Cases10 cases

6 to 10 of 10

1. IL-0132 A faulty resource cleanup process marked Vertex AI prediction endpoints as deleted worldwide, undeploying customers' models and failing predictions for 4 hours 20 minutes

   Google Cloud2022

   Symptoms

   From 10:10 US/Pacific on 2 June 2022, Vertex AI Online Prediction returned errors for 30% up to 100% of requests per region, depending on how each customer used the service.

   High error rateResources deletedService outage

   Trigger

   A faulty resource cleanup process marked Vertex Prediction Endpoints as deleted, globally.

   Root cause

   Software defect

   Component

   Model request routing

   Contributing factor

   Automation amplificationExcessive blast radius

   Detection

   The first public status update was posted at 13:58 US/Pacific and gave the start as 10:20 US/Pacific; the later mini incident report gives 10:10.

   Remediation

   • restored the Vertex Prediction Endpoints that had been marked as deleted

   View case
2. IL-0106 A malformed Traefik IngressRoute triggered a Traefik v2 bug once a node rollover restarted every Traefik pod, and all Kubernetes services returned 404

   DFDS2021

   Symptoms

   At 08:00 UTC on 2021-09-06 every service in DFDS's Kubernetes cluster answered with 404. Investigation started five minutes later.

   Service outageHigh error rateProbe failures

   Trigger

   The automated Kubernetes node rollover on 2021-09-06 restarted all the Traefik v2 pods, five days after a malformed IngressRoute manifest had been applied.

   Root cause

   Software defect

   Component

   Ingress controllerCloud load balancer

   Contributing factor

   Change with a delayed effectNo configuration testsExcessive blast radius

   Detection

   All services returned 404 from 08:00 UTC and root cause investigation started at 08:05. Time to detection was 5 minutes.

   Remediation

   • changed the ALB health check path, which made the targets healthy but did not stop the 404s
   • rolled back by hand to Traefik v1 as the main ingress controller
   • deleted the malformed IngressRoute by hand and confirmed the Traefik v2 bug
   • applied the changes upstream

   Action items

   • Upgrade Traefik to v2.5, where the bug is fixed, and verify that it removes the bug
   • Change the Traefik v2 health check path to something more robust
   • Set up an external liveness probe on the ingress controller
   • Consider workload segmentation to limit the blast zone
   • Look into monitoring the Traefik v2 API for rapid changes
   • Look into smoke testing during and after node rollover

   Lessons

   A malformed IngressRoute did nothing visible for five days and then took down all ingress when every Traefik v2 pod had restarted during a node rollover. Because Traefik v2 failed gracefully, the first signals pointed at the ALB health checks instead of at Traefik. On incident handling, the team concluded that appointing an incident manager could have helped, that the war room meeting should be recorded and a log book kept, that people not involved should only make suggestions in writing, and that a short first round of sharing thoughts and gut feelings is useful.

   View case
3. IL-0035 Worker nodes went NotReady after a new ElastAlert rule without memory limits used up node memory

   NU.nl2019

   Symptoms

   The NU.nl development team got reports that several services on one of its Kubernetes clusters behaved erratically: occasional error pages, slow responses and timeouts. The affected services were mostly internal tools and dashboards.

   Node NotReadyHigh error rateHigh latencyIntermittent timeoutsMonitoring outageOOMKilled

   Trigger

   A new ElastAlert rule, deployed shortly before from the command line, that fired if the Cloudfront-* Elasticsearch indexes stopped receiving documents; those indexes take a few million requests per hour, and the rule made ElastAlert's memory grow without bound.

   Root cause

   Configuration change

   Component

   Container memory limitkubeletPrometheus

   Contributing factor

   No resource requests or limitsUnreviewed defaultsNo alert on limitMonitoring blind spotExcessive blast radius

   Detection

   Users reported error pages, slow responses and timeouts from several services; no alert fired.

   Remediation

   • terminated the NotReady nodes one by one so the auto-scaling group replaced them (the new nodes also went NotReady, but Prometheus and Grafana landed on a stable node)
   • found the ElastAlert pods' memory growth on the Grafana pod memory dashboard
   • fixed the ElastAlert problem that caused the memory growth

   Action items

   • Define alerts on resource usage such as CPU, memory and disk space
   • Keep researching an alerting strategy that combines Elasticsearch, Prometheus Alertmanager and New Relic
   • Consider exporting metrics out of the cluster and running Grafana outside it, on EC2 or a separate Kubernetes cluster
   • Either visualise the important metrics in the ELK stack or make Prometheus and Grafana more available, and improve the metrics strategy
   • Set resource limits for ElastAlert through the Helm values
   • Configure namespaces with default memory requests and limits
   • Isolate customer-facing workloads from build, logging and monitoring workloads (a dedicated node group, or later a separate tools cluster)
   • Deploy everything through automation such as Jenkins pipelines, and consider a GitOps approach
   • Deploy through a pipeline that includes a smoke test step
   • Make Kubernetes and cloud infrastructure work part of the team's sprints, picked up by all team members while pairing with more experienced members, and hold workshops that go deep into specific topics

   Lessons

   Fixing the root cause covers only a small part of what an incident can teach; the chain of smaller failures around it is where most of the learning is. A pod without a memory limit can take whole nodes down, so limits need defaults at the namespace level. Monitoring that runs inside the cluster it watches goes blind with it. Changes installed by hand from the command line leave the rest of the team without a record of what changed.

   View case
4. IL-0001 Cluster-wide DNS outage after CoreDNS pods were OOM-killed again and again under a burst of DNS queries

   Zalando2019

   Symptoms

   Customers of the Zalando Fashion Store got a high rate of errors on all web product pages and outfit pages for more than an hour. The errors came from the main data aggregation layer service, which runs in one of Zalando's Kubernetes clusters.

   High error rateDNS resolution failuresOOMKilledMonitoring outage

   Trigger

   Calls from the aggregation layer to a single, not very important downstream service started timing out. The aggregation layer returned 404s for them, clients retried, and the resulting spike in requests became a spike in DNS queries to CoreDNS.

   Root cause

   Capacity exhaustion

   Component

   CoreDNSContainer memory limitClient-side DNS resolver and cache

   Contributing factor

   Unknown limitNo cachingRetry stormUnreviewed defaultsExcessive blast radiusMonitoring blind spot

   Detection

   The on-call engineer for the aggregation layer phoned the on-call Kubernetes engineer. Automatic paging did not happen because the cluster's monitoring and alerting ran inside the cluster and failed along with DNS. The Kubernetes engineer then saw that the CoreDNS pods were being OOM-killed.

   Remediation

   • Raised the CoreDNS memory requests and limits by hand from 100Mi to 2000Mi

   Action items

   • Introduce a more resilient DNS setup: CoreDNS fronted by dnsmasq on every node
   • Improve high-level alerting from outside the clusters, so that a cluster problem is known before or at the same time as applications are affected
   • Improve incident response processes and tooling so that any on-call engineer can escalate directly to other teams when automatic paging fails, for example through a chat bot

   Lessons

   The Kubernetes DNS setup was far from resilient enough: a single application could take down DNS for everything in the cluster. The blast radius of one application sending more DNS queries should be limited to one node at most, ideally to single pods. Known Kubernetes DNS problems, such as Linux kernel races in conntrack handling and the ndots 5 problem, were familiar to the team and partly mitigated, but their combined effect on the DNS infrastructure was worse than the team had understood. Monitoring that runs inside the cluster was not robust enough: when the cluster's DNS broke, it could not page anyone, and recovery took longer than it should have.

   View case
5. IL-0031 Kubernetes API server OOM-killed, leaving the Skipper ingress without routes; the talk points to a misconfigured CronJob manifest

   Zalando2018

   Symptoms

   Customers were affected, and Zalando's IAM service answered requests with HTTP 404. The talk shows this on slides titled "Customer impact" and "IAM returning 404", followed by a slide on the number of pods in the cluster.

   High error rateControl plane unresponsiveOOMKilled

   Trigger

   A CronJob (apiVersion batch/v2alpha1, schedule "*/15 9-19 * * Mon-Fri") whose concurrencyPolicy: Forbid and job history limits sat inside the pod template spec instead of the CronJob spec. The talk shows it right after the slides on the number of pods and the API server being OOM-killed, and the lessons add a quota on the number of pods.

   Root cause

   Capacity exhaustion

   Component

   Kubernetes API serverIngress controller

   Contributing factor

   Single point of failureExcessive blast radius

   Remediation

   • fixed the CronJob manifest: moved concurrencyPolicy: Forbid and the job history limits to the CronJob spec, added activeDeadlineSeconds: 120 and changed the schedule to "7 8-18 * * Mon-Fri"

   Action items

   • Make the ingress stay healthy during API server problems
   • Make the ingress keep the last known set of routes
   • Use a ResourceQuota on the number of pods (the example sets pods: "1500")

   Lessons

   An AWS ALB routes traffic to all hosts when all hosts report unhealthy, so the ingress should stay healthy during API server problems. An ingress controller that reads routes from the API server should keep its last known set of routes. Use a quota for the number of pods.

   View case

By root cause5