Cases14 cases

6 to 10 of 14

1. IL-0005 A new telemetry service overloaded the Kubernetes API servers of large clusters and broke DNS-based service discovery for all OpenAI services

   OpenAI2024

   Symptoms

   Alerts fired at 3:13 PM PST on 11 December 2024, and from 3:16 PM every OpenAI service was significantly degraded or completely unavailable, including ChatGPT, the API and Sora. The status page reported API calls returning errors and difficulties logging in to platform.openai.com and ChatGPT. Customer impact peaked at 3:40 PM.

   Service outageHigh error rateControl plane unresponsiveDNS resolution failures

   Trigger

   At 2:23 PM PST a change introducing a new telemetry service, which collects detailed Kubernetes control plane metrics, was merged and the deployment pipeline started. Between 2:51 PM and 3:20 PM the change was applied to all clusters.

   Root cause

   Configuration change

   Component

   Kubernetes API serverClient-side DNS resolver and cacheTelemetry agent

   Contributing factor

   Unrepresentative stagingNo staged rolloutCache-masked failureMonitoring blind spotLimited debug access

   Detection

   OpenAI's detection tools worked: alerts fired at 3:13 PM PST, a few minutes before customer impact began at 3:16 PM, and the root cause was identified at 3:16 PM.

   Remediation

   • Scaled clusters down to reduce the total Kubernetes API load
   • Blocked network access to the Kubernetes admin APIs so no new expensive requests arrived
   • Scaled up the Kubernetes API servers
   • Removed the telemetry service once enough control was back
   • Moved traffic from impacted clusters to healthy ones
   • Intervened manually in clusters where simultaneous resource downloads saturated resource limits

   Action items

   • Put every infrastructure configuration change through a phased rollout with continuous monitoring of both the service workloads and the clusters, including the Kubernetes control plane
   • Run fault injection tests in which the data plane runs without the control plane for longer, and tests that roll out bad changes on purpose to check detection and rollback
   • Build break-glass mechanisms so engineers can reach the Kubernetes API servers in any circumstances
   • Decouple the Kubernetes data plane from the control plane so that service discovery for critical workloads does not depend on the control plane
   • Speed up recovery with better caching and dynamic rate limiters for resources needed at cluster startup, and regular drills that rapidly replace an entire cluster

   Lessons

   Several systems and processes failed at once and interacted in unexpected ways: tests did not catch the change's effect on the Kubernetes control plane, DNS caching put a delay between the change and the failures, and being locked out of the control plane made remediation very slow. Because service discovery used Kubernetes DNS, the control plane was carrying load for product traffic even though the data plane was meant to be independent of it.

   View case
2. IL-0203 A routine Kubernetes cluster system update made the AWS Load Balancer Controller fail, so new pods were never registered with the ALB and Honeycomb's EU1 ingest and UI went down once the last old pods were replaced

   Honeycomb2024

   Symptoms

   Event ingestion and the Honeycomb UI in EU1 became unavailable at 22:00:10 UTC on April 9, 2024, and the status page reported that the EU1 instance was down. Both the ingestion service and the service behind the web UI stopped at roughly the same moment.

   Service outageStale endpoint routing

   Trigger

   A routine system update to the Kubernetes cluster, rolled out earlier on April 9, followed by a regular deployment that replaced pods one at a time.

   Root cause

   Breaking change in an upgrade

   Component

   Ingress controllerCloud load balancer

   Contributing factor

   Monitoring blind spotUnrepresentative stagingChange with a delayed effect

   Detection

   The first status post, saying the EU1 instance was down, went up at 15:08 PDT (22:08 UTC), about 8 minutes after the outage began. There was no alerting on the AWS Load Balancer Controller.

   Remediation

   • Restarted the AWS Load Balancer Controller
   • Rolled back the Kubernetes cluster system update

   Action items

   • Find the exact cause of the failure so the system update can be deployed safely
   • Add monitoring and alerting on the AWS Load Balancer Controller so on-call engineers learn of a failure quickly and can pause deployments

   Lessons

   A rolling deployment can look healthy while the component that wires new pods into the load balancer is broken; the failure only shows when the last old pod goes away. Honeycomb now treats the AWS Load Balancer Controller as a critical component that needs its own alerting, so deployments can be paused before they remove the last pods that still receive traffic. Extensive testing in development and staging did not reveal the bad interaction.

   View case
3. IL-0060 Kubernetes 1.24 upgrade removed the master node label that Calico route reflectors selected on, dropping all pod routes and taking Reddit down for 314 minutes

   Reddit2023

   Symptoms

   About two minutes after an engineer started the Kubernetes 1.23 to 1.24 upgrade of the cluster running Reddit's legacy stack, just after 19:00 UTC, the site stopped. The CDN edge request graph, which is kept separate from the cluster, showed traffic by status collapsing. Users saw the error page, a homepage with no posts, or plain errors.

   Service outageHigh error rateDNS resolution failuresMonitoring outagePods stuck in ContainerCreatingControl plane lag

   Trigger

   An in-place upgrade of the legacy cluster from Kubernetes 1.23 to 1.24; in 1.24 the node-role.kubernetes.io/master label is removed even from running clusters.

   Root cause

   Breaking change in an upgrade

   Component

   CalicoKubernetes API serverAdmission webhooketcd

   Contributing factor

   Configuration not in codeUnrepresentative stagingUntested backupMonitoring blind spotSingle point of failure

   Detection

   The site broke about two minutes into the upgrade; an incident was opened immediately and responders were on the call within three minutes, working from CDN edge graphs because the cluster's own metrics were gone.

   Remediation

   • deleted the Open Policy Agent webhook configurations to stop API server timeouts
   • terminated all worker nodes of the cluster
   • restored the control plane from the pre-upgrade backup onto the first control plane node
   • switched off the cluster autoscaler until the control plane was stable
   • generated new TLS certificates so replacement control plane nodes could join etcd
   • brought traffic back in steps from 1% to 100%
   • restarted legacy services that had been stopped by hand

   Action items

   • Codify cluster configuration so every choice has a record, and document in detail what cannot be codified
   • Standardise components across clusters and replace bespoke configurations
   • Build tooling to manage the lifecycle of whole clusters so they can be recreated or replicated
   • Add pre-production clusters that can be tested with confidence
   • Improve the backup restore process and make cluster upgrades safe

   Lessons

   Configuration that exists only in a live cluster, edited by hand through a special tool, can be forgotten, as the route reflector setup was, and then break on an upgrade nobody connects it to. A Kubernetes upgrade cannot be reverted without a restore, so the restore procedure has to be kept current and practised. Consistent, codified clusters are what make upgrades testable; bespoke clusters keep producing outages.

   View case
4. IL-0109 A retried node rollover on a DFDS EKS cluster set the Auto Scaling group back to 15 instances, and the default termination policy removed every healthy worker node, leaving the cluster with no nodes

   DFDS2023sev1

   Symptoms

   At 09:33 UTC on 14 February 2023 end users told the platform team that Kubernetes logs were not arriving in CloudWatch as expected. The investigation went through fluentd and found that the nodes were running containerd as their container runtime instead of docker-shim. A scheduled node rollover the day before, at 07:30 UTC on 13 February, had set containerd as the default runtime by accident.

   Service outageResources deletedMonitoring outage

   Trigger

   A scheduled node rollover had accidentally made containerd the default container runtime, which stopped log shipping to CloudWatch; the unscheduled rollover to revert it timed out, and the pipeline run that followed reduced the Auto Scaling group from 30 to 15 instances.

   Root cause

   Configuration change

   Component

   Container runtimeNode launch templateNode poolTerraformTelemetry agent

   Contributing factor

   Automation amplificationUnreviewed defaultsUnrepresentative stagingNo configuration testsNo runbookHarmful manual action

   Detection

   End users reported the missing CloudWatch logs that started the work. The postmortem lists 2 minutes as the time to detection of the node failure.

   Remediation

   • ran the node rollover script to add new nodes, which failed on a timeout from Docker Hub
   • ran the node rollover script a second time, which added new nodes that the load balancer marked healthy at 11:01 UTC

   Action items

   • Document the rollover process, especially what can be done when it goes wrong
   • Investigate options for improving the rollover process
   • Add fluentd log collection to the QA test suite so logs are covered end to end
   • Add alarms for fluentd when too little or no log data is sent
   • Implement a staging environment
   • Implement guards against selecting an invalid container runtime

   Lessons

   When a change goes wrong, rolling back to a known working configuration is safer than rushing a fix, which can make the damage worse. Launching new worker nodes depends on the cloud-init script not failing in any way. The QA configuration differed from production, there was no staging environment, the QA tests did not cover fluentd log collection, and the rollover process had no documentation for when it fails.

   View case
5. IL-0116 A CNI configuration change incompatible with a hardware variant that exists only in production cut network connectivity in an OpenAI GPU cluster and took text-davinci-003 fully down

   OpenAI2023

   Symptoms

   At 4:53 pm Pacific on 25 January 2023 one of OpenAI's GPU clusters had a major outage, and the text-davinci-003 model went fully down.

   Service outageThroughput degradation

   Trigger

   A configuration change to the CNI (Container Network Interface) plugins, which provide network connectivity to containers, was deployed to a GPU cluster that contained a hardware variant the staging environment did not have.

   Root cause

   Configuration change

   Component

   CNI plugin

   Contributing factor

   Unrepresentative stagingMonitoring blind spot

   Detection

   Engineers immediately identified the problem as network connectivity, but it took nearly an hour to identify CNI as the cause.

   Remediation

   • put a fix for the CNI configuration change in place, which restored 80% of traffic immediately
   • deployed models back onto GPUs over about three more hours until capacity was fully stable

   Action items

   • Make staging environments contain every type of hardware configuration so they better represent production
   • Monitor and alert on the health of the CNI system directly
   • Fix the network and storage bottlenecks that limit how quickly many replicas of a model can be deployed

   Lessons

   A change that had been deployed to other clusters over 24 hours and deemed safe still broke a cluster with different hardware, so staging needs every hardware configuration that production has.

   View case

By root cause5