Conditions and blockers

Inspect conditions and capacity status together:

kubectl --context YOUR_CONTEXT -n fleets get celldfleet my-fleet   -o json | jq '.status | {conditions, capacity}'

Signal	Meaning
Ready	Every replica of the observed workload generation is ready. Availability does not mean celld has finished recovering every session.
RoutingReady	Optional route status: current Gateway acceptance or an Ingress address. Independent of fleet readiness; not a DNS/TLS/connectivity test.
InfrastructureReady	Required Kubernetes objects match; Pods may still be Pending.
Blocked	A requested action cannot proceed; read its reason and message.
Progressing	Provisioning while a change rolls out or a down member waits to be replaced, or LifecycleProgress while the operator heals a member or deletes the fleet.
MaintenancePaused	Workload changes are suspended.
Deleting	A fleet deletion request is removing compute and, for PersistentFleet, disks.
OperationSizeWarning	The reservation’s capacity-policy history is nearing its bounded encoded-state limit.

Earlier releases also set DiskCleanupPending; it is removed as fleets reconcile. status.lifecycle is always empty. Editing status authorizes nothing.

Configuration blockers include InvalidConfiguration, NamespaceAccessDenied, ServiceAccountMissing, StorageClassMissing, InvalidStorageClass, IsolationUnverified and StorageScopeConflict. Fix the named prerequisite; do not bypass identity checks. Seeded fleets can also report SeedReservationBlocked, SeedInitializing and SeedCancellationBlocked; see preview seeding.

Both profiles converge drift in the workload’s template, replicas and update strategy, and in the NetworkPolicy and PodDisruptionBudget, back to the operator’s spec. Services stay verify-only; the one in-place change fills a field a newer release declares that the live object leaves unset, such as the peer port’s appProtocol. Objects without this fleet’s UID label, or with owner references, are refused: LifecycleBlocked for the workload, InfrastructureBlocked for prerequisites.

Reason	Meaning
Provisioning	Workload created, members rolling out, or a scale-in step waiting for the previous change to roll out. For PersistentFleet, also one member down and waiting for the replacement delay; the message gives the time it will be replaced on a fresh disk unless it returns. A member the scheduler cannot place is named with the scheduler’s reason.
Provisioned	The workload has rolled out and all replicas are ready.
LifecycleProgress	The operator took a self-healing action, named in the message: it force-deleted a Pod left on a node that no longer answers, or it is replacing a member on a fresh disk. Also reported while fleet deletion removes the workload and then, for PersistentFleet, every fleet PVC.
InfrastructureBlocked	A prerequisite object or the workload could not be created or converged.
LifecycleBlocked	The workload is not owned by this fleet.
StorageIdentityConflict	A PVC at a member’s name does not carry this fleet’s UID label; it is never adopted. Creation and growth wait.
CapacityUncertain	Automatic or External contraction lacks survivor-capacity evidence.
MemberUnschedulable	A PersistentFleet member has gone unscheduled for the replacement delay, and the operator does not replace it: another member is also down, or a fresh disk would not help. Blocked is true while the workload’s objects stay in place. The message quotes the scheduler. See a member that cannot be scheduled.
SchedulingBlocked	An Ordered Bucket zone scheduling gate could not be released.
MaintenancePaused	maintenance.paused suspends workload changes.
UnsupportedTransition	The runtime is not a digest-pinned fork pin.
DeletionBlocked	The workload to delete is not owned by this fleet.

The operator emits these Events on the fleet:

Event	Type	Meaning
MemberForceDeleted	Warning	A member Pod was still present more than two minutes after its termination grace ended, so its node no longer answers. The operator force-deleted it; the member returns on its own disk. PersistentFleet and Ordered Bucket.
MemberDiskLost	Warning	A member’s claim was Lost: its volume no longer exists. The operator deleted the claim and the Pod; the member returns on a fresh disk.
MemberReplaced	Warning	One member stayed down for the replacement delay while every other member had been ready for five minutes. The operator deleted its claim and Pod; the member returns on a fresh disk.
OperationSuperseded	Normal	The operator dropped an operation that an earlier release left in flight.

Other Events record changes to the Blocked condition’s status or reason. Earlier releases reported DiskRetired, LauncherBlocked and PodCompositionUnsupported; current releases do not. See self-healing.

Capacity reasons such as IncompleteMetrics, RepeatedSamples, PendingCapacity, IneffectiveCapacity, StabilizingOut, StabilizingIn, ObservingRedistribution and LoadNotRedistributed describe demand observation rather than durability. See capacity and troubleshooting.