Urgent.News

What's breaking now, across thousands of outlets.

Tech

We set maxUnavailable to zero, the platform team set a pod quota, and no release could start

Our checkout service runs twenty replicas, and its rolling update has maxUnavailable set to zero and maxSurge at twenty five percent. Zero unavailable was a rule we were proud of: a release never takes capacity away. It adds five new pods, waits for them to be ready, then removes five old ones. In October the platform team introduced resource quotas on every namespace, after one team's batch job…

Our checkout service runs 20 replicas, with rolling updates set to allow for a maximum of zero unavailable pods and a maximum surge of 25%. This rule was considered a priority to ensure releases never consumed capacity. In October, the platform team introduced resource quotas for each namespace, after a team's batch job filled the cluster. To prevent conflicts, each quota was determined by the actual running pods that week.

Checkout received a limit of 20 pods and a CPU request that matched. No changes were made on the day, as nothing was created. On Thursday, a fix for a voucher bug was deployed. The new ReplicaSet attempted to create five pods, but the quota admission refused each one, stating that the namespace already had 20 pods. With maxUnavailable at zero, the deployment couldn't remove an old pod until a new one was ready, and no new pod could exist without removing an old one. Neither side could progress.

The rollout timed out after 10 minutes, reporting a failure in the pipeline. The issue was traced back to the new ReplicaSet's events, where a FailedCreate error was logged. Unfortunately, this error was often overlooked during releases, and it took over an hour to identify. The voucher fix was delayed by two hours after the quota was manually increased.

Two settings, each intended to ensure release safety, ultimately prevented releases from starting. Neither team could claim responsibility for the combined issue. As a result, quotas are now calculated based on each deployment's replicas plus its surge, rather than the current cluster state. The platform repository now computes this from deployment manifests, and the pipeline checks quota availability before allowing a rollout to start.

Any failures include the warning events for the new ReplicaSet, providing immediate visibility into the reason for rejection. Sergey Shinder emphasized that steady-state configurations are not suitable for environments that constantly require changes.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in Tech

Before You Batch-Convert a Folder of Videos, Test One File from Each Source

Suppose your project folder contains phone clips, camera footage, and a screen recording. You want them all to work in one editor or upload destination.

  • Test one file from each source type before batch conversion
  • Check destination requirements for file formats, codecs, dimensions
  • Verify pilot file meets requirements before processing entire group

Will Your Photo Come Out Sideways? A Python EXIF Orientation Preflight Before You Pay for a Bama Bound Clip (Tested)

Checked November 17, 2026 (15:19–15:21 CST, UTC+8). I build bamabound.app , an independent studio that turns one photo of one person into a vertical clip of them holding a red I'M BAMA BOUND banner.

  • Studio bamabound.app creates vertical clips of people holding a "I M BAMA BOUND" banner
  • JPEG, PNG, or WebP formats accepted, max 10 MB, must display person's face
  • EXIF Orientation tag can cause photos to appear sideways if not adjusted

Fitting a raw-to-scaled score curve without an official table (Digital SAT)

College Board does not publish a raw-score-to-scaled-score conversion table for the Digital SAT. On the old paper test they did, one table per form.

  • College Board lacks official raw-to-scaled conversion table for Digital SAT
  • Developed method using practice tests and total-group report data
  • Output provides total score with error bar ranging ±30 to 60 points

More from Tuesday 17 November →