Limits

Size caps, budgets, lifetimes, and fixed maximums in RushDB, with their defaults. Limits marked configurable can be changed (see the configuration reference); the others are fixed in this release.

Requests and responses

Limit	Default	Configurable	Behavior when exceeded
HTTP request body	16 MiB	server.body-limit	413, invalid_request
Response size (all transports)	16 MiB	server.body-limit	HTTP 413, execution_failed; elsewhere resource_exhausted
gRPC message size (in and out)	16 MiB	server.body-limit	gRPC error
HTTP request bodies in flight (all connections)	256 MiB (never below the body limit)	http-inflight-body-bytes	503, resource_exhausted, before the body is read
Admitted concurrent requests (scheduler)	workers + queue; workers = CPU count clamped to 2–256, queue = 32 × workers clamped to 64–4096	threads	503, unavailable
Concurrent requests per tenant (multitenant)	unlimited	--max-concurrent-requests-per-tenant	unavailable
Request execution deadline (every transport)	none	--request-timeout-ms	cancelled
X-RushDB-Tenant header length	1024 bytes	no	400, invalid_request
pgwire connections	1024	no	connection refused
Bolt connections	1024	no	connection refused
Bolt message size	64 MiB	no	connection error
Bolt buffered rows per result	100,000	no	error
Bolt idle I/O timeout	60 s	no	connection closed
Bolt authentication timeout	10 s	no	connection closed
gRPC concurrent requests per connection	64	no	queued by HTTP/2
gRPC concurrent streams per connection	64	no	queued by HTTP/2

Queries

Limit	Default	Configurable	Behavior when exceeded
Query working set	256 MiB	query-memory-max (0 = off)	resource_exhausted
Query phase reservation	8 MiB per request, drawn from the managed envelope	no	resource_exhausted when the envelope is full
Records cursor page size	1–10,000	no	refused
Retained cursor state (node-wide)	64 MiB	no	new cursor refused
Cursor lifetime	as requested, capped at tx-max-age-seconds	tx-max-age-seconds	cursor expires
Aggregation buckets per query	1,000	no	refused
SQL: records per label (table)	100,000	no	query refused
Graph-language mutation: records created or changed	100,000	no	mutation refused
Graph-language mutation staging	64 MiB	no	resource_exhausted (not retryable)
Graph-language statement clauses	8,192	no	invalid_request
text query default limit	10	per request	—
vectorIndexCandidates default limit	500	per request	—
Vector spaces per tenant	256	no	refused
Records per transport response	bounded by the response size	server.body-limit	see above

Writes and transactions

Limit	Default	Configurable	Behavior when exceeded
Expanded size of one commit	512 MiB	max-expanded-commit-bytes	resource_exhausted (not retryable)
WAL record size	64 MiB (larger commits are chained automatically)	no	—
Active transactions	128	tx-max-active	resource_exhausted (retryable)
Operations per transaction	10,000	tx-max-operations	resource_exhausted (not retryable)
Staged bytes per transaction	8 MiB	tx-max-staged-bytes	resource_exhausted (not retryable)
Staged JSON nesting depth	64	no	resource_exhausted (not retryable)
Transaction pool (all transactions)	128 MiB	transaction-pool	resource_exhausted (retryable)
Staged bytes per tenant (multitenant)	shared pool	tx-max-staged-bytes-per-tenant	resource_exhausted
Default transaction lifetime	60 s	tx-default-ttl-seconds	transaction expires
Maximum transaction lifetime	60 s	tx-max-age-seconds	requested TTL is capped
Write-lock wait	unlimited	lock-timeout-ms	resource_exhausted (retryable)
Idempotent retry window	24 h	idempotency-window-seconds	a retry becomes a new commit
Free disk for writes (every transport)	10%	disk-space-threshold-percent	unavailable (HTTP 503)

Memory budgets

Budget	Default	Configurable
Managed envelope	512 MiB	managed-limit
Storage page cache	128 MiB (minimum 64 KiB)	storage-cache
Logical caches (vector, text, planner statistics)	128 MiB	logical-cache-bytes
Planner statistics	32 MiB (child of the logical caches)	planner-stats-bytes
ANN segment cache	256 MiB, and at most 12 cached generations	no
Resident tenant handles	512 MiB (16 KiB minimum charge per tenant)	tenant-resident-bytes
Tenant handle idle time	300 s	tenant-idle-seconds
Memory headroom inside a declared deployment limit	128 MiB	memory-headroom

Retention and storage

Limit	Default	Configurable
WAL commit cap	100,000 commits	tx-log-retention-commits, tx-log-keep-all
WAL byte cap	4 GiB	tx-log-retention-bytes
WAL keep window without consumers	0 commits	tx-log-keep-commits
WAL segment size	64 MiB	no
Retention consumer lease	24 h	retention-consumer-lease-seconds
Abandoned read view lease	60 s	no
WAL archive window	7 days	wal-archive-retention-seconds
WAL archive byte cap	none	wal-archive-retention-bytes
Physical storage shards	1 (maximum 256, Preview above 1)	shards
Compaction trigger	file grew to 150% or ≥ 10,000 keys purged	compact-growth-percent
Replication page	2,000 WAL records per poll	no

Background intervals

Task	Interval	Configurable
Maintenance tick (GC, retention, index builds, checkpoint)	5 s, up to 512 keys per step	checkpoint-interval-seconds
Index build budget per tick	250 ms	no
Idle compaction check	300 s	compact-interval-seconds
Compaction quiet period	10 s	compact-quiesce-seconds
Expired-transaction reaper	1 s	no
Text index applier	250 ms	no
WAL archiver	5 s	no
Storage byte statistics refresh	15 min	no
RSS watchdog (when enabled)	10 s	rss-max
Replication follower poll when caught up	200 ms	no

Shutdown

Phase	Deadline
HTTP connection drain	30 s
Background maintenance drain	30 s
Final durability checkpoint	5 s
pgwire / gRPC / Bolt drain	5 s each, inside a 7 s transport budget

See also

• Configuration reference
• Capacity planning
• Error codes