Skip to content
Drydock
ConfigurationUpdate Bouncer

Security fast lane

Let an update that fixes critical or high vulnerabilities skip the maturity, snooze and maintenance window holds.

Scans normally inform an update but never change when it applies. The security fast lane is an opt-in policy that changes that for one case: a pending update that fixes critical or high vulnerabilities in the image you're running now. When that's true, the update isn't held back by its maturity min age, a snooze, or a closed maintenance window.

It's off by default.

Turning it on

WhereSettingEffect
Container labeldd.security.fastlane=trueOpts that container in. false opts it out of a global default.
EnvironmentDD_SECURITY_FASTLANE=trueOpts every container in that has no dd.security.fastlane label.
services:
  app:
    image: ghcr.io/example/app:1.4.0
    labels:
      - dd.security.fastlane=true

The fast lane needs Update Bouncer scanning to be on, because it works from scan results.

When an update qualifies

Drydock compares two scans that already exist: the latest scan of the running image, and the scan of the pending candidate. The update is on the fast lane when all of these hold:

  • the candidate has been scanned by digest (repo@sha256:..., never by tag, so the result describes exactly that image), and that scan is for the digest that's pending now;
  • the running image's scan is for the digest that's running, and was taken by digest as well. Each scan records the reference it actually scanned, and it has to be a digest reference for the digest being claimed, so a scan of a tag, a copy of one, or an older record with no reference never counts. Both digests have to be real sha256: content digests. If a registry doesn't return one and Drydock falls back to the tag, there's no fast lane for it;
  • both scans are complete: neither carries an error, even a partial one from one of two scanners, and both recorded their full critical and high ID lists before the stored list was cut off;
  • the candidate scan is by the same scanner and is not older than the running scan, so a stale candidate scan can't be credited against a fresh running scan;
  • at least one critical or high vulnerability ID (a CVE or GHSA) in the running image's scan is missing from the candidate's critical and high IDs;
  • the candidate has no critical ID that the running image didn't already list as critical or high.

An update that only fixes medium or low findings doesn't qualify, and neither does one that brings in a new critical, even if it fixes others.

The fast lane never starts a scan storm. An unscanned candidate simply isn't on the fast lane. For opted-in containers, the scheduled scan scans each pending candidate once, with the same worker count as the rest of the scheduled scan, and the on-demand scan of a container scans its candidate even when a snooze or maturity hold is hiding the update. Containers pending the same candidate digest share one scan, and a candidate is scanned once per candidate digest and running scan, so a failed scan isn't retried until the candidate or the running scan changes.

What it lifts

HoldFast lane
Maturity min age and release-count maturityLifted
SnoozeLifted
Maintenance windowLifted (the auto-update dispatches inside a closed window)

The maintenance window is only lifted for the automatic install. Under scope=scan, the window still decides when the scheduled scan runs.

What it doesn't lift

  • dd.action.exclude and the trigger include rules
  • The trigger threshold
  • Skip and reject entries (skip-tag, skip-digest)
  • Digest pinning
  • A blocked scan of the candidate, a rolled-back candidate, and an update already in progress
  • Approval. In manual mode, or under any group or trigger rule that needs approval, a fast-laned update still waits in the approval queue. The row is marked Security fix so it's easy to spot, and it carries the same hold reasons as before.

The scan gate is not lifted either. A candidate is scanned before the pull and the image that's deployed is only known after it, so the install still runs its own scan of the exact pulled digest, and a blocked running image keeps security-scan-blocked unless the no-worse-than-current gate passes.

The install checks the digest

The update pulls by tag, so after the pull Drydock compares the digest that arrived with the candidate digest that qualified. If the tag was moved in between, or the pulled digest can't be determined, that install is aborted before anything is recreated, a security-fast-lane audit entry records the digest mismatch, and the next scan re-evaluates the new candidate under the normal holds. This applies to Docker and Docker Compose actions alike. On a multi-arch image the tag resolves through a manifest list, and a pull records the list's digest locally rather than the platform image's. Drydock records the list digest at discovery next to the platform digest, and the install accepts either one. Anything else, including a list digest left over from an older candidate or one the registry didn't return, aborts the install and the update goes through the usual holds.

What gets recorded

Every automatic update that only got through because the fast lane lifted a hold writes a security-fast-lane entry to the audit log. It names the container, the versions, the trigger, the fixed IDs and which holds were lifted.

The container, and so every update notification and event that carries it, gets a securityFastLane object while the candidate qualifies:

FieldMeaning
fixedIdsThe fixed critical and high IDs, criticals first, capped at 10
fixedCountHow many were fixed in total
fixedCriticalHow many of them were critical
fixedHighHow many of them were high

The default simple update message adds (security fast lane: fixes N critical or high vulnerabilities) when it's set. In a custom template you can test it like any other field:

${container.securityFastLane ? "Security update: fixes " + container.securityFastLane.fixedCount + " critical or high vulnerabilities" : ""}

The container's update status in the UI shows a Security fast lane row explaining why the usual holds don't apply, and GET /api/containers returns updateEligibility.securityFastLane with the list of bypassedHolds.

The fast lane follows the scans. If a newer scan changes the picture, for example the running image is rescanned or the candidate moves, the container drops off the fast lane and the usual holds apply again.