Cases13 cases

8 to 12 of 13

1. IL-0084 Packet drops on a bare metal provider's vSwitch forced Namespace to rebuild its fra3 region and left global capacity short, and the next day slow NVMe writes on a control plane node drove the Kubernetes API server to very high CPU

   Namespace2024

   Symptoms

   From 17 October 2024, packets through the vSwitch that joined Namespace's bare metal machines at one provider in the fra3 region were dropped at random. Retries in TCP, QUIC and the Namespace control plane hid most of it, but the SLO error budget for latency targets on key interactions began to burn.

   Intermittent connection failuresHigh latencyControl plane lagService outage

   Trigger

   Random packet drops on the bare metal provider's vSwitch from 17 October 2024; for the second outage, a rate of Kubernetes changes that the slow NVMe on one control plane node could no longer keep up with.

   Root cause

   Cloud provider outage

   Component

   Kubernetes API server

   Contributing factor

   No capacity planningNo headroomOverlapping failures

   Detection

   SLO-based monitoring of latency targets showed the error budget burning while retries hid the individual packet drops.

   Remediation

   • diverted all customer traffic from fra3 to other regions
   • connected each provider machine to the zrh1 cluster over Wireguard instead of the vSwitch
   • repaved every fra3 machine with a new identity, certificates and encryption-at-rest keys, dropping regional caches to recover faster
   • cloned existing datasets in a more distributed way to speed up repaving the machines
   • added machines back to the new region in batches and returned the region to the compute pool
   • redirected workloads away from zrh1 during the second outage
   • cloned the failing control plane node's key state to a higher-capacity node that took over its role and IP address

   Action items

   • Credit all instances that ran during the outages back to customers
   • Build up backup regions and solutions so a similar event has options that do not affect customers

   Lessons

   Running hot with a delayed hardware delivery left no room to take a region out, so fixing a provider network fault turned into a global capacity shortage. After one outage the team was anchored on its cause and first looked for network problems in the second outage, which had a different cause: a control plane node whose disk could not keep up with the Kubernetes change rate.

   View case
2. IL-0495 HubSpot's internal load balancers on Kubernetes were falsely marked unhealthy and the rest collapsed under the traffic

   HubSpot2024

   Symptoms

   Between 14:20 and 15:16 UTC on 27 August 2024, some HubSpot customers could not log in to their accounts, reach their content or use HubSpot tools. Requests were slow or failed because HubSpot's internal load balancers, which run on Kubernetes and spread traffic across the backend services, were overloaded.

   Probe failuresHigh latencyHigh error rate

   Trigger

   Routine operation of a critical component of HubSpot's traffic management system, from about 14:08 UTC, ran path handling updates in the load balancer stack, and an issue in that stack made those updates block health check responses.

   Root cause

   Software defect

   Component

   Ingress controllerLiveness and readiness probes

   Contributing factor

   Automation amplificationNo headroom

   Detection

   Monitoring detected widespread performance degradation across the platform at 14:23 UTC and the incident response team was mobilised. HubSpot says it had identified the issue internally by 14:20 UTC. The status page was not updated until 14:44 UTC, which left customers with outdated information.

   Remediation

   • Tried to increase the capacity of the remaining load balancer instances
   • Brought new load balancer instances online, which relieved the pressure and let traffic return to normal

   Action items

   • Change the health check configuration so health checks do not time out falsely
   • Redesign the health check system so health checks get priority and keep running during operations such as configuration updates
   • Increase the capacity of the load balancers for sudden traffic spikes
   • Change the load balancing architecture to detect traffic surges earlier and distribute load dynamically in real time
   • Update the status page automatically for large-scale outages
   • Integrate the monitoring systems more tightly with the communication tools
   • Expand load testing to simulate high-traffic scenarios
   • Introduce chaos engineering to test platform resilience regularly
   • Add traffic pattern metrics, including the ratio of current connections to the maximum allowed
   • Add auto-scaling that adjusts resource capacity to traffic patterns

   Lessons

   HubSpot concluded that earlier detection of the abnormal traffic surge could have prevented the cascading failure, and that health checks have to keep running during operations such as configuration updates so that working load balancers are not removed together. The late status page update caused confusion for customers.

   View case
3. IL-0044 Prometheus OOMKilled in a loop while replaying its write-ahead log after a kube-prometheus-stack upgrade

   Saifeddine Rajhi2024

   Symptoms

   After an upgrade of the kube-prometheus-stack Helm chart, several pods stayed not ready. Among them was prometheus-prometheus-operator-prometheus-0, showing 3/4 Running and a recent termination.

   OOMKilledProbe failuresMonitoring outage

   Trigger

   During the kube-prometheus-stack Helm chart upgrade, Prometheus had to replay its write-ahead log on startup, and the replay needed more memory than its limit allowed.

   Root cause

   Capacity exhaustion

   Component

   PrometheusContainer memory limit

   Contributing factor

   Unknown limitNo headroom

   Detection

   The team noticed pods not ready after the Helm chart upgrade and found the OOMKilled termination with kubectl describe pod.

   Remediation

   • ran promtool tsdb analyze inside the Prometheus pod to find high-cardinality metrics such as haproxy_server_http_responses_total

   Action items

   • Verify that the memory overhead during WAL replay stays within acceptable limits, for example 10 to 15%, and do not run Prometheus close to its memory limit
   • Optimise storage, scraping and remote write configuration and upgrade to newer Prometheus releases
   • Recover automatically from OOM crash loops, for example by deleting the WAL on OOM events
   • Enforce scrape limits so Prometheus stops scraping targets when memory use passes a threshold

   Lessons

   Size Prometheus for its restart, not only for steady state: WAL replay can need far more memory than normal running, and a Prometheus that already runs near its limit can be OOM-killed during the replay. Much of Prometheus memory use comes from cardinality, which promtool tsdb analyze can measure even when Prometheus itself is down.

   View case
4. IL-0070 Critical pods on core nodes evicted when Prometheus began scraping about 200 nodes after a dask-gateway scale-up

   2i2c2023

   Symptoms

   The hub pod of the prod JupyterHub on the leap cluster (leap.2i2c.org) restarted unexpectedly during a large node scale-up, and a user opened a support ticket. The ingress-nginx and proxy pods kept running.

   Pod evictionPods Pending

   Trigger

   A user's dask-gateway workload added about 100 dask-worker nodes within a minute, and Prometheus started scraping the metrics exporters on about 200 nodes.

   Root cause

   Capacity exhaustion

   Component

   PrometheusNode pool

   Contributing factor

   No headroomExcessive blast radiusMonitoring blind spot

   Detection

   A user support ticket (2i2c Freshdesk ticket 414) about the leap hub; the engineer then looked at Grafana and the cluster's nodes and events.

   Action items

   • Configure CPU and memory requests for the hub and proxy pods (#2127; at least 100m CPU was proposed)
   • Make the Grafana OOMKiller dashboard work (#2213)
   • Move LEAP cloud infrastructure to shared nodes (#2209)
   • Update resource requests and limits, mainly for ingress-nginx and Prometheus (#2324)
   • Decide whether Prometheus should scrape dask worker pods (#2366)

   Lessons

   Prometheus load grows with the number of nodes, so a burst of autoscaled nodes can push a nearly full core node over the edge and evict the pods that matter most. Core pods that request 10m of CPU get a tenth of the CPU share of busy neighbours that request 100m. Kubernetes events expire after an hour, so evidence of OOM kills needs a dashboard that keeps it.

   View case
5. IL-0115 One of OpenAI's three production AKS clusters went offline through a human error, and rebuilding it took 10½ hours because an AKS Kubernetes upgrade broke the ingress and AKS cannot be downgraded

   OpenAI2022

   Symptoms

   At 18:56 PDT on 23 June 2022 (01:56 UTC on 24 June) several models went offline at once: some fine-tuned curie and davinci models, the Codex beta models code-davinci-001 and code-cushman-001, the legacy GPT models curie, babbage and ada, and all Embeddings models.

   Service outageHigh error rateIntermittent timeouts

   Trigger

   A destructive operation on one of the three production clusters, caused by human error that the existing safeguards did not catch, took the whole cluster offline.

   Root cause

   Operator error

   Component

   Ingress controllerModel hub

   Contributing factor

   No safeguard on destructive changeNo headroomDeferred upgradesOverlapping failures

   Remediation

   • re-allocated the models from the lost cluster to the remaining two clusters
   • rebuilt the VPN link with Azure support after losing the connection to AKS during cluster setup
   • built workarounds with Azure support after losing the connection to Azure Private Link
   • tore down the cluster and rebuilt it from scratch after the upgraded Kubernetes version broke the ingress
   • moved traffic back into the rebuilt cluster between 05:30 and 06:30 PDT

   Action items

   • Add checks and locks so that this and similar production-critical infrastructure cannot be modified by accident
   • Add better visibility into the operations being performed and make parts of the system more lexicographically distinct
   • Keep working with Azure on the AKS network connectivity issues so they do not disrupt new cluster operations
   • Update playbooks with the new cluster failure modes and change the setup steps for new clusters and infrastructure
   • Upgrade Kubernetes systematically and set aside time so the clusters do not fall behind again
   • Shard fine-tuned models across multiple clusters and work towards more automatic cross-cluster replication
   • Distribute and replicate Azure Blob Storage so fine-tuned models load faster and more reliably
   • Finish rolling out node-to-node caching to relieve Azure Blob Storage bottlenecks
   • Change how model weights are stored so they are more efficient and faster to load

   Lessons

   Losing one of three clusters exhausted the GPU headroom, and GPUs could not be added on short notice, so full recovery depended on bringing the same cluster back. Rebuilding it exposed a deferred Kubernetes upgrade: the newer version broke the ingress, and because AKS cannot downgrade there was no way back except a full rebuild. Recovering many large models at once is limited by the bandwidth of the storage they load from, and the caches that normally hide this do not help when everything moves to a different cluster.

   View case

By root cause7