Cases10 cases

1 to 5 of 10

1. IL-0160 A repaired legacy GPU monitoring script sent a firmware query that triggers a race condition in the GPU System Processor of NVIDIA RTX 4000 Ada GPUs, locking up GPU Linodes and LKE worker nodes in several regions

   Linode2026

   Symptoms

   From about 21:00 UTC on March 4, 2026, customers using Linodes backed by NVIDIA RTX 4000 Ada GPUs saw the GPUs lock up. At first the problem seemed limited to worker nodes of the Linode Kubernetes Engine (LKE); later Linode confirmed that every Linode with this hardware was affected, while Standard Compute and other GPU instances were not.

   GPU unavailableControl plane unresponsive

   Trigger

   Deployment of a repaired legacy GPU monitoring script, fixed during a routine improvement effort to restore a missing metric on internal observability dashboards.

   Root cause

   Faulty deployment

   Component

   Telemetry agentGPU driver

   Contributing factor

   Missed in reviewUnverified assumption

   Detection

   Customers saw the lockups from about 21:00 UTC; the first status post went up at 02:07 UTC on March 5, about 5 hours after the lockups began.

   Remediation

   • Disabled the monitoring script across the GPU fleet
   • Rebooted the affected nodes

   Action items

   • Take improvement and prevention measures to stop a recurrence; the postmortem does not list them

   Lessons

   A monitoring script can take down the hardware it watches. A fix meant only to restore a missing metric brought back a firmware query written for an earlier GPU generation, and on the RTX 4000 Ada architecture that query against an active GPU triggers a lockup. The script ran across the GPU fleet, so every Linode with this GPU was affected in several regions, and the investigation first suspected a regression in the host hypervisor or the GPU firmware.

   View case
2. IL-0134 A power event at Lambda's AUS01 data center rebooted 246 GPU nodes, all spine switches and the VAST storage, failing customer workloads and leaving stale NFS mounts on VMs

   Lambda2026sev1

   Symptoms

   At 21:06 UTC on 14 February 2026 every GPU node in Lambda's AUS01 data center rebooted at the same moment, without warning or plan, affecting multiple customers. The nodes came back online quickly, but the storage they used had disconnected, and customer workloads failed.

   GPU unavailableStorage disconnected or stale mounts

   Trigger

   A power event at the AUS01 data center caused a loss of power.

   Root cause

   Dependency failure

   Component

   Data center power and coolingInfiniBand fabricShared network filesystem

   Contributing factor

   –

   Remediation

   • restored the compute fabric and InfiniBand connectivity, bringing all spine switches and most leaf switches back online
   • restored part of the UPS redundancy
   • powered up all nodes in AUS01, completed physical checks and started the storage back up
   • restored Starbase and the other storage-dependent services in us-west-3
   • told affected customers to reboot their VMs, or the hypervisor if needed, to clear stale or empty NFS mounts

   Lessons

   A power loss that reboots the network spine and the shared storage together with the GPU nodes takes longer to recover than the nodes themselves. The GPU nodes were back quickly, but the fabric was restored first, storage was started only after physical checks, and some VMs were left with stale NFS mounts that customers were told to fix with a VM reboot or, if that was not enough, a hypervisor reboot.

   View case
3. IL-0171 Karpenter node auto-repair replaced EKS GPU nodes about 10 minutes after they became Ready, in a loop, because the Bottlerocket NVIDIA AMI lacked libdcgm.so and set AcceleratedHardwareReady to False

   Inflab2025

   Symptoms

   GPU nodes for an AI video upscaling job were created normally and reached Ready, and then about 10 to 11 minutes later Karpenter treated each one as unhealthy and replaced it. Every new node went the same way, so the GPU pods kept restarting and the upscaling work could not make progress.

   Node recreation loopGPU unavailable

   Trigger

   Enabling Karpenter's node auto-repair feature gate (nodeRepair: true) and then running GPU pods on nodes from the Bottlerocket NVIDIA AMI.

   Root cause

   Software defect

   Component

   KarpenterNode operating system imageGPU driver

   Contributing factor

   Automation amplificationUnverified assumption

   Detection

   The team noticed GPU nodes being replaced repeatedly after running GPU pods, then found the DCGMError message in the node events.

   Remediation

   • Disabled Karpenter node auto-repair by setting the nodeRepair feature gate to false

   Lessons

   Automatic node repair acts on every unhealthy node condition, including AcceleratedHardwareReady, so an image defect that only breaks GPU health monitoring is enough to have every GPU node terminated and recreated in a loop. Before turning on auto-repair, check which conditions it watches and whether the node image reports them correctly. Disabling auto-repair stops the loop at once but is only a temporary measure; installing the missing library or changing the AMI address the cause but cost more on an immutable OS such as Bottlerocket.

   View case
4. IL-0120 A daily scheduled OS update restarted systemd-networkd on cloud-hosted GPU nodes, which conflicted with OpenAI's networking agent and removed all routes, cutting capacity for ChatGPT and the API

   OpenAI2025

   Symptoms

   At 11:36 PM PDT on 9 June, alerts fired across multiple OpenAI services as a large number of cloud-hosted GPU nodes lost network connectivity, and the engineering team was escalated to immediately. Losing those nodes cut the capacity available to serve ChatGPT and the API.

   High error rateGPU unavailable

   Trigger

   A routine daily scheduled update of the host operating system on the cloud-hosted GPU servers restarted systemd-networkd on the affected nodes.

   Root cause

   Breaking change in an upgrade

   Component

   systemd-networkdNode operating system image

   Contributing factor

   No staged rolloutLimited debug access

   Detection

   Alerts fired immediately across multiple services when the infrastructure began losing network connectivity at 11:36 PM PDT, and the engineering team was escalated to at once.

   Remediation

   • re-imaged the affected GPU virtual machines
   • deployed recovery automation to restart the remaining impacted nodes
   • halted the background update mechanisms on the GPU VMs

   Action items

   • Disable automatic daily updates on GPU VMs so updates run in a controlled, scheduled way
   • Update system configurations to prevent conflicts between systemd-networkd and the networking service
   • Audit VM configurations across the fleet to find and reduce similar risks
   • Improve recovery speed for critical infrastructure such as GPU VMs and clusters
   • Run regular disaster recovery drills

   Lessons

   An unattended operating system update that restarts systemd-networkd can wipe the routes another networking agent manages, and it hits every node that takes the update at about the same time. Without break-glass tooling to restore connectivity, nodes that lost their network had to be re-imaged, which stretched recovery over many hours.

   View case
5. 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

By root cause5