node, a gp-1 or gs-1 type; every node has the same one. hp-1 types take single-node clusters only. Can change later. |
storage |
Each node’s storage: sizeGiB, and optionally iops and throughputMBps. Can grow later. |
subnet |
The subnet it’s in. Fixed. |
allowedSourceCidrs |
Who can connect. |
plaintextPort |
Whether port 9092 (Kafka) or 5672 (RabbitMQ) also serves without TLS. See The plaintext port. |
settings |
Kafka: autoCreateTopics, defaultRetentionMs, defaultPartitions. See Topics. |
maintenance.windowUtc |
The maintenance window. |
deletionProtection |
Refuses deletes while on. |
Storage
Each node keeps everything it’s sent on its storage. A cluster is storage_low when more than 85% of
its storage is used, and the project’s notifications say so. Grow it while it
serves:
ezgh hosted-queues clusters update hqc_4k2m9x0a7qpe --storage-size-gib 200 --waitStorage only grows. When it’s full, RabbitMQ blocks publishers until there’s room, and Kafka stops
taking writes. Kafka topics keep messages for 7 days by default (defaultRetentionMs, or a topic’s own
retention.ms).
Who can connect
By default, any address in the cluster’s VPC can connect to its client ports. allowedSourceCidrs
narrows that to the ranges you list:
ezgh hosted-queues clusters update hqc_4k2m9x0a7qpe --allowed-source-cidr 10.0.1.0/24 \
--allowed-source-cidr 10.0.2.0/24 --waitAn instance’s own firewall also has to allow its outbound connection. A firewall
rule can name the cluster as its peer (hqc_4k2m9x0a7qpe) in place of its addresses.
Maintenance
Patch releases and other maintenance happen in the cluster’s maintenance window
(maintenance.windowUtc, such as sun:06:00-sun:07:00). A replicated cluster is maintained one node at
a time while the others serve. See Timeouts and reconnects.
Change the size
ezgh hosted-queues clusters update hqc_4k2m9x0a7qpe --instance-type gp-1.large --waitA replicated cluster moves one node at a time while the others serve.