Cases4 cases

1 to 4 of 4

1. IL-0489 Rejoining a failed node of Magic Pages' three-node MySQL cluster broke its metadata and took customer websites down

   Magic Pages2025

   Symptoms

   On 15 January 2025, from about 09:55 to 11:36 CET, a database outage at Magic Pages took down customer websites. It began when the founder tried to reconnect a failed third node to the high-availability MySQL cluster and the database server reported ambiguous errors about missing metadata information.

   Service outage

   Trigger

   One of the three nodes of the MySQL cluster went down, and reconnecting it failed with errors about missing metadata information.

   Root cause

   Dependency failure

   Component

   DatabasePersistent volumes and stateful sets

   Contributing factor

   Harmful manual actionSingle point of failureMonitoring blind spot

   Remediation

   • Restarted the storage server to fix PVC provisioning
   • Let the MySQL nodes restore and sync with each other for about 45 minutes to rebuild the cluster metadata
   • Cleaned up the corrupted cluster metadata
   • Redeployed MySQL Router

   Action items

   • Monitor the health of the storage system more closely
   • Improve the failover procedures for MySQL Router
   • Investigate and fix the status page so it shows reliable information

   Lessons

   The cluster survived the first node failure as designed and lost no data, but its metadata turned out to be a single point of failure. Better monitoring of PVC health could help prevent similar issues, the 45-minute recovery leaves room for improvement, and the status page has to detect failing customer sites reliably.

   View case
2. IL-0110 A Grafana Agent pod's high memory use and crash-looping Hangfire pods drove a Kubernetes node into system OOM and NotReady, twice in one day

   DFDS2024

   Symptoms

   At 10:01 UTC on 2024-03-13 a Kubernetes node entered NotReady state, with CPU requests at 78% and memory requests at 82%. An alert fired in Slack at 10:03 and was picked up at 10:05. An attempt to SSH into the node at 10:13 failed because it was unresponsive; it started responding again at 10:19.

   Node NotReadyContainer crash loopOOMKilledMonitoring outage

   Trigger

   Pods of a dev container that used Hangfire to trigger jobs failed with OOM and crash-looped on a node where a Grafana Agent pod was already using a lot of memory.

   Root cause

   Capacity exhaustion

   Component

   Telemetry agentContainer memory limitDatabase

   Contributing factor

   Excessive blast radiusNo resource requests or limitsHarmful manual actionMonitoring blind spot

   Detection

   A Slack alert fired two minutes after the node went NotReady and was picked up two minutes later; the second node's NotReady at 15:03 also raised a Slack alert.

   Remediation

   • marked the EC2 instance unhealthy so a new node started
   • cordoned and force-drained the NotReady node and terminated its EC2 instance
   • removed the new LimitRange from the Grafana Agent namespace and restarted the Grafana Agent pod
   • scaled the dev container to 0 replicas, after which EC2 network traffic returned to normal
   • after the second NotReady, scaled the dev container to 0 replicas again and added a new node
   • emailed the team that owns the dev container, which confirmed it was working on a fix

   Action items

   • Move grafana-agent to the monitoring node group
   • Add alerts for when bandwidth across the cluster reaches a certain threshold
   • Investigate whether memory limits can be set on the grafana-agent Helm chart
   • Investigate grafana-agent memory usage
   • Ask Grafana support whether grafana-agent should use a PVC
   • Consider pushing node logs to Loki with limited retention
   • Create a wiki page with common metrics and log queries to check when investigating issues with Kubernetes nodes

   Lessons

   Dev workloads mixed with production let one crash-looping dev container take down nodes that production also used, and a monitoring agent outside its dedicated node group added to the memory pressure. Scaling the suspect container back up before its owners had fixed it brought the failure back on the replacement node. The team also noted that action items from earlier post-mortems need follow-up, and that knowledge of useful CloudWatch metrics had not been shared, which made the diagnosis harder.

   View case
3. IL-0037 Image registry rate-limited a whole region after a crash-looping DaemonSet with imagePullPolicy Always pulled through three NAT instances

   Datadog2019

   Symptoms

   Users reported that their jobs were not starting, with some sign that image pulls were involved. Many pods across the cluster were in error states related to image pulls, while applications that were already running were not affected.

   Image pull failuresService outage

   Trigger

   A change to the permissions on a storage bucket that one application needed to read. The application could no longer start, and its DaemonSet, running on roughly a thousand nodes of the cluster, went into a crash loop with imagePullPolicy: Always, so every restart pulled the image again although it was already on the node.

   Root cause

   Capacity exhaustion

   Component

   Container registry and its credentialsNAT gatewayDatabaseAdmission webhook

   Contributing factor

   Automation amplificationSingle point of failureUnknown limitHarmful manual actionExcessive blast radius

   Detection

   Users reported that their jobs were not starting; monitoring had not caught it first.

   Remediation

   • replaced the affected NAT instances so registry traffic came from new source IPs
   • reconfigured the static IPs and firewall rules so applications could reach the Cloud SQL database again

   Action items

   • Add an admission webhook that rejects floating image tags, starting with latest
   • Apply the image tag webhook to workload objects only, not to pods, after it blocked pods created by in-cluster controllers

   Lessons

   The speakers' takeaway is to be very careful with DaemonSets in large clusters: they say they had broken everything DaemonSets depend on, including DNS, Vault and the image repository, and that it is very easy to break things when managing them by the thousand. imagePullPolicy: Always made every crash-loop restart pull the image again. Sending all registry traffic through three NAT instances put the whole region behind three source IPs, which the registry rate-limited, and changing those IPs cut the connections to a critical database reached over the internet through them. An admission webhook on pods rejects the pods that controllers in the cluster create, where no user sees the error.

   View case
4. IL-0506 UC Berkeley DataHub logins failed and students got empty home directories after the JupyterHub database was overwritten

   UC Berkeley2017

   Symptoms

   At about 1 AM on 9 February 2017, UC Berkeley's DataHub was reported down: users trying to log in got a proxy error. The hub pod was running, but its log was full of sqlite errors.

   Service outageData lossMisrouted requests

   Trigger

   While debugging, an operator ran `.save jupyterhub.sqlite` in sqlite3 inside the hub pod without having attached the database, which replaced the JupyterHub database file with an empty one.

   Root cause

   Operator error

   Component

   DatabasePersistent volumes and stateful setsJupyterHub

   Contributing factor

   No safeguard on destructive changeUnsafe failure behaviourHarmful manual action

   Detection

   DataHub was reported down at about 1 AM, when users got a proxy error at login. At 1:12 AM a user reported that he could log in but all his notebooks were gone.

   Remediation

   • Scaled the hub to 0 replicas
   • Rebuilt the JupyterHub sqlite database by matching Kubernetes PVC names back to user ids
   • Set cookie_id to null instead of a dummy value and attached the rebuilt database to the hub
   • Changed the cookie secret in the hub pod and restarted it to log all users out
   • Deleted all user pods to log everyone out
   • Ran a script that filled in cookie_id for all users, restarted the hub and deleted all user pods again
   • Dropped the new users with a '.' in their name and renamed the older users from '-' to '.'

   Action items

   • Report upstream that JupyterHub uses only part of the cookie secret and discards the rest
   • Report upstream that the cookie_id column of the users table should have UNIQUE and NOT NULL constraints
   • Ask KubeSpawner to support username hashes in PVC and pod names so the names stay the same when the database is deleted
   • Ask OAuthenticator to support using the id from the Google authenticator as the user id, giving stable user ids
   • Switch to Google Cloud SQL, a hosted and managed MySQL database
   • Take regular and tested backups of the database
   • Start an operational FAQ of things to do and not to do
   • Set up better monitoring and paging
   • Document escalation procedures explicitly

   View case

By root cause3