Staging

This section examines a standard deployment with a DEV (Development) and a PROD (Production) instance. If necessary, an additional TEST (Testing) instance may also be useful.

Here are the recommendations that have proven effective in practice. Guiding principle: PROD is an exact replica of an approved repository state. Nothing is developed on PROD, and nothing is modified locally.

1. Version status of the flows

Every Process Group on PROD is versioned and marked as “Up to date.” This is the green checkmark. The following statuses should not persist on PROD:

  • Locally modified (asterisk): This is always a process error. Hotfixes run on DEV, are committed, and then deployed to PROD. If a local change was made anyway, the change is propagated to DEV and resolved on PROD via “Revert local changes” or a version change.

  • Sync failure: This is an incident and must be resolved immediately.

  • Stale: This is only permitted if a deployment is intentionally pending. PROD does not automatically follow “latest.” It runs on an explicitly approved version.

  • Unversioned Process Groups: They do not belong on PROD.

Separate DEV and PROD in the repository. Both must not point to the same state uncontrollably. In DEV, completed states are explicitly marked as releases via the commit message. Only these are deployed to PROD. It is also conceivable to restrict PROD to read-only access.

Top-level Process Groups with clear business boundaries are versioned. Nested versioned Process Groups should only be used deliberately (e.g., for reuse), as they make deployment more complicated.

2. Environment-Dependent Configuration

Anything that differs between DEV and PROD belongs in Parameter Contexts. This applies to hosts, paths, credentials, and topics. There are no hard-coded values in the flow itself. The parameter Context has the same name in both environments but contains environment-specific values.

3. Processor Status

All processors belonging to the production flow are running (RUNNING). Stopped processors on PROD are a warning sign because queues fill up beforehand and backpressure blocks the upstream.

Processors that are not supposed to run do not belong in the PROD state. They are removed on DEV. If a branch is intentionally kept inactive—for example, as a feature under development—it is set to DISABLED rather than just STOPPED. The DISABLED status is part of the release, whereas STOPPED or RUNNING are not. This ensures that “intentionally disabled” is documented in the repository, and no discrepancy appears on PROD.

Debug and test components are removed before a version is released to PROD. These include LogAttributes for debugging, GenerateFlowFile as a test source, and dead-end Funnels.

On PROD, there are no invalid components (yellow warning triangle) and no persistent Bulletins.

4. Controller Services

All Controller Services used by the flow are ENABLED. The “Enabled” status is not versioned and must therefore be part of the deployment step after an import.

Controller services are defined within the versioned Process Group. They are then included with the flow. References to services outside this group—that is, at the root or controller level—can easily lead to missing or incorrectly assigned Services during import. Where global Services are required, such as a shared SSLContextService, they must have identical names on DEV and PROD and be clearly documented.

Unused Controller Services are removed rather than simply left deactivated.

5. Validation

  • Automated checks after every deployment:

  • All versioned Process Groups are marked as “Up to date” or, if applicable, “Stale.”

  • The Controller Services are enabled.

  • There are no invalid components.

  • All processors are running, with the exception of those that are versioned and disabled.

  • Regular drift check: The same check also runs periodically via the REST API, for example as a monitoring job. This ensures that local changes are detected before they cause problems during the next deployment.

Summary: On PROD, every flow is versioned and set to an explicitly approved “Up to date” status, with no local changes. All processors are running, unless they have been intentionally and versioned as disabled. All Controller Services are enabled and reside in the Process Group. Environment values come exclusively from Parameter Contexts. Every change follows the path DEV → Repository → PROD.