Cases19 cases

12 to 16 of 19

1. IL-0193 A GCP persistent disk degradation took down a Cassandra StatefulSet whose OrderedReady policy blocked recovery, and Metrictank could not reconnect to the recreated pods, causing a 23-hour Graphite query outage

   Grafana Labs2019

   Symptoms

   An alert fired because some Metrictank instances could not drain their write queue, which meant they could not write to Cassandra. Google had not yet published an incident, but a team member's Discord server, also hosted on GCP, was down, which pointed at GCP; Google confirmed an SSD persistent disk incident about 90 minutes after it began.

   High error rateNode NotReadyPods PendingService outage

   Trigger

   A Google Cloud incident in which SSD persistent disks performed so poorly that many I/O operations failed for an extended period.

   Root cause

   Cloud provider outage

   Component

   Persistent volumes and stateful setsDatabaseNode root disk

   Contributing factor

   Overlapping failuresUnpatched known bugUnverified assumption

   Detection

   An alert that Metrictank write instances could not flush their write queue to Cassandra. Error rates alone rose too slowly to notice quickly, because most queries were served from memory.

   Remediation

   • Added compute nodes hoping new nodes could attach the disks, which did not work
   • Deleted and recreated the Cassandra StatefulSet with an updated pod management policy, so all pods but one started
   • Raised Kafka topic retention to avoid data loss during the long downtime
   • Temporarily added nodes so Metrictank restarts were not slowed by CPU starvation
   • Restarted all Metrictank instances, which took more than 20 hours, so they connected to the new Cassandra IPs

   Action items

   • Re-instantiate the gocql connection in Metrictank when it is detected dead, so it reconnects even when all Cassandra IPs change
   • Fix the Metrictank write-queue deadlock by updating an internal data structure with atomics instead of a lock

   Lessons

   An OrderedReady StatefulSet cannot recover past a single pod whose volume will not attach; when every pod is down, the ordering that protects normal start-up blocks the whole cluster. A client that learns peer addresses only from live peers needs a way to go back to the service name when all peers are replaced at once. A long-tolerated bug with a config workaround can surface at the worst time; this one turned out to have a simple fix. Writing incoming data to a Kafka write-ahead log first meant all data could still be ingested after recovery, and Kafka on non-SSD disks escaped the disk incident. With both Metrictank bugs fixed, the team expects a similar event to take less than 3 hours to recover from instead of 23.

   View case
2. IL-0039 New Kubernetes worker nodes could not join any cluster because a package URL hardcoded in kops nodeup broke

   FREE NOW2019

   Symptoms

   At 16:00 CEST a developer reported in the #ask-platform Slack channel that a deployment to the Prelive cluster was failing, and a second report followed at 16:06.

   Pods PendingScale-up failure

   Trigger

   Roughly two to three days earlier, the CentOS repository mirror was restructured and container-selinux-2.68-1.el7.noarch.rpm moved, which made the static download URL hardcoded in kops nodeup invalid.

   Root cause

   Dependency failure

   Component

   Node operating system imageNode pool

   Contributing factor

   Monitoring blind spot

   Detection

   Developers asked for help in the #ask-platform Slack channel because their deployments were failing.

   Remediation

   • downloaded the package by hand on the Prelive nodes so their bootstrap could finish
   • changed the cluster AMI from Amazon Linux 2 to Debian and applied it to all clusters

   Action items

   • Create an alert that shows when nodes cannot join the cluster, possibly from information the cluster autoscaler already has
   • Document the external dependencies of node startup and look for ways to reduce them
   • Change the AMI back to Amazon Linux 2 Minimal once kops fixes the issue
   • Document node troubleshooting better: what happens during bootstrap and where the logs are

   Lessons

   Applying cluster changes through kops and Jenkins worked well, and troubleshooting quickly showed that the failing deployments came from missing cluster capacity. What went wrong was that nobody noticed, over a long period, that some autoscaling groups held more worker instances than Kubernetes had nodes. The team was lucky that no large number of production spot instances was terminated during the problem and that it was noticed during office hours.

   View case
3. IL-0014 A GKE upgrade to a withdrawn version stalled on slow drains and a stuck node pool, then broke an ingress and confused the HPA

   loveholidays2019

   Symptoms

   On the evening of a long-overdue upgrade of a loveholidays production GKE cluster from 1.10.X to 1.12.5-gke.10, planned to take 2 to 2.5 hours, the master upgrade and the first two 7-node NodePools went smoothly. After that the NodePool upgrades slowed to a crawl. Each node is cordoned, drained and upgraded in turn. Even a 14-node cluster at about 5 minutes per node took over an hour, and with about 100 more nodes to go, some of them taking over 15 minutes to drain, the sequential drains were too slow to finish at a reasonable hour.

   Upgrade stallPods PendingScale-up failureService outageAutoscaler misbehaviour

   Trigger

   A manual upgrade of the GKE masters and NodePools from 1.10.X to 1.12.5-gke.10, two versions at once, on 19 March 2019; GKE's release notes of 14 March had already said that version was no longer available for new clusters or master upgrades because masters showed elevated error rates, but it could still be selected.

   Root cause

   Cloud provider outage

   Component

   GKE control planeNode poolCluster autoscalerIngress controllerHorizontal pod autoscaler

   Contributing factor

   Unrepresentative stagingOverlapping failuresIgnored early signal

   Detection

   The slow drains, zombie nodes and stuck NodePool were seen by the engineer running the upgrade; the broken Ingress was noticed after the upgrade and explained by Google support; the HPA problem was noticed the next day.

   Remediation

   • drained the remaining NodePool nodes ahead of the upgrade with kubectl drain --ignore-daemonsets --delete-local-data
   • deleted the zombie instances from the managed instance group with gcloud compute instance-groups managed delete-instances
   • cancelled the stuck UPGRADE_NODES operation with gcloud beta container operations cancel
   • used kubectl port-forward in place of the broken internal Ingresses
   • scaled down the older of two active ReplicaSets, or triggered a dummy deployment, to bring the HPA back

   Action items

   • Alert when a HorizontalPodAutoscaler is at 70% and at 100% of its capacity
   • Alert when a deployment has had two active ReplicaSets for 30 minutes and for 60 minutes
   • Re-evaluate termination grace periods that were set to arbitrary large values
   • Read the GKE release notes before every cluster upgrade
   • Use auto-upgrade in non-production environments, and manual cluster upgrades with automatic node upgrades in production
   • Keep environments as identical as possible, including resource names

   Lessons

   Read the release notes before upgrading: the target version had been withdrawn five days earlier. Drain time depends on pod configuration, so emptyDir volumes, long grace periods and missing disruption budgets turn a NodePool upgrade into hours. Pre-production only catches problems like the ingress bug if it matches production down to resource names. Auto-upgrades in non-production surface problems early, and a mix of manual cluster upgrades and automatic node upgrades in production keeps clusters current without much effort.

   View case
4. IL-0466 Prow scheduled test job pods onto its service cluster, leaving hundreds Pending and starving Prow's own services

   Kubernetes project2019

   Symptoms

   On 8 February 2019 the Kubernetes project's Prow instance at prow.k8s.io became very slow: its controllers took about five minutes to respond to the hook, bot commands such as the cat command stopped working, and the prow.k8s.io site was hard to reach. GitHub status was normal, and the Prow service nodes were found running above 100% of CPU and memory.

   Pods PendingSustained high node CPUHigh latency

   Trigger

   At 13:20 PST an auto-generated bump PR for Prow was merged and deployed, and Prow started scheduling workloads onto the service cluster.

   Root cause

   Faulty deployment

   Component

   CI system

   Contributing factor

   Unrepresentative stagingNo resource requests or limitsLimited debug access

   Detection

   While working on a PR, a maintainer noticed the bot's cat command failing, found GitHub status normal, learned on Slack that the Prow nodes were above 100% CPU and memory, and then found 600 pending pods in the test-pods namespace.

   Remediation

   • Rolled Prow back to v20190206-b01b288 and cleaned up the pods and ProwJobs in the service cluster
   • Applied the revert by hand with kubectl apply after the deployment permission was missing
   • Sent a fix, PR 11209, so the client no longer overwrites a provided literal default client
   • Rolled Prow back to v20190125-2aca69d, from before the bug was introduced
   • Cleaned up the wrongly scheduled pods and ProwJobs again and confirmed Prow was healthy

   Action items

   • Find out why the bug did not show deterministically
   • Be more careful with sweeping Prow changes, with more end-to-end tests or a real staging Prow

   Lessons

   The behaviour was not deterministic, and the change was deployed on a Friday afternoon. Prow instances are set up differently, for example in how build clusters are defined, so another project's Prow running a version fine does not show it is safe for the Kubernetes Prow, which needs a staging environment that looks like it.

   View case
5. IL-0504 mybinder.org launches failed for about five hours as undeleted user pods piled up and filled the cluster

   Binder Project2018

   Symptoms

   On Saturday 8 July 2018 at about 18:30 CEST, the launch success rate of mybinder.org dropped to zero and stayed there. Suspicious behaviour was first reported on Gitter at 22:48, more than four hours later.

   Service outagePods PendingStale or inconsistent API reads

   Root cause

   Software defect

   Component

   JupyterHub

   Contributing factor

   No alert on limit

   Detection

   Suspicious behaviour was reported on Gitter at 22:48 CEST, more than four hours after launch success dropped to zero; no downtime notifications were set up.

   Remediation

   • Deleted user pods older than four hours with the delete-pods.py script
   • Culled pods of the most-used images, such as ipython-in-depth, more aggressively
   • Restarted the Hub pod after about 600 pods were culled, which brought launches back to normal
   • Cordoned some of the extra nodes so the autoscaler could reclaim them
   • Ran manual max-age culling the next morning, letting the cluster scale back to 3 nodes

   Action items

   • Set up notifications of downtime
   • Automate cluster scale-down
   • Document the downtime recovery steps so every team member knows them
   • Fix KubeSpawner's failure to delete pods when the pod reflector fails
   • Make KubeSpawner more robust to pod reflector failure
   • Refactor the event reflector into a single reflector instead of one per pod
   • Detect a likely unhealthy Hub state and restart the Hub without manual intervention

   Lessons

   The delete-pods.py script was useful for bulk-deleting old pods, and once the problem was noticed the cluster was healthy within a few minutes. The outage lasted five hours before anyone noticed because no downtime notifications existed, and scaling the cluster back down still took several manual steps of cordoning and draining nodes.

   View case

By root cause6