Cases29 cases

7 to 11 of 29

1. IL-0027 All CoreDNS pods landed on one node, the node ran out of memory, and DNS for the whole service went down

   Backpackr2025

   Symptoms

   Services on the production Amazon EKS cluster behind the idus marketplace threw intermittent errors whose cause was hard to pin down. Calls to external services such as RDS and Redis failed now and then, and whenever DNS traffic spiked, domain lookups answered slowly or failed outright.

   DNS resolution failuresIntermittent connection failuresHigh latencyOOMKilledService outage

   Trigger

   All CoreDNS pods were scheduled onto the same node, and that node's memory usage spiked until it ran out of memory.

   Root cause

   Capacity exhaustion

   Component

   CoreDNSClient-side DNS resolver and cachekubeletContainer memory limitNode pool

   Contributing factor

   Unreviewed defaultsSingle point of failureNo caching

   Detection

   The write-up does not say how the full outage was noticed. The earlier DNS trouble showed up as intermittent service errors, and investigating failed calls to RDS and Redis led to DNS resolution failures.

   Remediation

   • deployed NodeLocal DNSCache as a DaemonSet in front of CoreDNS
   • added a PodDisruptionBudget (maxUnavailable 1) and topologySpreadConstraints across zones for CoreDNS
   • lowered ndots from 5 to 2 with dnsConfig in the Deployments

   Lessons

   The CoreDNS setup that EKS creates by default (two pods) can be enough for a small or development cluster but becomes a bottleneck in a large production cluster, and DNS is a single point of failure that needs high availability and room to scale. Tuning the CoreDNS cache and buffer size only relieved the symptoms; the real load came from the extra queries that ndots:5 generates for external names. A CoreDNS pod without a memory limit can share a node's fate with everything else on it.

   View case
2. IL-0085 A control plane node of Magic Pages' Kubernetes cluster on Hetzner Cloud ran out of memory, the cluster's own calls then used up the Hetzner API rate limit, and all customer websites were down for about 7.5 hours

   Magic Pages2025

   Symptoms

   At around 09:33 UTC one of the three control plane nodes of the Kubernetes cluster that hosts Magic Pages went offline. Looking at the node showed that it had run out of memory.

   Service outageNode NotReadyCloud API rate limit exhausted

   Trigger

   One of the three control plane nodes ran out of memory and went offline at about 09:33 UTC; the attempts to replace it with hetzner-k3s failed while the cluster's own components were exhausting the Hetzner API rate limit.

   Root cause

   Capacity exhaustion

   Component

   Cloud controller managerPersistent volumes and stateful sets

   Contributing factor

   Unknown limitRetry stormMonitoring blind spotLimited debug accessSingle point of failureNo runbook

   Detection

   The founder found the control plane node offline at about 09:33 UTC. Monitoring alerts about unreachable customer websites started around the time the replacement attempts failed, and about half an hour of investigation identified the exhausted Hetzner API rate limit.

   Remediation

   • powered off all servers of the Kubernetes cluster to stop the API requests so the Hetzner rate limit could recover
   • waited until ratelimit-remaining had built up a buffer before starting recovery
   • restarted the control plane servers
   • started the essential Longhorn components by hand so Longhorn could heal and make storage volumes available
   • restarted the remaining websites by hand

   Action items

   • Monitor the Hetzner API call rate coming from the Kubernetes cluster nodes themselves to get early warning of high usage
   • Get a higher Hetzner API rate limit; Hetzner support was contacted
   • Investigate how to reduce the cluster's reliance on a single API, including alternatives to hetzner-k3s
   • Document and test a procedure for recovering Longhorn after a full cluster shutdown

   Lessons

   The running cluster, not only the deployment tooling, consumed the cloud provider's API quota, and in a failure it consumed it fast enough that the quota never recovered. The default limit of 3600 requests per hour was not enough for failure and recovery scenarios, and API usage from inside the cluster was not monitored. A tool that hides the provider's API errors slows diagnosis. After a full, non-graceful shutdown, Longhorn needed manual intervention to come back, which lengthened the outage.

   View case
3. IL-0138 A failed Nvidia driver build left broken kernel packages on the only Kubernetes master node, which hit a kernel panic on reboot after a power outage and took the homelab cluster down for six days

   Barush Mendez2025

   Symptoms

   Every service in the homelab Kubernetes cluster stopped answering: the V-Rising server, Grafana, Jellyfin and Linkding were all unreachable, and the Kubernetes API was down, so the cluster could not be managed at all. The outage was found by hand, about 3 hours after it began, when someone tried to use the services.

   Control plane unresponsiveService outageMonitoring outage

   Trigger

   A power outage forced the Kubernetes master node to reboot.

   Root cause

   Software defect

   Component

   GPU driverLinux kernelKubernetes API server

   Contributing factor

   Ignored early signalChange with a delayed effectSingle point of failureMonitoring blind spot

   Detection

   Manually, when someone tried to access the services; there were no automated alerts.

   Remediation

   • Connected a monitor to the master node and read the boot logs, which showed the kernel panic
   • Booted the previous kernel version, which brought the system back
   • Removed the broken Nvidia packages with sudo apt-get remove --purge '^nvidia-.*' and ran sudo dpkg --configure -a
   • Found remaining broken packages with dpkg -l | grep -v ^ii and removed them by hand
   • Ran sudo apt-get install -f and reinstalled linux-headers-$(uname -r)
   • Rebooted and confirmed the kernel panic was gone

   Action items

   • Make sure GPU driver installations fully succeed, or revert immediately when they fail
   • Move the cluster to a high-availability control plane
   • Expose the Kubernetes API through a load balancer in the HA setup
   • Set up a ticketing system to record incidents and their resolution
   • Write predefined troubleshooting workflows to shorten response time
   • Add infrastructure monitoring such as Prometheus and Grafana
   • Add blackbox HTTP and TCP checks from outside the cluster for critical services

   Lessons

   A driver installation that fails half-way can leave kernel packages broken without any visible effect until the next reboot, which may come weeks later and at the worst time. A failed DKMS build should be fixed or reverted at once. With a single master node and no monitoring from outside the cluster, one unbootable machine meant a total outage that nobody was told about; the fix itself was simple once the boot logs were seen.

   View case
4. IL-0127 Llama 3 405B pre-training on 16K H100 GPUs was interrupted 419 times unexpectedly in 54 days, mostly by faulty GPUs, HBM3 memory and other hardware

   Meta2024

   Symptoms

   Over a 54-day snapshot of Llama 3 405B pre-training the training job was interrupted 466 times. 47 interruptions were planned: automated maintenance such as firmware upgrades, or operator actions such as configuration or dataset updates. The other 419 were unexpected. Because training is synchronous, a failure on a single GPU can force a restart of the whole job.

   Training job failureGPU unavailableNCCL timeout

   Root cause

   Hardware failure

   Component

   GPU hardware and memoryNCCLPyTorch

   Contributing factor

   Single point of failure

   Detection

   The communication library's state is monitored and a stalled collective is timed out automatically; on an NCCLX watchdog or heartbeat timeout the PyTorch NCCL flight recorder dumps its tracing data, including collective metadata and stack traces.

   Remediation

   • handled almost every interruption through automation, with significant manual intervention only three times in 54 days

   Action items

   • Reduce job startup and checkpointing time
   • Use the PyTorch NCCL flight recorder to record every communication event and collective duration and to dump traces on NCCLX watchdog or heartbeat timeouts
   • Turn on heavier tracing and metadata collection in production through online configuration changes, without a code release or job restart
   • Give PyTorch access to NCCLX internal state so stalls are detected and localised faster, with a snapshot of the failing collective's finished and pending transfers

   Lessons

   Training on 16K GPUs has more failure modes than Meta's much larger CPU clusters, and because training is synchronous a single GPU failure can force a restart of the whole job. Meta raised effective training time by cutting job startup and checkpoint time and by building tools for fast diagnosis, and kept it above 90% while automated firmware and kernel maintenance caused at least one interruption a day.

   View case
5. IL-0169 The Scaleway Kubernetes API in FR-PAR ran out of memory under simultaneous deployments, and because it also handled node authentication, Kapsule and Kosmos nodes went NotReady

   Scaleway2024

   Symptoms

   At 23:48 CET on 24 January 2024 the Scaleway Kubernetes API for the FR-PAR region ran out of memory. At the same minute Scaleway saw requests to its API Gateway spike.

   Node NotReadyOOMKilledIntermittent timeoutsMonitoring outage

   Trigger

   Simultaneous deployments in the FR-PAR region pushed the Scaleway Kubernetes API out of memory.

   Root cause

   Capacity exhaustion

   Component

   Container memory limitkubelet

   Contributing factor

   Single point of failureRetry storm

   Detection

   Increased load on the API Gateway was detected at 23:48, the minute the incident began, and the incident was escalated to on-call engineers at 00:00:10.

   Remediation

   • raised the memory limit and garbage collection thresholds of the Kubernetes managed services
   • deployed additional replicas of key services
   • purged the registry cache and monitored the clusters as they stabilised
   • adjusted service configurations after the incident to keep the platform stable

   Action items

   • Implement a local authentication cache
   • Further develop node authentication
   • Enhance monitoring, adjust rate limiting strategies and improve incident response protocols

   Lessons

   Because the Kubernetes API also handled node authentication, its memory exhaustion made some nodes NotReady, and the endless retries of those nodes added load on the API Gateway. Scaleway plans a local authentication cache and further work on node authentication.

   View case

By root cause11