Limits and Constraints

This page lists every hard limit and configurable constraint in ConvergeDB. Other pages link here rather than repeating these values.

Hard limits

These limits are fixed by the wire protocol and storage format. They cannot be changed through configuration.

Resource	Limit	Reason
Sources (SourceId)	64 (0 to 63)	Source set is stored as a u64 bitmask.
Fields per kind	64 top-level fields	PATCH presence mask and ChangedFields bitmask are u64.
Array elements per field	1,024	Validation cap enforced at schema registration.
String / VarString / Bytes / VarBytes max length	65,535 bytes	u16 length prefix (String/Bytes) or u16 count in directory entry (VarString/VarBytes).
Entity ID size	Exactly 32 bytes	Fixed-width row layout for cache-aligned storage.
Registered kinds	65,535	KindId is a u16.
Metadata per write	65,535 bytes	u16 length prefix on the wire.
Kind name prefix	convergedb_ is reserved	System kinds managed by the server use this prefix. User-defined kinds with this prefix are rejected at registration.

Configurable constraints

These values have defaults that work well for most workloads but can be tuned.

Parameter	Default	Description
Coalescing window	20 ms	Duration over which writes to the same entity are merged into a single flush. Lower values reduce notification latency. Higher values increase the coalescing ratio for hot entities.
Tombstone retention	5 minutes	How long tombstoned entities remain visible to queries and subscriber notifications before being removed by compaction. Subscribers offline longer than this period must re-bootstrap.
Liveness deadline	30 seconds	If a source’s connection is silent for this duration, the server considers the source dead. Entities with only that source’s bit are tombstoned.
Heartbeat interval	15 seconds	How often the client sends keep-alive frames. Must be less than the liveness deadline.
Reconnect max delay	10 seconds	Maximum exponential backoff between reconnection attempts.
Subscriber buffer size	4,096 notifications	Per-subscription bounded buffer. If the subscriber falls behind and the buffer fills, the subscriber is disconnected. On reconnect, the subscriber must re-subscribe (and optionally re-bootstrap).

Derived constraints

These are not configured directly but follow from the limits above.

Constraint	Value	Notes
Maximum entity row size	Header (50 bytes) + field data + padding to 64-byte boundary	Depends on the schema. The header contains the entity ID (32 bytes), source set (8 bytes), version (8 bytes), schema version (1 byte), and flags (1 byte).
Maximum concurrent streams	32,768	Wire protocol uses a 16-bit stream ID for request-response multiplexing.
Struct nesting depth	1 level	Sub-structs cannot contain other sub-structs.