The direct answer
Declare how the worker consumes configuration before deciding what an update means. For the narrow profile below, reference one reviewed immutable ConfigMap revision through the Pod template’s environment inputs, and review a new reference for the next release. A ConfigMap update or rollout status does not establish which complete configuration an individual worker or job actually used.
Choose one configuration path for a fixed worker
A fictional preprocessing worker W-A produces bounded plain-text summaries under a trusted executable and a complete settings revision C-A. The application needs to distinguish “new settings published” from “this job used those settings.” Those are different observations.
This profile selects one ordinary, effectively unpaused Deployment application container, one namespace and environment-only consumption of one application-owned ConfigMap. C-A contains non-confidential preprocessing settings. It carries no prompts, credentials, customer records or secrets. Trusted code owns the key mapping and interprets the values under a separately reviewed contract.
All eight cards below are PROPOSED_UNRUN. No manifest, cluster, ConfigMap, Pod, environment, mounted file or rollout was created or inspected. Actual support, ownership, complete configuration, worker association, bounds and current rights remain unimplemented prerequisites; unknown evidence blocks real admission.
Separate three delivery behaviors
The official ConfigMap guide describes non-confidential key-value data and requires the referenced ConfigMap and Pod to share a namespace for the selected consumption path. It distinguishes updates to volume-projected keys, which are eventual, from environment consumption, which does not automatically update and requires a Pod restart. A subPath mount does not receive ConfigMap updates.
Therefore an observation about one path cannot be borrowed for another. A changed file projection is not evidence that a process reread or accepted it. A successful data update is not evidence that an existing environment-consuming worker changed its settings. This article supplies neither a refresh interval nor an application hot-reload mechanism.
The positive chooses only environment consumption. Projected volumes and subPath appear as contrasts, not alternative implementations. A team switching delivery modes needs a fresh review of its actual consumer and lifecycle, rather than assuming a familiar ConfigMap name preserves the original behavior.
Use a new immutable data revision rather than editing the old one
The same guide’s immutable section says immutable status cannot be reverted and the data and binaryData contents cannot then be changed; deletion and recreation remain possible. That is not a guarantee that a name identifies trustworthy bytes forever.
Our stricter application policy uses separately reviewed, distinct revision references C-A and C-B. It does not delete and recreate C-A to smuggle new data under an old association. Source ownership, actual contents and preservation through use require independent evidence. Immutability supplies no permission decision or proof that a program interpreted every required value correctly.
Review the complete configuration at once. If the application requires parser profile P-A, output policy O-A and route profile R-A, its declared C-A inventory must account for all three. These are symbolic profile references, not real settings or a parser implementation. A convenient release label cannot repair a missing key, an unexpected override or values borrowed from another revision.
Bind the configuration reference inside the Pod template
The Deployment documentation ties rollout triggers to changes in the Deployment’s Pod template, .spec.template, rather than other changes such as scaling. The selected plan additionally requires independently supported effective unpaused controller state; a template edit is not evidence that a rollout started. It describes new ReplicaSets scaling up while older ones scale down at a controlled rate.
For proposed template T-A, trusted environment mappings select C-A’s declared keys. A reviewed T-B instead selects C-B. Changing that reference within the template is a template change; editing another object’s data without changing the template is a different operation. This worksheet constructs no template or command and adds no automatic reload controller.
During replacement, do not assume every running worker already belongs to T-B. Track actual worker/configuration association and job requirements separately from desired state. A new rollout does not transfer an old job’s approval, settle an external effect or establish that live work completed.
Keep a per-job association ledger
The useful conditional positive is an ordinary summary job J-A whose required configuration matches an independently established W-A/C-A association, whose declared unpaused deployment prerequisites hold, and whose current policy permits the work. Only then is its bounded preprocessing plan eligible. No job execution or summary result is reported.
PROPOSED_UNRUN — configuration association only
Worker/build: W-A / independently admitted executable
Configuration: C-A / complete reviewed non-confidential data
Consumption: environment only / explicit trusted key mapping
Pod template: T-A / reviewed reference to C-A
Actual worker association: separate evidence, not a name
Job: J-A / required configuration and current rights
Next release: C-B with separately reviewed T-BThe evidence record must distinguish intended reference, effective startup input, application interpretation and job admission. It should retain minimal profile and disposition references rather than environment dumps. This article supplies no cluster query, runtime introspection, job router or coordination implementation that establishes those associations.
Walk eight inert configuration cards
| Declared condition | Proposed disposition |
|---|---|
| Complete admitted C-A, matching T-A and actual worker/job association; all current prerequisites hold | Consider only the bounded C-A job plan; claim no observed execution. |
| A mutable ConfigMap’s data changed while a worker consumes environment values | Do not infer a changed running environment. |
| Projected volume replaces the selected environment path | Hold changed consumption; eventual projection is not verified application reload. |
| A subPath mount is introduced | Do not borrow ordinary projection-update expectations. |
| ConfigMap data changed, but the Deployment template did not | Do not claim that observation triggered a Deployment rollout. |
| T-B references C-B while older workers or live jobs remain unresolved | Keep associations separate; no atomic fleet activation or job-transfer claim. |
| C-A is deleted/recreated or keys come from mixed revisions | Hold identity or completeness drift; familiar names are insufficient. |
| Template and configuration match but actual worker state or current job rights are unknown | Withhold admission; object state and rollout status cannot replace those gates. |
Limit the rollout claim to supported evidence
The two maintained official bodies showed v1.37 navigation when read on 30 October 2026. No installed cluster release, controller behavior, startup environment or application parser was inspected. No candidate Kubernetes API, CLI, watch, environment, file, rollout, runtime fixture or model operation occurred.
Boundary is prelaunch and does not ship this configuration-delivery or worker-admission service. Its existing synthetic policy evidence concerns different behavior. Keep current request rights separate: an older configuration is not permitted merely because its data cannot change. No zero-downtime, all-worker activation, atomic file projection, secret protection or universal revocation guarantee is claimed.
Primary sources
Official references supporting the guidance. Examples in this guide are synthetic and are not reports of a production incident.