Engine and storage settings

Listeners

server.listen_address: HTTP and WebSocket listener. Default 127.0.0.1:8080.
server.grpc_addresses: Nonempty list of unique, nonzero-port gRPC listeners. Default [127.0.0.1:50051]. The first is the primary control endpoint.
server.cors_allowed_origins: Exact HTTP origins, with no path or trailing slash. Default empty: same-origin access only; no wildcard.

store.backend: haematite (durable, default) or memory (ephemeral; loses state at shutdown).
store.owned_shards: Static shard assignment. Empty means this node owns all shards. Meaningful for haematite, not memory.
store.data_dir: Haematite data directory. The merged config defaults it under Aion home. Existing data is reopened; a fresh directory creates a database.
store.shard_count: Number of haematite shards to create on a fresh database. Defaults to 64.
store.cluster: Optional cluster membership. Omit for single-node storage. Requires haematite.
store.node_cache_budget: Required for haematite. Use an object with a positive bytes integer, or the explicit string unlimited. File boot healing supplies unlimited when missing; a constructed config must provide it.
store.lock_acquisition_patience_ms: Retired. Warned and ignored; the server waits for the kernel writer lock and reports the holder.
store.lock_acquisition_retry_cadence_ms: Retired. Warned and ignored; the kernel lock controls acquisition.

store.cluster.node_id: Required unique distribution node name.
store.cluster.bind_address: Required replication and quorum listener address.
store.cluster.members: Complete quorum membership by node id. An empty list means this node alone. Do not replace full membership with only reachable peers.
store.cluster.peers: Peer connection entries. Each gives its name, replication endpoint, optional gRPC endpoint and owned shards.
store.cluster.failover_poll_interval_ms: Peer liveness cadence in milliseconds. Default 500.
store.cluster.failover_confirmations: Consecutive disconnected observations before adoption. Positive integer; default 3.
store.cluster.cancel_relay_backoff_base_ms: Required positive first-retry delay for relaying a remote descendant cancellation. No default.
store.cluster.cancel_relay_backoff_multiplier: Required retry growth factor, at least one. No default.
store.cluster.cancel_relay_backoff_max_ms: Required delay ceiling, at least the base delay. No default.

store.cluster.peers.name: Required peer node name, matching its configured node id.
store.cluster.peers.address: Required peer replication endpoint.
store.cluster.peers.grpc_address: Optional peer client API endpoint for forwarding. Without it an incorrectly addressed request returns NotOwner.
store.cluster.peers.owned_shards: Shards assigned to this peer. Automatic adoption requires these to be declared.

runtime.scheduler_threads: Positive engine scheduler thread count. Default 1.
runtime.jit_threshold: Optional positive compilation threshold. When absent, beamr chooses its own default. A large value defers compilation; it does not disable it.
runtime.query_timeout_ms: Positive query reply deadline in milliseconds. Merged default 10000; zero is refused.
runtime.workloop_sweep_interval_ms: Positive workloop sweep interval in milliseconds. Merged default 1000. It contributes to health-alarm detection latency.
runtime.stop_drain_timeout_ms: Positive no-progress bound while stopping engine workers. Merged default 30000 ms. Each progress report resets the bound.

drain.timeout_seconds: Positive graceful transport drain window. Default 30 seconds.

namespaces.default: Default namespace name. Default default.
namespaces.auto_create: open (default) allows an unseen namespace to be created during worker registration or workflow start; closed requires explicit creation through POST /namespaces.
namespaces.max_in_flight_activities: Default 1024. Caps claimed outbox rows only when outbox.enabled=true, for parallel/collect members. Ordinary single activity calls bypass it. The typed outbox default is false; the first-run template enables it. A successful worker handoff marks the row done while its handler may still execute, so running handlers can exceed this claim ceiling. Cluster nodes receive a rounded owned-shard share.

namespace.mode: SharedEngine by default. SingleTenant carries one explicitly named namespace.