Cases43 cases

1 to 5 of 43

1. IL-0179 With DEBUG logging and synchronous log output, a high-traffic service filled the 64 KB stdout pipe, so request threads blocked in write() while holding database connections and HikariCP ran dry, failing requests in 30-second bursts

   MUSINSA2026

   Symptoms

   The on-call engineer for a high-traffic service was alerted that requests were failing in large batches every 30 seconds.

   High error rateIntermittent timeoutsConnection pool exhaustion

   Trigger

   DEBUG level logging was on in production when traffic surged, raising log volume to about 100 times normal while the service logged synchronously.

   Root cause

   Configuration change

   Component

   Application logging libraryContainer runtimeTelemetry agentDatabase

   Contributing factor

   Unverified assumptionExcessive blast radius

   Detection

   The on-call engineer was alerted that requests were failing in large batches every 30 seconds.

   Action items

   • Replace synchronous logging with TraceAwareAsyncAppender, which captures trace_id and span_id on the calling thread and drops logs instead of blocking when the queue is full (neverBlock=true)
   • Keep the amount of log output at a sustainable level, since an asynchronous queue still fills under sustained overload

   Lessons

   Removing AsyncAppender to keep trace IDs also removed the protection that kept request threads away from stdout back-pressure. A company-wide standard applies the same side effect to dozens of services, so check what a standard rests on and what working behaviour it takes away. ThreadLocal context does not cross into an asynchronous worker, but capturing it on the calling thread before handing the event off keeps both the IDs and the asynchronous output. Under extreme load, dropping log lines is preferable to stopping request handling.

   View case
2. IL-0058 Argo CD deleted namespaces shared by several applications, leaving about 97 applications stuck Missing and OutOfSync

   Argo CD users2026

   Symptoms

   After a rolling restart of the Argo CD application-controller, namespaces that several applications shared went into the Terminating state. About 97 applications were then stuck in a Missing/OutOfSync deadlock.

   Resources deleted

   Trigger

   A manual bulk sync with force: true from the Argo CD UI, run during the window after an application-controller restart, on an application whose manifests declared a namespace whose app.kubernetes.io/instance label pointed to a different application.

   Root cause

   Software defect

   Component

   Argo CD

   Contributing factor

   No safeguard on destructive changeUnverified assumptionHarmful manual action

   Action items

   • Refile the problem as #29739: a manual force sync hard-replaces shared cluster-scoped resources through a tracking-ID conflict, ignoring Prune=false

   Lessons

   The sync option Prune=false only protects the prune path; it does not stop a force sync from deleting and recreating a resource. When two Argo CD applications declare the same namespace or other cluster-scoped resource, force-syncing the application that does not own the tracking label can delete it at any time.

   View case
3. IL-0470 Buildkite notifications were delayed after Prometheus storage filled and EKS notification workers could not autoscale

   Buildkite2026

   Symptoms

   From 20:01 UTC on 28 May 2026, Buildkite build and job notifications (GitHub commit status, email, Slack and webhook notifications) reached a subset of customers late. Internal queue latency monitoring showed the delay as queues grew while notification-processing workers stayed underprovisioned.

   High latencyAutoscaler misbehaviour

   Trigger

   The Prometheus service used by Buildkite's EKS autoscaling path ran out of storage, so some EKS-based notification workers could not autoscale while their queues were growing.

   Root cause

   Capacity exhaustion

   Component

   Prometheus

   Contributing factor

   No alert on limitPartial rolloutIneffective rollbackUnverified assumptionHarmful manual action

   Detection

   Internal queue latency monitoring detected the rising notification delivery latency.

   Remediation

   • Moved the affected notification-processing workloads from EKS back to the ECS-based infrastructure
   • Adjusted scaling configuration and worker counts by hand
   • Completed the rollback and scaled the remaining affected ECS services
   • Increased Prometheus storage capacity and recorded the change in infrastructure as code

   Action items

   • Alert before Prometheus storage exhaustion can affect autoscaling
   • Fix GitHub notification retry behaviour for a class of errors that could cause repeated retries
   • Make the EKS-to-ECS rollback verify destination capacity, autoscaling configuration and traffic movement
   • Review the CPU, memory and storage allocations of EKS control-plane dependencies such as KEDA and Karpenter
   • Move customer-critical workloads to EKS more gradually, with clearer settling periods
   • Improve customer-level monitoring of notification delivery latency to detect customer-impacting regressions earlier
   • Review which notification types sit on customers' scheduling or CI hot path and need tighter latency, separate queueing or specific alerting
   • Update the status page earlier when notification latency is likely to affect customer workflows
   • Describe the customer-visible impact in status page updates, not only the affected internal service
   • Improve internal escalation for customers who report critical CI impact

   Lessons

   Buildkite said recovery took longer than expected because the rollback path from EKS to ECS did not fully handle this failure, and that its public status page did not show the customer-visible impact as quickly or clearly as it should have, since notification latency affects customers differently depending on how their CI and deployment workflows use notifications.

   View case
4. IL-0097 A cluster autoscaler version bump was incompatible with the one cluster still on Kubernetes 1.33, so no scale-up events fired and new nodes were not created

   2i2c2026

   Symptoms

   At 15:10 (Bucharest time) on 2026-03-31 an alert fired for the prod hub on the berkeley-geojupyter cluster on AWS: "[FIRING:1] At least two servers failed to start in the last 30m cluster=berkeley-geojupyter hub=prod (provider=aws immediate action needed)". It opened PagerDuty incident #2057.

   Scale-up failure

   Trigger

   An upgrade merged on 2026-03-30 bumped the cluster autoscaler version, and the new version was incompatible with Kubernetes 1.33, which the berkeley-geojupyter cluster still ran.

   Root cause

   Breaking change in an upgrade

   Component

   Cluster autoscaler

   Contributing factor

   Deferred upgradesUnverified assumption

   Detection

   An alert that at least two servers failed to start in the last 30 minutes fired at 15:10 and opened PagerDuty incident #2057. The report notes it was triggered by an automatic health check run, not by users being unable to start servers.

   Remediation

   • upgraded the cluster's Kubernetes version from 1.33 to 1.34
   • downgraded the cluster autoscaler by one patch version to 9.55.0, the version meant for Kubernetes 1.34

   Action items

   • The change in https://github.com/2i2c-org/infrastructure/pull/8008 (the report gives only the link)
   • Use recurrent issues, such as the recurrent-k8s-aws-upgrades workflow, to standardize Kubernetes versions across clusters

   Lessons

   Because the context of the autoscaler bump was fresh, the team went down the right path from the start. The upgrade batches had been designed with only 1.32 clusters in mind, so the one 1.33 cluster was missed and got an incompatible autoscaler. 2i2c notes it was lucky that an automatic health check, not real users, hit the failure first.

   View case
5. 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

By root cause10