Who can deploy. Organisation administrators and site managers, with a recent second-factor code if they have enrolled one. Other members see the same card and the reason it is or is not ready, but never the button; the card says Only organisation administrators and site managers can deploy.
Deploy to production. An open pull request card shows Deploy to production only when everything is in order. Otherwise it shows, in plain words, the one thing in the way and what to do about it. The first problem found is the one shown:
- A required check never reported, for example Required check
phpunithas not reported. A check that never reported is not a pass. The workflow that provides it may have been renamed; fix the branch protection or the workflow, then the button appears. The card lists each required check as passed, still running, failed or has not reported. - A required check is still running, or has failed.
- The branch conflicts with the base branch, or is behind it where the host requires it to be current. Use Rebase with Sentinel or update it on the host.
- A reviewer requested changes (the reviewer is named), or an optional check failed on GitHub and the host reports the pull request as unstable (the check is named). If the host says unstable only because non-required checks are still running, the card shows Checks running, not Optional check failed. Once required checks are turned on for your organisation, a failed optional check is a warning, not a refusal (see Required checks).
- A required approval is missing (see Overriding a missing approval), or conversations on GitLab are still unresolved.
- The host reports the pull request as blocked, or has not worked out yet whether it can be merged. Wait a moment or look on the host.
- The pull request goes into a branch other than the production branch, for example a maintenance update into your branch for maintenance updates. Sentinel deploys only into the production branch, so the card says where it goes and asks you to merge it on the host and promote it yourself.
- The pull request targets neither the production branch nor the branch for maintenance updates. Merge it on the host yourself, or change the site's branches and run Patching again.
- The site's production environment is not mapped to a hosting platform (see The site Patching tab). You can instead choose "Sentinel merges, my pipeline deploys" or "I merge and deploy myself" under Who deploys updates.
- The site is set to "I merge and deploy myself". The card shows a sentence instead of the button: Sentinel does not merge or deploy this site's pull requests; merge the pull request on your Git host and deploy it the way you usually do.
- The Automated tests in CI item is not adequate. The card shows the same notice as that item.
- The deployment terms are not published yet.
- Another deployment for the site is queued or running. The card says "Another deployment is in progress for this site" with its step (Queued, Backing up, Merging or Deploying), and a link, See that deployment, opens that deployment's own page (on a job page it scrolls to the card below). Only one deployment runs per site at a time. While this pull request's own deployment runs, the card says "This pull request is being deployed" with the step and a See the deployment link. It keeps saying so even if a check turns pending part way through. Once the merge lands, the card closes and the job page's Deployment section follows the deployment to the end.
- The pull request is a draft, or is already merged or closed.
- A runtime pull request is waiting for the Drupal one (see Deploying a runtime pull request).
If the card was out of date when you pressed the button, Sentinel checks again, refuses and shows the current reason.
Deploying a runtime pull request. The Deploy to production button on a runtime card stays hidden until the Drupal pull request from the same run has been deployed by Sentinel and confirmed by the site. The card shows the reason instead:
- Waiting for #231: the Drupal pull request from the same run deploys first. Deploy the Drupal pull request first.
- Waiting for #231 to be verified: the site has not confirmed the Drupal update yet. The Drupal deployment is done, but the site has not confirmed it yet. If it fails or is not confirmed, the card asks you to check that it reached production.
- The Drupal pull request was closed without merging: run Patching again. The next run links the runtime pull request to a new Drupal one, or lets it deploy alone.
- Someone pushed to the runtime branch outside Sentinel: the listed changes may be out of date. Rebase it with Sentinel to rebuild it, or close it and run Patching again.
- The Drupal pull request was merged outside Sentinel: the runtime pull request is free to deploy. Confirm the Drupal update yourself.
- A Patching run for the site is still in progress: the button appears when the run finishes.
A runtime deployment names no packages, so the site confirms it when it next sends a full report. Only organisation administrators and site managers can deploy or rebase, as for any pull request, and Patching must be enabled for the organisation.
The deployment terms. Pressing Deploy to production opens a dialog with the deployment terms. Read the terms to the end, tick the box and type the site name exactly, then confirm. On a runtime pull request, the step where you type the site name also shows the consent line, for example: Deploying runtime update #232 to production after #231: PHP uselagoon/php-8.3-fpm:25.9.0 → uselagoon/php-8.3-fpm:25.10.1. A database release line change adds: A database upgrade cannot be downgraded. A backup is taken first. Sentinel takes no backup itself: any backup comes from your hosting platform, your pipeline or you. Your consent is recorded with the deployment. The card then says Deploying: queued. and then Deploying: step N of 4 with the step name, a status badge and a small four-step bar. See the deployment opens the deployment's own page, described under Deployments page below (on a job page, it scrolls to the card below).
What happens next. For a site that follows a hosting platform, the deployment runs in this order, and each step appears on its card in the Deployments panel:
- Backup on the hosting platform, where the platform has a backup service. Sentinel asks for the backup and then waits until the platform reports it as complete before anything is merged. While it waits, the status is Backing up and the card says "Waiting for the pre-deployment backup to finish." Sentinel checks every 10 seconds. It waits up to 10 minutes on Platform.sh and up to 30 minutes on Quant Cloud. The deployment stops here, and nothing is merged, if the backup fails, if it is still running at the limit ("did not finish within" that many minutes), or if it only completes after the limit. It also stops if the environment is on a different branch than the site's production branch, or the platform is already running a deployment that Sentinel did not start. It stops, with nothing merged, if during the wait someone starts a deployment on the platform, the branch or the environment mapping changes, the site moves to another organisation, or Patching or its subscription ends for the organisation. For the last two cases (the organisation no longer has Patching, or the site moved) no pull request comment or email is sent. The Activity page records
backup.startedwhen the backup is requested, andbackup.takenonly once it is complete, with its size where the platform reports one. - Merge, pinned to the commit that was reviewed. If anyone pushed to the branch after the check, the merge is refused: review the new commit and deploy again. Sentinel squashes the commits unless a merge commit was asked for.
- Deployment on the platform. Sentinel follows the platform's progress. If it reports no final result within 90 minutes, the deployment is marked failed; Sentinel does not cancel anything on the platform.
- Deployed, awaiting site confirmation. The platform finished. The deployment is Confirmed when the site's connector next reports that every package named in the pull request is at or above the expected version. A connector report that is still below the expected version keeps waiting and says why. If 24 hours pass without confirmation the status becomes Deployed, not confirmed: check the site and its connector.
While the card says Deployed, awaiting site confirmation, a box reads "Sentinel waits for the site until about" a time 24 hours after the deployment, then says this deployment becomes Deployed, not confirmed and organisation administrators get an email. To report sooner, run drush sentinel_connector:deploy --force on the deployed Drupal environment, not in CI: --force sends a report even when the site thinks nothing changed. Open this site's Connector log shows whether a report arrived; a report the site skipped is not in the log. For Deployed, not confirmed the box gives no deadline. It says Sentinel stopped waiting, so a new report cannot confirm the deployment, though it still updates the versions Sentinel holds for the site; check the deployment reached the site, then send a report with the same command. The same box shows for deployments through a pipeline and through a hosting platform. See How the site reports.
The other statuses are Queued, Backing up, Merging, Deploying and Failed. If your organisation uses Uptime monitoring, its result may also be shown, but it never fails a deployment.
Overriding a missing approval. When the only thing missing is an approval, the card may offer Override and deploy. This depends on the Git host:
- GitHub with the Sentinel App allowed to bypass the rule (a bypass actor in the ruleset or classic branch protection): available to administrators.
- GitHub classic branch protection that is not enforced for administrators: available if you connected your own GitHub account under GitHub identity for override merges (see Connections) and you administer the repository. The card links to the Patching connections in Settings if you have not.
- GitHub rulesets without the App as bypass actor, or classic protection enforced for administrators: not available. Ask a reviewer to approve, or change the rule.
- GitLab: never. Ask for the approval on GitLab.
An override first shows the override terms, then the deployment terms, and records both. After the merge Sentinel comments on the pull request naming the administrator who instructed it and the terms version they accepted.
Deployments panel. Once a repository is chosen, the Deployments section of the site's Patching tab has a Deployments panel, newest first, which refreshes every 30 seconds while one is running. Each card shows the status, who requested it (or Sentinel), a four-step tracker (Backup, Merge, Platform deploy, Site confirms), a link to open it on the platform, and the Expected versions with what the site reported. Under each step the tracker says its state in words (done, in progress, failed, not confirmed or not started) and its time when one was recorded (under Backup, the time the deployment started). A pipeline deploy, or a platform without backups, says "no backup was taken" under Backup. On a phone the four steps stack. While a deployment runs, the pull request's checks and reviews fold into one line, "The checks and reviews are folded while this pull request deploys.", with a Show checks and reviews button. Each card has an Open deployment link to its own page, and so does each deployment on the job page. The panel shows the latest 20. Under the list, Show older deployments loads earlier ones, a page at a time, and disappears when nothing older is left. A saved link to a deployment card that is older than the ones shown still opens it: the card appears at the top under The deployment you opened. With no deployments it says No deployments yet. Anyone who can see the site's Patching tab can read this list.
Deployments page. Where Patching is on for your organisation, Deployments at /app/deployments lists every Patching deployment of the organisation, running ones first. The chips Running, Waiting for the site, Needs a look (failed or not confirmed), Done and All each show a count, and a Site filter narrows the list. It shows 25 at a time. The columns are Site (with the pull request number; the site name opens the deployment), Progress (a four-step bar of Backup, Merge, Platform deploy and Site confirms, with a status badge), Job (opens the job), Started and By (the person, or Autopilot). It refreshes every 30 seconds while a deployment runs or waits for the site. There is no sidebar item for it yet; open it from a deployment link or a message.
One deployment. /app/deployments/(reference) is titled "Deployment on (site)", with a status badge and the four steps named. Its section of links is headed Where to look: Site's deployments (the site's Patching tab, Deployments view), the job, the pull request and, when there is one, the platform log. The four-step tracker sits in the deployment card below (packages, verification and rollback guidance) and a Timeline of what was recorded. A failed deployment says "Organisation administrators are told when a deployment fails, in the dashboard and by email unless they turned that email off." A reference from another organisation, or one that no longer exists, shows "This deployment is not in your organisation, or it no longer exists." A deployment of another of your organisations asks you to choose that organisation in the top bar first. It refreshes every 30 seconds while the deployment runs or waits. Members with Patching read access (organisation administrators, site managers and viewers, the same people who see Jobs) can open both pages. When Patching is off for the organisation they show the dashboard's not-found message. Deploying still needs deploy permission and still happens on the pull request, not on these pages.
When a deployment fails. The card names the step that failed (Backup, merge, Platform deployment or Site confirmation) and shows the platform's or host's message with secrets removed. It also gives guidance for rolling back or redeploying on your platform: on Platform.sh restore the snapshot or redeploy the previous commit; on Quant Cloud redeploy the previous image tag; on Lagoon revert and push, or restore the backup. A link to the support plan page is shown once, as help. Sentinel also comments on the pull request, and the failure is logged on the Activity page and the release timeline.
Notifications. Deployment of pull request #N for (site) failed is sent by email to every organisation administrator. ...not confirmed is also emailed, after the 24 hours. ...confirmed is an optional bell notification that you can control with your notification rules. The in-app messages and emails for all three end with Open the deployment, which opens that deployment's page.
Required checks
Required checks are being rolled out. Until they are turned on for your organisation, any failed check stops Autopilot, including an optional one, and the published Autopilot terms stand. The rest of this section describes how Patching works once required checks are turned on for your organisation.
A required check is a check your repository says must pass before a pull request can merge into the production branch. You set it on your Git host: on GitHub with branch protection (Require status checks to pass before merging) or a ruleset (Require status checks to pass), on GitLab with Pipelines must succeed under Settings, Merge requests, Merge checks. Sentinel reads it from there and never decides for itself that a check matters. If the rule names the app that must post the check, a passing check of the same name from another app does not count. Security, maintenance and runtime updates follow one rule.
- The repository has required checks. Autopilot merges when every required check passed. A failed optional check does not stop it and is reported.
- The repository has no required check. Autopilot merges even when CI is red, once CI has finished or the 30-minute wait has passed. Failures are reported.
- The rules could not be read, or were read only in part. Autopilot never merges. It tries again, then pauses, and no acknowledgement can be given. This never counts as "no required check".
The acknowledgement. Autopilot needs an acknowledgement for each website, as well as the Autopilot terms. An organisation administrator or site manager gives it, with a recent second-factor code if they have enrolled one, on the wizard's Mode step or when you select Switch to Autopilot in the Mode card under Configuration. The Required checks panel names the checks and, for each, the app that must report it (Any app when the rule accepts any), or says plainly that there are none and that Autopilot will merge and deploy even if CI/CD checks fail. It also shows the customer (the legal name on your Data Processing Agreement, or your organisation's name with a note while none is recorded), the repository and branch, the full wording, and how long Autopilot waits for checks that have not finished ("Autopilot waits up to N minutes"). Tick the box and select Confirm for this website. Other members can see the required checks but cannot acknowledge. A profile never gives it. Each acknowledgement is recorded in the Patching activity, kept for 13 months. If the repository changed while the panel was open, nothing is recorded and you are asked to review it again.
History. Organisation administrators and site managers can select History in the Required checks panel to see every acknowledgement, invalidation and added requirement for the website, newest first, each with the deployments that relied on it. Download CSV saves a copy. The history cannot be changed or deleted and is kept as long as your Patching consent records. Members do not see it.
When the repository changes. Sentinel reads the rules while an Autopilot run is active and again just before the merge. A required check that is added is followed, with no new acknowledgement. If one is removed, its expected app changes, or the repository moves to having no required check, the Autopilot card says The required checks need a new review. An organisation administrator selects Review required checks, confirms, then selects Resume. Changing the website's Git host connection, repository or production branch also ends the acknowledgement, and going back does not bring it back. A change made and undone between two readings cannot be seen.
Merge queues are not supported. If the production branch requires a GitHub merge queue, or the GitLab project uses merge trains, the acknowledgement cannot be given and Autopilot pauses. Merge the pull request on your Git host or remove the rule.
Copilot. No acknowledgement is needed. A failed optional check is a warning: the card names it and the deployment records it. If the failing optional checks change between opening the card and pressing Deploy, nothing is deployed: the card shows the new list and asks you to press Deploy again. The deployment shows Optional checks failing when deployed and Optional checks found failing just before the merge separately. A failed required check refuses the deployment, and an override never waives one.
Websites never acknowledged keep the strict rule, so any failed check pauses them. The Autopilot card has a Required checks row with Review. Accepting new Autopilot terms does not relax it.
Automated tests. A current acknowledgement replaces the rule in Automated tests in CI for that website. You can give it on Copilot, with Review required checks in the Mode card under Configuration, so that a website with no tests can run an update and its first deployment. Tests that fail in Sentinel's own update job, and pull requests holding changes that need a person's review, still stop Autopilot. With an acknowledgement, Sentinel's CI fixing works only on required checks.
