Skip to content

Patching

51 help pages

Patching is a separate product for applying security and maintenance updates to your sites through your own repository. It is in testing and open to every account. If Patching is turned off for your organisation, or off on this deployment, nothing about it is visible: there is no Patching group in the sidebar, no Patching connections entry under Settings, no Patching tab on a site and no entry in the product switcher, and opening a Patching address shows This feature isn't available for your organisation yet with a link to open a support ticket (or a note to contact support where the support portal is off). If it is turned on again it returns on your next page load. If it is turned off, its pages disappear at once, any scheduled work is paused and your connections and site choices are kept. Having a plan that includes Patching does not by itself switch it on, and Agency coverage does not carry it from one organisation to another. Patching is in testing and open to every account, so there is no invitation, application or approval. If you see that message, an administrator turned Patching off for your organisation or it is off on this deployment; use the ticket link to ask why.

Patching has two modes for a site:

  • Copilot: Sentinel prepares a pull request on your repository. A person reviews it and presses Deploy, which opens the deployment terms dialog. Accept the terms there for that deployment. If you choose "I merge and deploy myself" for the site (see Who deploys updates), there is no Deploy button: you merge and deploy the pull request yourself.
  • Autopilot: Sentinel prepares the update, waits for your CI, merges, deploys and verifies it, with no person in the loop. A site set to "I merge and deploy myself" never gets an automatic merge.

Both modes run on the same two schedules, one for security updates and one for maintenance updates (see Schedules).

What Autopilot does after the pull request. This part is off until it is rolled out to your organisation, and Autopilot cannot be turned on in production until its terms are published. When it is on, the website's Patching tab shows an Autopilot card with the current run and what it waits for: the checks, an administrator's approval (only if the website asks for one), the quiet window, or another deployment. Turning Autopilot on for a website also accepts the Deployment terms that the Autopilot terms name; the acceptance step has a collapsed Read the Deployment terms you also accept section. Autopilot applies to the updates you chose for the website: security updates, maintenance updates in the scope you set, and runtime components you switched on for Autopilot. When the Deployment terms change, the Autopilot terms change too, and Autopilot pauses on each website at its next run until an organisation administrator or site manager accepts the new terms again. Autopilot merges only the exact commit it checked, then deploys and waits for the site's connector to confirm the new versions. If the Autopilot terms change, the Autopilot card shows Accept the new terms to organisation administrators and site managers, and Resume stays hidden until someone accepts.

  • It never merges when a check or test fails (tests that already fail without the update included), when no check reports within 30 minutes, when the connector has not reported for 48 hours, or while another deployment runs. Once required checks are turned on for your organisation, a site with a current acknowledgement follows the repository's required checks instead (see Required checks). On a website whose deploy waits for its tests, an organisation administrator can let it merge before the tests run, once that is turned on for your organisation (see Autopilot when your deploy waits for tests).
  • If something goes wrong, it pauses, comments on the pull request and tells the organisation's administrators. The card says Autopilot paused: it needs a person with the reason. Running Autopilot runs for the site wait until an administrator or site manager selects Resume, with a recent second-factor code if they have set one up; nothing merges or deploys meanwhile. Scheduled runs are skipped during the pause, and Resume does not run them again. To catch up a skipped security update, merge or close any open security pull request, then use Run update. If the deployment failed after the merge, Resume ends that run and the merged change waits for you to deploy it. If an approval link expired, Resume sends a new one once the checks pass again. Merging or closing the pull request yourself also ends the run.
  • If the site asks for approval, administrators get an email with a link. Sign in and select Approve the merge. Administrators and site managers can approve, with a recent second-factor code if they have set one up. The link lasts 14 days and covers only the commit it names: a new push asks again.
  • Before you can switch a site to Autopilot, deploy one update on it yourself through Sentinel. Enrol it on Copilot first.
  • The Patching overview's On Autopilot card shows how many sites are paused.

A site has no Patching settings or schedule until an organisation administrator or site manager enrols it (see Enrolling a site).

Set up starts where you connect accounts (Git hosts and hosting platforms), with a checklist that shows when a site is ready. Pull requests that Sentinel opens are shown on the site's Patching tab (see Pull requests below). Running an update opens a pull request on your repository (see Run an update below). With Copilot, a person reviews it and presses Deploy to production (see Deploying a pull request below). Autopilot deploys sites whose required checks pass, under the consent you set, once merging and deploying without a person is rolled out to your organisation (see What Autopilot does after the pull request above); until then Autopilot runs stop at the pull request, as in Copilot. A site set to I merge and deploy myself always stops at the pull request, in both modes (see Who deploys updates below). The product tour has one extra stop, Patching, that points at the Overview link.

Who can do what

  • Organisation administrators manage everything in Patching. Only they can manage AI keys, Git hosts, hosting connections and a website's hosting mapping, profiles, Included AI and the Patching activity log.
  • Site managers run Patching for their websites. They can enrol and unenrol a website, change its Patching settings, run updates, deploy (including Deploy anyway), accept the Autopilot terms, switch to Autopilot, resume it, approve a merge from an approval link and acknowledge required checks. They cannot manage the items above, and they cannot let Autopilot merge before the tests run (see Autopilot when your deploy waits for tests).
  • A recent second-factor code is asked for when you enrol or unenrol a website, switch to Autopilot, resume Autopilot, use an approval link, turn on Included AI, remove a Patching connection or change a website's hosting mapping, and when you deploy. This applies only if you have set up two-factor sign-in.
  • Billing access is separate. On first-site-free Pay as you go and Agency accounts, the billing account owner must also administer the billing organisation to manage the Patching subscription. That owner sees Subscription and billing and Plans. Another organisation administrator does not, once billing access has been checked. Retained legacy subscriptions use organisation administrator billing access.
  • Members and viewers can read Patching pages, the Patching plan card and each site's readiness. Viewers cannot change Patching. Members cannot change site setup. Site managers and members cannot change billing.

Deployments held for configuration drift

Where configuration drift checks are switched on for your deployment, Patching will not merge or deploy while the live site's configuration differs from the code in your repository, because a deployment could overwrite the live changes. Sentinel still opens the pull request. Take a backup first, or make sure your deployment process takes one.

  • What you see. While a hold is in effect, the site's Patching tab shows a pause notice at the top and on each open pull request's deploy card: "Deploys are paused: production has configuration drift (N objects)", or the wording for unknown, with Sentinel's message and what to do. It links to the Drupal configuration section of the site's History tab and says "Deploys resume on their own when the next complete report shows no drift." Nothing appears while drift checks are off. Updates still run and open pull requests; only merge and deploy wait.
  • Copilot. The deploy check refuses with one of two reasons. Configuration drift means the production configuration differs from code. Configuration drift unknown means Sentinel has no complete report from the last 48 hours (or none yet), the Connector is older than 1.3.0, or the latest report was incomplete or could not compare the configuration. The check is made after Sentinel has confirmed that no other deployment is running, and before it asks for an approval.
  • Deploy anyway. Organisation administrators and site managers see a Deploy anyway button on the pull request. Members and viewers see the reason but no button. The button opens a dialog showing the drift state and the number of objects, a warning that changes on production may be replaced, and "Take a backup first, or make sure your deployment process takes one." Enter a reason of up to 500 characters and type the site name, then accept the usual deployment terms. A recent second-factor code is needed, as for any deployment. It covers this one deployment only; the next deployment meets the hold again. If the drift changed while the dialog was open, the dialog comes back with the new state and empty fields, and nothing is deployed. If the hold no longer applies, the dialog closes with a message.
  • Record. The deployment and the Patching activity show "Deployed with configuration drift overridden by NAME on DATE (N objects differed from code, reason: ...)", or "drift state unknown". It is also in the audit log.
  • Autopilot. Automatic Autopilot merging is not available yet. When it is, Autopilot opens the pull request and then waits before merging. The tab says "Autopilot opened the pull request and is waiting before merge." and when it stops waiting. It continues by itself after the first clean report. If the hold lasts longer than the hold period (7 days unless your administrator has changed it), the run ends as held expired and the pull request stays open for a person to deploy with Copilot. An administrator can also use Deploy anyway on a held run.
  • Other holds. A check failing on the production branch holds deploys in the same way, with its own notice and no override. See Checks failing on the production branch.
  • Readiness. The Readiness card lists Deploy holds separately from the setup steps. A hold does not change a site's readiness, and you can still enrol it or run an update. The Patching sites list shows that a deploy hold is in effect, and the Patching overview counts the sites held by drift.

Checks failing on the production branch

Before each security or maintenance update, Sentinel reads the checks and statuses your Git host shows for the latest commit on your production branch. This is the pipeline you see on that branch, for example the one that deploys it, not Sentinel's own test run. It applies to GitHub projects. A GitLab project gets no reading and no hold. A failing check, required or not, never stops the update: Sentinel still opens the pull request. It holds the deployment instead, because deploying an update onto a branch that is already failing is not safe.

  • What is held. Copilot's Deploy and Autopilot's merge and deploy. Updates keep running and keep opening pull requests. This hold cannot be overridden: there is no Deploy anyway button for it. Fix the failing check on your production branch.
  • Where you see it.
    • Job page: a warning above the findings, "A check is failing on the production branch", with the check's name and what to do. Under it, up to three links, each "Open (check name) on GitHub", open the failing check's page in a new tab. A link is shown only for a page on github.com, so a check run by another service (an external CI, for example) has no link, and a job that ran before this was added has none. The same row, with the same links, is in the list under What this job did, and the job ends as "Done, 1 item needs you".
    • Site Patching tab: a pause notice at the top, "Deploys are paused: a check is failing on the production branch", and a row in the Deploy holds list of the Readiness card. A hold does not change a site's readiness, and you can still enrol it or run an update.
    • Pull request: the deploy card shows the same pause notice and a message such as "The deploy check failed on main. Fix it before deploying this update." instead of the Deploy button. The top of the pull request description on your Git host starts with Check failing on the production branch.
    • Email: organisation administrators get "A check is failing on the production branch of" the site's name, with an Open Patching link. It belongs to the Patching needs attention group on the Email settings page, where the service has turned that page on.
  • How it ends. Sentinel reads the branch again about every five minutes while a hold is in effect, when someone selects Deploy, and just before Autopilot merges. When a reading shows the checks passing, the hold lifts by itself and the notice goes. Autopilot waits, then carries on without anyone selecting anything. The notice says "Deploys resume on their own when the production branch's checks pass. Sentinel reads them every few minutes and before every deployment." If a reading cannot be made, the last result stays until a reading shows the branch passing.
  • Who. Anyone who can see the site's Patching tab or the job page sees the notice. Organisation administrators and site managers can deploy once the hold is gone; members and viewers can read but never deploy. The email goes to organisation administrators.

Patching plan

Patching needs active access, through a plan bought under Billing or access arranged with support; see Patching in the Billing section. A Patching plan card in the Patching section of Home and on the Billing overview shows, for every member:

  • the plan name;
  • running jobs of the limit;
  • this month's estimated AI spend, in US dollars, charged by your own AI keys (each key has its own cap; see AI keys);
  • sites enrolled of the sites covered. The caption reads "enabled of those paid for", or "enabled of those free on your plan" when only the Pay as you go free site is covered.

If Patching is not on the subscription, the card in the Patching section of Home and on the Billing overview shows Add Patching under Billing. On Billing → Patching the card shows Add Patching below instead, which moves you to the Plan section. If no plan is offered for your subscription's currency and billing period, the page says No Patching plan is currently available for this organisation and the card says Patching cannot be added yet; use Contact support to ask when it will be. Everyone in the organisation sees the card, but adding the plan needs billing rights. Two things are refused on a site's Patching tab:

  • Enrolling a site is refused when the organisation has no active Patching access (the message says to add it under Billing), and when the sites covered are used up (approve more sites under Billing first).
  • Choosing Autopilot needs active Patching access (the plan or access arranged with support) and a passing CI tests gate, then the Autopilot terms. The wizard disables Autopilot with one of two plan reasons: "Your organisation does not have active Patching access. Contact your billing account owner or support." or "The Patching plan could not be verified. Try again before selecting Autopilot." A failed or missing plan response does not enable Autopilot. Retry after the plan can be loaded; contact the billing account owner or support if access is missing. There is no separate Autopilot plan, and a Sentinel trial does not switch it off.

Review Patching pricing appears in the wizard's Mode step only when Autopilot is blocked by the plan, billing management is allowed and a current, changeable subscription has a Patching offer. It opens Billing → Patching. Adding the plan does not enrol the site or accept deployment terms. Return to the wizard, check the site's readiness and CI tests, then choose the mode and accept the separate deployment terms. Copilot remains an option when its own access and readiness checks pass.