Hazelcast

REST API supports cluster group name + password (legacy) or token-based auth (Hazelcast 5.x). Native clients authenticate with username/password or TLS client certificates. Hazelcast Cloud uses API key authentication from the Cloud Console. RBAC available in Hazelcast Enterprise for fine-grained permissions (read/write per map, queue, topic). TLS enforced on Hazelcast Cloud; operator-configured for self-hosted. Cluster security config via hazelcast.yaml.

Hazelcast provides a distributed in-memory data grid: distributed maps (key-value cache with TTL, eviction, persistence), queues, topics (pub/sub), locks, and atomic counters — all accessible from any cluster node via the client API or REST. For agents, the most useful primitives are distributed maps (shared cache between agent instances), queues (task distribution), and topics (event broadcast). Hazelcast Cloud is the managed offering; self-hosted via Docker or Kubernetes. Embeds well into Java services; multi-language clients (Python, Node.js, Go, .NET) available. Strong consistency (CP subsystem) available for leader election and distributed locking. Confidence is docs-derived.

REST Management API at {host}:5701/hazelcast/rest. Resources: maps, queues, cluster, members. GET /hazelcast/rest/maps/{mapName}/{key} reads a map entry. POST /hazelcast/rest/maps/{mapName}/{key} writes a value. DELETE /hazelcast/rest/maps/{mapName}/{key} removes entry. GET /hazelcast/rest/cluster returns cluster metadata and member list. Native Java/Python/Node.js/Go clients preferred for full feature access (IMap, IQueue, ITopic, etc.) with binary protocol performance. REST is suitable for simple get/put operations; native clients required for advanced features (entry listeners, predicates, transactions).

Distributed operations are resilient: failed member operations automatically re-route to other cluster members. Client reconnect logic built into all official clients. REST operations return HTTP 200/400/503 (cluster unavailable). Map TTL and max-idle configuration handle stale entry cleanup. Hazelcast Cloud SLA: 99.99% for Enterprise. Self-hosted reliability depends on cluster size and deployment. CP subsystem (Raft-based) provides linearizable operations for lock/counter; AP subsystem for maps and queues (eventual consistency in split-brain scenarios).

docs.hazelcast.com provides comprehensive documentation: data structures guide (IMap, IQueue, ITopic, ISet, IList, MultiMap), client library guides for all supported languages, REST API reference, security configuration, Hazelcast Cloud setup, and CP subsystem guide. Documentation quality is high. Getting started requires a Hazelcast Cloud account (free tier) or Docker self-hosted. Java client is the most feature-complete; other language clients have narrower feature coverage. Community support via Hazelcast Slack and GitHub.