Tuning

A few knobs worth thinking about before turning on the exporter at scale.

MONGO_POOL_SIZE (default 50)

Sets maxPoolSize on the mongoose connection. The advanced scrape fans many aggregations out in parallel via Promise.all; if you see queries serialize on small deployments, the pool is the first place to look. Most LibreChat installs are fine at the default.

ADVANCED_REFRESH_INTERVAL (default REFRESH_INTERVAL × 10, i.e. 300 s)

The heavy $facet aggregations run at this cadence. On very large message/transaction collections you may want this longer (e.g. 600000 for 10-minute granularity) to reduce Mongo load. The basic counts continue to refresh on REFRESH_INTERVAL.

REFRESH_INTERVAL (default 30000)

Cheap countDocuments per collection. Going much lower than 10 s is usually wasted load — Prometheus typically scrapes every 15 – 30 s anyway. There's no benefit to refreshing faster than the scrape cadence.

EMIT_PER_USER_METRICS=true

Only enable on small / single-tenant deployments. Three metrics gain an email label and emit one Prometheus time series per user (unbounded growth). The corresponding *_by_email_domain variants stay enabled regardless and are bounded by company-domain count.

TENANT_ID

When set, installs schema-level mongoose hooks that inject { $match: { tenantId } } into every aggregate / find / count. Use this when one MongoDB serves multiple LibreChat tenants and you want metrics scoped to one of them.

Reentrancy

Each tier (basic / advanced) holds a per-tier "running" flag and skips a tick if the previous one is still in flight. You will see a warn log on each skipped tick (basic scrape still running, skipping this tick); these are informational, not errors.

Persistent skipping means the scrape budget is too tight — either:

• raise ADVANCED_REFRESH_INTERVAL, or
• look at index coverage via the librechat_exporter_missing_indexes gauge and add the recommended indexes Mongo-side.