Azure App Service is a service used to create and deploy scalable, mission-critical web apps.
A successful OneDeploy operation confirms that the package was processed and stored, but runtime activation should be verified by starting the app and executing a health check or representative trigger while reviewing Application Insights telemetry.
What OneDeploy Succeeded establishes
For Flex Consumption code deployments, OneDeploy processes the submitted package and stores the ready-to-run ZIP in the configured deployment container. The function app subsequently runs directly from that package. Uploading a ZIP directly to the container is not equivalent to deploying it.^1^
Therefore, Succeeded plus Pipeline completed for FDM deployment establishes successful package ingestion and publication to deployment storage. It does not, by itself, demonstrate that:
- an instance started successfully;
- the Functions host loaded the package;
- Python imports and dependencies loaded;
- triggers were indexed;
- an invocation completed successfully.
Those require runtime validation.
Identifying the effective package
The supported operator-facing surface is:
- Function App → Diagnose and solve problems.
- Search for Flex Consumption Deployment.
- Review deployment history, package status, and the associated deployment details.^2^
- Correlate the deployment ID with Application Insights deployment traces:
traces
| extend deploymentId = tostring(customDimensions.deploymentId)
| where deploymentId != ""
| project timestamp, deploymentId, message,
severityLevel, customDimensions, appName
| order by timestamp desc
This is Microsoft’s documented query for Flex Consumption code-deployment logs.^3^
The published documentation does not define a supported runtime API or app setting that returns an immutable package hash or exact blob name currently mounted by a particular instance. For deterministic release identification, include a commit SHA or release ID in the package and expose it through a health/version endpoint or telemetry. That application-level identifier should be treated as the runtime proof of which build executed.
Meaning of the diagnostic result while stopped
Microsoft documents that a site update occurs for code or configuration changes and that rollout behavior operates on running instances. It also states that there is no built-in completion signal for site updates; instance logs can only provide a best-effort estimate.^4^
The documentation does not define Function app updates summary: Failed or Site update / config change: Not detected as authoritative evidence that a successfully processed package was rejected. It also does not explicitly guarantee that these exact detector results are expected when the app is stopped or has zero instances. In that state, there is no runtime activation to observe, so do not use this detector result alone as either success or failure.
Start the app and perform runtime validation before interpreting the deployment as activated.
Which artifacts and permissions must remain
Two artifacts have different lifetimes:
- Original package source referenced by
packageUri: after OneDeploy has successfully downloaded, processed, and copied it, the runtime executes the processed package from the deployment container—not directly from the original source URL.^5^ - Processed package in the configured deployment container: this is runtime content and must remain available. Flex Consumption stores and runs the current package there.^1^
Consequently, temporary source URLs, SAS tokens, or staging blobs can be retired after confirmed deployment provided they are not the configured deployment container or credentials used by the app. Do not delete the processed package, deployment container, storage account, or remove the function app identity’s required access. The app retrieves the package on startup, including after scaling from zero.^2^
Recommended activation and rollback procedure
- Deploy and require OneDeploy to finish successfully.
- Query deployment telemetry by
deploymentId; fail the pipeline on validation, storage, or build errors. - Start the app if the workflow stopped it.
- Invoke a dedicated health/version HTTP endpoint, or submit a controlled event to a representative non-HTTP trigger.
- Verify:
- the expected release ID is returned or logged;
- the invocation succeeds;
- Application Insights has no startup, import, indexing, or dependency exceptions.
- Only then mark the release active and remove temporary source access.
- Retain the commit SHA, dependency lock file, configuration, and successful pipeline run.
Flex Consumption has no deployment slots or built-in package revision history. Each code-package deployment overwrites the current package. Rollback is another deployment of the previous state—not a platform-side package selection operation.^6^
For rollback:
- Re-run the previous successful GitHub Actions or Azure Pipelines run.
- Ensure it checks out the original commit SHA and restores both code and versioned settings.
- Repeat the same health/version invocation and telemetry checks.
- For higher assurance, validate releases in a separate Flex Consumption app because deployment slots are unavailable.^6^
References
- Deployment technologies in Azure Functions (flex-consumption-plan)
- Azure Functions Flex Consumption plan hosting
- Analyze Azure Functions telemetry in Application Insights
- Site update strategies in Flex Consumption
- Package-based deployment for Azure Functions | Microsoft Learn
- Recover from a bad Flex Consumption plan app deployment