
It should identify the versions and dependencies, affected workflows and owners, release-note changes, backup and restore evidence, test scenarios, change window, stop conditions and post-upgrade acceptance checks.
A Helm rollback targets a release revision. Whether and how application data or schema can be restored depends on the application and database. Confirm the vendor’s recovery guidance and test the complete procedure for the deployment.
Compare representative findings before and after the change. Test an import and, where relevant, a reimport; follow a finding through assignment, ticket synchronization, remediation evidence and reporting. Check exceptions and access permissions separately.
Pinning can give the team time to prepare. The decision should consider security fixes, vendor support and the organisation’s exposure, with a named owner and a date to reassess the pin.
No. A successful deployment establishes only the checks represented by that deployment process. Data, permissions, integrations and finding behaviour still need to be verified against the baseline and the applicable release notes.
A scanner sends a new report on Monday morning. The security team expects to find its results in the platform, assign the critical items and follow the existing tickets through to closure. After a major upgrade, the application login works, but the import has stalled and the team does not yet know which findings reached the ticketing system.
That scenario is illustrative, but the operating problem is real: a security platform carries work between scanners, engineers, risk owners and reviewers. An upgrade plan needs to account for those handoffs. The practical goal is to prepare a recoverable change, test the paths that carry remediation work and make any interruption visible to the people who depend on them.
The release notes tell you what the software changes. Your own records tell you which teams, integrations and decisions depend on the current behaviour. Both belong in the upgrade plan.
Start with one recent finding and follow it from intake to closure. Identify the scanner or API that submitted it, the product or service it belongs to, the person or team responsible for triage, the ticket created for remediation and the evidence used to close it. Follow a second finding through an exception or risk-acceptance path. The examples expose different dependencies: one may rely on a webhook; another may rely on an approval and an attachment.
Record the systems and owners beside those paths. Include scheduled imports, credentials and service accounts, authentication, ticket synchronization, dashboards, notifications, exports, retention and any service-level commitments the team tracks. Mark the points where data can wait safely and the points where delays need a human decision.
A baseline should be small enough to use during a change. Capture a dated sample of representative finding IDs, status, severity, ownership, timestamps and linked tickets; add the most recent successful import and the relevant dashboard or report. Protect this evidence because it can contain vulnerability details and customer information.
“A healthy login confirms that people can reach the platform. It does not confirm that a finding can travel from a scanner to the engineer responsible for it.”
The same distinction appears in Das Meta’s Security Hub migration guide: the team checks where findings go and whether ticketing and reporting still receive them. A security application upgrade has its own technical path, but the downstream ownership question is familiar.
Write down the running application version, chart or package version, database version, enabled features and custom settings. Then identify the target version and read the vendor’s notes for the releases in between. A chart version and an application version are different identifiers; record both rather than assuming their numbers move together.
Product changes can alter a remediation workflow even when the service starts normally. DefectDojo’s open-source 3.0 upgrade notes describe changes to authentication and authorization, removed API endpoints and changes to how a Dependency Check parser groups findings. For teams using the affected features, that makes login, integrations and reimport behaviour specific items to test. These are product-specific examples, not a prediction about every security platform.
Use the inventory to turn each relevant release note into a decision: Does this feature exist in our installation? Who uses it? What should a successful test show? If the answer depends on edition, custom settings or an undocumented integration, record the uncertainty and resolve it before production.
A version pin gives a team time to prepare, provided the pinned release remains suitable for its security and support needs. Record the pin, its owner and a review date. Check vendor security notices and support information before choosing how long to stay on it; a pin can defer an upgrade and also defer fixes.
In infrastructure-as-code environments, inspect the selected module version and explicit overrides before reviewing the planned change. Terraform’s plan command previews proposed operations, and a saved plan can be used for a later apply. The saved file can contain sensitive values, so store it with the same care as other change evidence.
A plan is one part of the approval package. Add the tested starting state, target configuration, dependencies, maintenance window, named decision-makers and a recovery plan. If the platform is installed with Helm, inspect the chart values and release history as well. The Helm upgrade reference documents the available flags and value-handling options; their effect must be checked against the version and provider in use.
Ask what must come back for the team to continue its work. The answer may include a database, uploaded evidence, configuration, encryption material and integration credentials. The exact components depend on the installation. A database snapshot alone may leave a restored application unable to read attachments or reconnect to the systems that send findings.
Take a consistent backup using the method appropriate to the database and storage, then restore it into an isolated environment. Verify the version and configuration used for the restore. Check that representative findings, ownership, history, attachments and access controls are present, and that the restored system cannot send real notifications or modify production tickets.
DefectDojo’s self-hosted Pro restore guidance distinguishes database recovery from restoring files and external storage. It is a useful example of why those components need separate checks. The guide is for a particular product and edition, so the platform owner must confirm the correct procedure for the deployment being reviewed.
Measure the restore in rehearsal and compare it with the agreed recovery objectives. A successful test on a copy provides evidence for a specific backup and procedure; it does not prove that every future restore will succeed. Keep the backup location, access, timestamps, test result and recovery owner in the change record.
Rehearse with a representative copy of the data and configuration. Include the jobs, connectors and access paths that matter in production, using test destinations or disabled outbound actions. Run the upgrade, record migration and startup behaviour, and compare the resulting workflows with the baseline. A small test database may miss the timing or data-shape issues that appear in a larger environment.
Before the production window, agree who can pause the change and what will trigger that decision. Examples include a failed migration, unavailable backup, a critical connector that cannot authenticate, missing ownership on sampled findings or a queue that cannot drain. Decide how intake will be handled during a pause: hold reports in a known location, assign a person to monitor them and agree how they will be replayed safely.
Recovery needs its own decision point. Helm can return a release to an earlier revision, but application data and database schema may have changed separately. The DefectDojo Pro upgrade guidance explicitly tells operators to address the database when a migration has run. Confirm the recovery sequence for your platform and edition before relying on a rollback command.
When the platform is available, repeat the baseline checks with the people who own each path. Import a representative report in a controlled way, confirm the expected findings appear and check whether identifiers, severity, status and ownership match the intended behaviour. Reimport where that is part of the workflow. Investigate unexpected duplicates or closures before allowing scheduled imports to resume at full volume.
Follow a finding into the ticketing system and back. Confirm the assigned team, link, due date or SLA field where applicable, and the action that records remediation. Check an exception or accepted-risk record, an attachment and a report used for review. A dashboard with missing labels or broken alert routes can also obscure the work: Das Meta’s Grafana dashboard upgrade guide shows how to validate the signals and context that users rely on.
Finally, check access for an administrator, a triage user and a read-only reviewer, using approved test accounts. Confirm API credentials and notification destinations, inspect failed jobs and queues, and record any unresolved discrepancies with an owner and a deadline. Resume each integration deliberately, then watch the first scheduled runs.
Identify the source and target versions, edition, dependencies and relevant release notes.
Record representative findings, owners, tickets, exceptions and integration paths before the change.
Restore a protected backup into an isolated environment and verify data, files and access.
Rehearse the change, define a maintenance window, stop conditions and a recovery decision owner.
Compare intake, remediation, evidence, reporting and access against the baseline before closing the change.
DefectDojo’s 3.0 open-source upgrade notes identify changes to SSO, authorization, API endpoints and Dependency Check finding grouping. Check the features your deployment uses.
Terraform can save a plan for a later apply; that file can contain sensitive configuration and values. Protect it as a change artifact.
Helm release rollback and data recovery are separate concerns. Confirm the platform’s database migration and restore procedure before the production change.
The useful acceptance evidence is a tested path from import to owner, ticket, remediation record and report, using a representative baseline.