Skip to content
Cyntex Docs

Storage Model

Cyntex uses a three-tier storage architecture with clearly defined responsibilities at each layer.

Three-Tier Storage

L1 — Hazelcast (Distributed In-Memory)

  • Task runtime state, Ringbuffer CDC data stream
  • Forked from Hazelcast 5.7.0 (strips hazelcast-sql / HCL extensions — pure Apache 2.0)
  • Note: CP Subsystem (FencedLock / leader election) was removed from OSS entirely since version 5.5 — EE only

L2 — MongoDB (Persistent Source of Truth)

  • Single source of truth for connection and task definitions
  • CDC offset, checkpoint, task HA lease (CAS document lock + fencing epoch)
  • Connector jar distribution (GridFS)

L3 — Paimon (Incremental Data Lake, external add-on at end of GA)

  • Finance-grade low-latency historical audit
  • Not included in the initial rollout (POC/Alpha/Beta); added at the end of GA

DSL Artifact Dual-Layer Model (ADR-0021)

Local .cyn.yml file          MongoDB cyntex.resources
(authoring draft)    →       (official store, single source of truth)
                    apply
  • cyntex apply <file> = validate → canonicalize → id upsert (same hash = no-op)
  • cyntex export <id> = pull canonical YAML from MongoDB to local
  • cyntex diff <file> = compare local draft against canonical version in the store

Task HA Strategy

Because Hazelcast CP Subsystem is unavailable in OSS, task HA uses:

  • MongoDB document CAS lease (findOneAndUpdate with $inc fencing epoch)
  • Heartbeat suicide (lease not renewed on timeout → process exits voluntarily)
  • Monotonic fencing token to prevent split-brain dual scheduling