High-cardinality metrics ingestion, EU-hosted.
The platform has been in continuous operation since its first regional collector went live. Ingest capacity is provisioned per customer, and retention of raw samples is deliberately short — we are a transport layer, not a storage product.
Between 09:12 and 10:48 UTC a subset of eu-west collectors returned elevated p99 latency. Agents retried as designed and no samples were lost. Cause was a saturated buffer volume after an unusually large backfill...
Delivery to downstream backends lagged by up to 20 minutes for roughly two hours. Buffering held all samples and delivery caught up without intervention once the upstream backend recovered.
Rolling upgrade of eu-central collectors. Agents failed over to eu-west for the duration. No customer action was required.