Ingest Outage in US and EU

Final update

On June 24, we experienced approximately 30 minutes of severe data ingestion degradation in Honeycomb’s US and EU instances. During this degradation, Honeycomb rejected a significant percentage of inbound telemetry, and customers would have seen up to a half-hour gap in ingested telemetry. Additionally, during this same time window, customers would have experienced data processing degradation and service disruption within Anomaly Detection and Service Maps. A deployment containing a change to how services retrieve dataset schemas into local caches caused a cascading set of failures, resulting in failed remote cache retrievals, and subsequently, a MySQL stampede due to several services falling back to retrieving schemas from the database. This stampede resulted in significant memory utilization growth, causing some services to crash loop with out-of-memory exceptions while retrieving schemas. Customers would have seen 5xx errors from our ingestion APIs when this occurred, because ingestion services were among those that were crash looping. This also included crash looping from services responsible for Anomaly Detection and Service Maps. Right after these service crash loops began, a

1. Postmortem · Jul 2, 20:10 UTC
   On June 24, we experienced approximately 30 minutes of severe data ingestion degradation in Honeycomb’s US and EU instances. During this degradation, Honeycomb rejected a significant percentage of inbound telemetry, and customers would have seen up to a half-hour gap in ingested telemetry. Additionally, during this same time window, customers would have experienced data processing degradation and service disruption within Anomaly Detection and Service Maps. A deployment containing a change to how services retrieve dataset schemas into local caches caused a cascading set of failures, resulting in failed remote cache retrievals, and subsequently, a MySQL stampede due to several services falling back to retrieving schemas from the database. This stampede resulted in significant memory utilization growth, causing some services to crash loop with out-of-memory exceptions while retrieving schemas. Customers would have seen 5xx errors from our ingestion APIs when this occurred, because ingestion services were among those that were crash looping. This also included crash looping from services responsible for Anomaly Detection and Service Maps. Right after these service crash loops began, alerts fired for the crashing services, and procedures were run to pin all services back to the last known good build before the breaking change landed. Once services were running the previous deploy’s build, the 5xx error responses from ingestion APIs returned to nominal baseline levels, and service
2. Resolved · Jun 24, 19:04 UTC
   All services have been fully recovered. Impact window: 17:55–18:17 UTC Regions affected: Primary impact in US. The EU region was affected to a much lesser degree. During this period, the following effects may have occurred: Ingest: Some data was not ingested, resulting in complete or partial data loss for events sent during the window. Triggers: Triggers may have failed to fire within the impact window. SLOs: SLI values across the affected window are skewed by the missing data and may show artificial dips or accelerated budget burn. Service Maps: Maps covering the outage window are incomplete. Services and dependencies may be under-counted, as their traces were not ingested. We apologize for the disruption. Please reach out if you have any questions about how this may have affected your data.
3. Monitoring · Jun 24, 18:31 UTC
   Triggers, SLOs and Service Maps are recovered. We are continuing to monitoring the Ingest Service
4. Monitoring · Jun 24, 18:22 UTC
   We have rolled back the deploy and ingest service has recovered. There has been an ingest outage from 17:55 - 18:14 UTC. We have also observed an delay for Service Maps.
5. Investigating · Jun 24, 18:14 UTC
   We are investigating an issue with delayed ingest in the US and EU region. Our engineers are rolling back a recent deploy. SLOs and Trigger evaluations may also be delayed.