Harvest peak: half of the computation engine's pods stay Pending

Zitoun Analytics, a start-up in Meknes, computes yield forecasts and collection routes for 1,200 olive-growing cooperatives. Its engine moteur-recolte runs on the regional GKE Standard cluster gke-zitoun-prod (europe-west1, three zones), node pool pool-calcul with n2-standard-8 machines and autoscaling (3 to 30 nodes in total).

It is the harvest peak: since 13:45, the HPA has scaled the deployment up to 40 replicas. Only 22 are running, the others stay Pending; the afternoon routes are not computed and the trucks are waiting. Reda, the platform engineer, already “raised the pool maximum” once last week and thinks it needs raising again.

You are on the bastion, with kubectl pointing at gke-zitoun-prod (namespace calcul) and gcloud configured on the project zitoun-prod-7406.