Observability & Alerts
Both hosts expose operational endpoints on their loopback management_listen addresses and never log or label metrics with client-supplied bytes.
Endpoints
| Host | Listener | Endpoints |
|---|---|---|
tachyon-pir-publisher | management_listen | GET /health, GET /ready, GET /.well-known/pir-service-discovery |
tachyon-pir-publisher | each product listen | GET /metrics (Prometheus text format) |
tachyon-pir-query-server | management_listen | GET /health, GET /ready, GET /metrics |
/ready returns HTTP 200 with per-product "phase": "running", "ready": true once initial sync/publication has committed, and 503 while starting or degraded.
Publisher metrics
- Memo (
tachyon_memo_*):ingested_height,anchor_height,entries_total,open_scope_entries,segments_overflow_total,publication_seconds,sessions_revoked_total,envelopes_invalid_total,state_entries_total. - First-contact (
tachyon_first_contact_*):open_span_entries,open_span_epochs,segments_overflow_total,publication_seconds,sessions_revoked_total. - Tachygram (
tachyon_tachygram_*):ingested_height,anchor_height,entries_total,open_scope_entries,segments_overflow_total,publication_seconds,sessions_revoked_total. - Registry (
tachyon_registry_*):log_sequence,open_chunk_entries,registrations_total,registrations_duplicate_total,registrations_rejected_total,rate_limited_total,publication_seconds,sessions_revoked_total.
Key alerts
- Ingestion stalled:
ingested_heightfalling behind the Zakura node tip, or ingestion halted by a duplicate/all-zero tachygram integrity fault. - Segment overflow: any publication failing because a bucket exceeded
MAX_SEGMENTS(8). The previous publication continues serving while operators investigate. - Query host staleness: a query server failing to confirm publisher authority within
max_unconfirmed_seconds(default600s) transitions out of ready. - Registry write abuse: spikes in
tachyon_registry_rate_limited_totalortachyon_registry_registrations_rejected_total.