NautheraAuditQueueBacklog
Docs version v1 (main)Nauthera is pre-1.0, in a design-partner program
An audit queue is more than 40% full.
Meaning
nauthera_ for one queue (lane:
priority for mandatory events, buffered for the rest) has been above the threshold
(0.4) for five minutes. Events arrive faster than the writer stores them: writes are
slow or failing, or the event rate jumped (an attack producing many refusals, a raised
audit.level). This alert fires before any event is dropped.
Impact
Nothing is lost yet. Records reach the database and the exporters later, so the SIEM sees
them late. On the priority queue, unauthenticated callers' mandatory events are refused
room once it is half full, and at four fifths the pipeline turns degraded, which with
audit.fail_closed refuses administrators' privileged changes.
Diagnosis
- Which queue and replica:
nauthera_.audit_ buffer_ depth\{job="<name>"\} / nauthera_ audit_ buffer_ capacity\{job="<name>"\} - Write health:
rate(nauthera_and the p99 ofaudit_ flushes_ total{job="<name>",result="error"}[5m]) nauthera_.audit_ flush_ duration_ seconds - Arrival rate by category:
sum by (category, outcome) (rate(nauthera_. A jump in failures points to an attack; check NautheraThreatSignalCritical and NautheraSignInFailureSpike.audit_ events_ total\{job="<name>"\}[5m]))
Mitigation
- Slow or failing writes: follow NautheraAuditWritesFailing; check database load.
- Higher volume than the queue absorbs: raise
config.audit.buffer_sizeorconfig., or loweraudit. priority_ buffer_ size config.audit.level. - An attack: handle it as its own alert says; refusals from the edge are coalesced, so the backlog falls once the source is blocked.
Escalation
No page on its own; escalate to the identity platform owner if the backlog keeps growing towards drops (NautheraAuditEventsDropped, NautheraAuditMandatoryEventsDropped).
Alert runbooks · Generated from internal/ and operator/ in nauthera-server.