Metrics

Every Prometheus metric the server exports at GET /metrics. For scraping setup and alerting guidance see Monitoring.

Scraping

curl -s -H "Authorization: Bearer $RUSHDB_TOKEN" http://127.0.0.1:3000/metrics

• The endpoint requires the boot token or a capability with read permission.
• Format: Prometheus text exposition, version 0.0.4.
• --metrics-prefix prepends a string to every name; --metrics-filter keeps only families whose name contains a substring.
• Values are refreshed when the endpoint is scraped, except the storage byte breakdown (rushdb_storage_data_bytes, _history_bytes, _log_bytes, _receipt_bytes), which a background task refreshes every 15 minutes, starting 15 minutes after boot.
• Every family ending in _total is a counter; it resets when the process restarts.

Histogram buckets (seconds) for every histogram below: 0.001, 0.005, 0.01, 0.025, 0.05, 0.1, 0.25, 0.5, 1, 2.5, 5, 10.

Requests

Metric	Type	Labels	Meaning
rushdb_requests_total	counter	route	Requests executed, by route.
rushdb_request_duration_seconds	histogram	route	Request latency, measured around execution.
rushdb_rows_returned_total	counter	route	Rows/records returned by queries.
rushdb_operation_work_total	counter	operation, counter	Bounded work units. operation is the protocol family (query, mutate, describe, ingest, index, transaction, purge_tenant, backup) with counter response_rows or response_bytes; or operation="sql_scan" with counter candidate_rows, reconstructed_rows, filtered_rows, posting_bytes.
rushdb_scheduler_inflight	gauge	—	Jobs admitted to the bounded request scheduler.
rushdb_scheduler_capacity	gauge	—	Scheduler admission capacity (workers + queue).
rushdb_http_inflight_body_bytes	gauge	—	HTTP request-body bytes currently reserved.

route values: one per Protocol v1 operation (for example query.records, mutate.putRecord, transaction.commit; the full list is in HTTP endpoints), invalid for an HTTP body that is not a valid envelope, transport.execute for gRPC and Bolt, and pgwire.execute for PostgreSQL wire.

Commits and write path

Metric	Type	Labels	Meaning
rushdb_commits_total	counter	—	Durable commit index: commits since the store was created.
rushdb_commit_latency_seconds	histogram	—	Time from enqueue to durable acknowledgement per commit.
rushdb_commit_phase_seconds	histogram	phase	Commit pipeline time by phase. Per commit: admission_wait, build, expand, durable. Per publish group: wal_encode, wal_append_sync, backend_apply, derive, publish.
rushdb_storage_group_commits_total	counter	—	Group-commit write transactions persisted.
rushdb_sequencer_inflight_keys	gauge	—	Keys assigned to commits that are not yet published.
rushdb_conflict_window_bytes	gauge	—	Memory held by the conflict-validation window.
rushdb_storage_admission_waits_total	counter	—	Waits for the storage admission lock.
rushdb_storage_admission_wait_ms_total	counter	—	Milliseconds spent waiting for storage admission.
rushdb_storage_backend_mutex_wait_ms_total	counter	—	Milliseconds spent waiting for the storage backend lock.
rushdb_storage_backend_mutex_hold_ms_total	counter	—	Milliseconds the storage backend lock was held.
rushdb_storage_backend_mutex_hold_max_ms	gauge	—	Longest single hold of the storage backend lock.

Storage, WAL, and retention

Metric	Type	Labels	Meaning
rushdb_storage_file_bytes	gauge	—	Size of the primary store file (0 for in-memory).
rushdb_storage_data_bytes	gauge	—	Live data row bytes (key + value), excluding history and log. 15-minute refresh.
rushdb_storage_history_bytes	gauge	—	Retained MVCC history bytes. 15-minute refresh.
rushdb_storage_log_bytes	gauge	—	Command-log bytes inside the store; normally 0 for a durable store because the log is the WAL. 15-minute refresh.
rushdb_storage_receipt_bytes	gauge	—	Retry-receipt bytes. 15-minute refresh.
rushdb_wal_retained_bytes	gauge	—	WAL bytes retained on disk.
rushdb_bytes_stored	gauge	tenant	Approximate stored bytes; tenant="all" is the node total.
rushdb_records_stored	gauge	tenant	Records stored. tenant="all" always; per-tenant series (internal tenant id) only with --metrics-tenant-limit.
rushdb_storage_checkpoints_total	counter	—	Durability checkpoints taken.
rushdb_storage_checkpoint_wait_ms_total	counter	—	Milliseconds spent in checkpoints.
rushdb_storage_compactions_total	counter	—	Completed storage compactions.
rushdb_commit_log_rows_pruned	gauge	—	Command-log rows reclaimed by retention.
rushdb_receipts_swept	gauge	—	Retry receipts removed after the idempotency window.
rushdb_gc_tick_seconds	histogram	—	Duration of one maintenance tick (GC, retention, checkpoint).
rushdb_retention_consumers	gauge	—	Active retention consumers (backups, WAL archive, replicas, cursors).
rushdb_retention_protected_floor	gauge	—	Lowest commit protected by a retention consumer.
rushdb_retention_oldest_reader_commit	gauge	—	Oldest commit pinned by a live read view (0 = none).

Memory

Metric	Type	Labels	Meaning
rushdb_budget_used_bytes	gauge	pool	Bytes currently reserved in a managed pool.
rushdb_budget_peak_bytes	gauge	pool	Peak reservation of a pool since start.
rushdb_budget_limit_bytes	gauge	pool	Configured limit of a pool.
rushdb_allocator_allocated_bytes	gauge	—	Bytes allocated by the application (jemalloc stats.allocated).
rushdb_allocator_active_bytes	gauge	—	Bytes in active pages (stats.active).
rushdb_allocator_metadata_bytes	gauge	—	Allocator metadata (stats.metadata).
rushdb_allocator_resident_bytes	gauge	—	Physically resident bytes (stats.resident); the closest in-process view of RSS.
rushdb_allocator_retained_bytes	gauge	—	Bytes retained by the allocator but not active (stats.retained).
rushdb_allocator_bytes_allocated_total	counter	—	Cumulative bytes allocated; only populated in special measurement builds, otherwise 0.
rushdb_planner_stats_bytes	gauge	—	Bytes in the planner-statistics cache.
rushdb_planner_stats_entries	gauge	—	Value distributions in the planner-statistics cache.
rushdb_tenants_resident	gauge	—	Tenants whose in-memory handles are resident.
rushdb_tenant_resident_bytes	gauge	—	Bytes charged by resident tenant handles.

pool values: managed (the whole managed envelope), transaction (transaction staging pool), and, when the logical cache is enabled, vector, sql, graph. Per-query pools (sql, graph) are reserved only while a query runs, so their used is usually 0 between scrapes; read their peak. The rushdb_budget_* gauges track reservations, not mapped memory; compare them with the rushdb_allocator_* gauges. The allocator gauges stay 0 in builds without jemalloc.

Indexes and search

Metric	Type	Labels	Meaning
rushdb_text_pending	gauge	index	Changes not yet applied to each text index. In multitenant mode, summed by index name over the resident tenants.
rushdb_ann_exact_fallbacks_total	counter	—	Vector searches answered exactly because the cache budget could not admit the ANN index.
rushdb_unindexed_walks_total	counter	—	Queries whose unindexed predicate leaf walked more than 10,000 owners; describe.explain names the ordered index to declare.

Startup

Metric	Type	Labels	Meaning
rushdb_boot_phase_duration_seconds	histogram	phase	Startup phase durations: config, storage, raft (Preview only), total.

See also

• Monitoring
• Storage and compaction
• Configuration reference