Skip to main content

Troubleshooting scanning

This page covers the scans of a merge request (secret scan, dependency scan with its dependency source check, CI configuration check) and the analysis of a project's default branch, with its SBOM export. For each problem: what you see, why, and what to do. How scanning works is described in Security overview.

A rule holds for every scan: a scan that could not finish never counts as clean. Its control fails, and in Enforce mode the merge waits until a scan of the latest commit completes.

Where to look first​

  • The scan's card on the merge request page (Secret scan, Dependency scan (SCA), CI configuration check) shows the result or the error.
  • Insights → Jobs: the scan's job (secret_scan, sca_scan, workflow_scan, project_analysis) with its last error and next attempt (see Jobs).
  • The company's Scanning tab, where the settings named below live (see Scanning).

A merge request scan failed​

Scan failed: … on the Secret scan card, The secret scan failed: … on the Policy card​

Why: the scan could not read what it scans. The message says what could not be read and from where, and whether Merge Pilot tries again by itself, for example:

MessageWhy
The changes of the merge request could not be read from GitLab. (or of the pull request … from GitHub)The VCS did not answer within External call timeout (seconds), answered with an error, or refused the connection's token
The title, description and commit messages of the merge request could not be read from …The same, for the texts scanned with Scan texts
The merge request could not be read from …The dependency scan could not read the merge request itself
The CI configuration file … could not be read from … / … listed only part of the changed files, so changed CI configuration may have been missed / N CI configuration files changed; at most 50 are checked / <file> includes N local files; at most 50 are followed / The project's CI/CD configuration path could not be read from GitLabThe CI configuration check could not read a changed CI configuration file, or not all of them
Merge Pilot could not use the merge request's VCS connection (check it with Test connection on the company's Connections tab).The connection is turned off, deleted, or its saved credentials cannot be used
Merge Pilot could not load the merge request and its policy to run the scan. / The secret scanner could not scan the changes. / The scan stopped because of an error in Merge Pilot.A problem in Merge Pilot itself; if it happens again, contact your Merge Pilot administrator

The message ends with Merge Pilot tries again in about N minutes; Re-scan tries at once. when When the scan cannot finish retries automatically, and with Use Re-scan to try again. (or Re-scan dependencies, Check the CI configuration again, Scan the container images again, Check the infrastructure files again) otherwise. A scan failed only by a limit or by a file list the VCS cut short ends with Trying again would not change this: change the merge request or the company's scan settings, then use Re-scan. instead: it is never retried by itself, because it would fail the same way. The company's When a scan cannot see everything can let such scans check what they could and name the rest (Not looked at: …) instead of failing (see Scanning).

The scan could not finish, even after Merge Pilot tried it again several times. Use Re-scan to try again. (or Re-scan dependencies, Check the CI configuration again, Scan the container images again, Check the infrastructure files again) means the scan kept failing for a reason Merge Pilot retries by itself, for example the VCS still giving the changes of an older commit. The technical reason is in the audit log (job_failed) and on the Jobs page. Start the scan again from the merge request; if it keeps failing, check the connection with Test connection.

What to do:

  1. Fix the cause if there is one (the connection, see Test connection fails).
  2. Select Re-scan on the card. A failed scan is retried on its own only when When the scan cannot finish is Fail, retry automatically (after a pause, 10 minutes by default); with the default Fail, rescan by hand it waits for someone to re-scan.
  3. For large merge requests, raise External call timeout (seconds) on the company's Scanning tab.

A VCS rate limit does not fail a scan: the job waits for the limit to reset (shown as Retry on the Jobs page).

The dependency scan could not be completed: … Fix the cause and rescan.​

The card says The scan could not be completed, so the change is not known to be safe with the reasons:

ReasonWhyWhat to do
N manifests changed; only the first 20 were scannedMore than 20 dependency manifests changed in one merge requestSplit the change into smaller merge requests
N new or changed dependencies; only M were checked (limit)More new or changed dependencies than Max dependencies per scanRaise the limit on the company's Scanning tab → Dependencies, or split the change
<file> could not be read: … or <file> on the target branch could not be read: …A manifest could not be fetched or parsed (a manifest over 5 MB cannot be read)Fix the file, or exclude its path with the company's Blocked paths if it should not be scanned
the VCS listed only part of the changed files, so changed manifests may have been missedThe change is too large for the VCS to list every fileSplit the change
OSV could not be asked about the new dependencies: …The vulnerability database could not be reached or did not answer in timeRe-scan later; raise External call timeout (seconds) if it keeps timing out

Then select Re-scan (Re-scan dependencies). See Dependency scan → Limits.

The dependency source check could not be completed: … Fix the cause and rescan.​

The Dependency sources section of the Dependency scan (SCA) card says Could not be checked: …:

ReasonWhyWhat to do
more than 20 dependency files changed; only the first 20 were checkedMore than 20 changed lockfiles and manifestsSplit the change
<file> could not be read: … / <file> on the target branch could not be read: …The VCS did not give the file, or the file is over 5 MB (… exceeds max size …)Re-scan; check the connection if it keeps failing. A file over 5 MB cannot be checked: leave its path out with the dependency scan's Blocked paths if it should not be
reading <file>: … / <file>: more than 50000 entriesThe file at the head commit does not parse, or has more than 50,000 entriesFix the file, or leave its path out with Blocked paths
the VCS listed only part of the changed files, so changed dependency files may have been missedThe change is too large for the VCS to list every fileSplit the change

It follows the dependency scan's When the scan cannot finish setting. The limit of 20 files and a cut-short file list follow the company's When a scan cannot see everything, as the dependency scan does: with fail (the default) such a failure is final (Trying again would not change this); with note the check checks what was listed, or the first 20, and says what it did not look at. See Dependency sources → Limits.

The container image scan could not be completed: …​

The card says why for each image:

MessageWhyWhat to do
unable to find the specified image …The image or tag does not exist, or the registry hides it from an anonymous pullCheck the reference; add the registry under Company → Scanning → Container registries
signing in to …: the registry refused the credentialsThe registry's sign-in is wrong or expiredFix it under Container registries and use Test
… private network address of a host that is not configuredThe image is on a host on a private network that is not a registry of the companyAdd the registry, or the host to Internal hosts
the vulnerability database is not available yetMerge Pilot is still downloading its vulnerability data (after a start)Re-scan in a few minutes; if it persists, ask the administrator: the installation may not reach the vulnerability data
the scan took longer than …A very large imageRe-scan; the administrator can raise the scan timeout

A reference that depends on a variable or template (N image reference(s) depend on a variable or template …) is not an error of the scan: give the reference a default (ARG BASE=alpine:3.21), or set Image references that cannot be resolved to List only.

A project analysis says images or infrastructure files were not checked​

The image and infrastructure scanner is not installed on the server …: this installation of Merge Pilot cannot scan images or infrastructure files. Nothing of the images or infrastructure is taken as fixed meanwhile. Ask the administrator. The vulnerability database is still being downloaded …: Merge Pilot was just started; run the analysis again in a few minutes. N images could not be scanned: the images keep their findings; the reasons are those of the table above (an image not found, a registry sign-in refused, a host not allowed), and a merge request that refers to such an image shows them per image on its Container image scan card.

N changed infrastructure file(s) could not be checked​

The Infrastructure check card lists each file and why: the file could not be parsed: … (fix the file), or the Helm chart could not be rendered: … (the chart needs values only given when it is installed: give them defaults in values.yaml, leave the chart out with Files not checked, or set Changed files that cannot be checked to List only). A module could not be loaded … means a Terraform module was not downloaded (a git:: source, or a host the scans may not reach); the file was still checked.

The CI configuration check could not be completed: …​

The CI configuration check could not read the changed CI configuration files (see the table above). It follows the company's When the scan cannot finish setting, like the secret scan. Fix the cause, then Re-scan on the CI configuration check card.

The scan stays "in progress"​

Why: the scan's job is waiting in the queue, retrying, or waiting for a VCS rate limit.

What to do: look for the job on the Jobs page. A job runs for at most 10 minutes; a scan whose job failed after its retries is marked failed instead of staying pending, and Re-scan starts it again. While a scan is waiting or running, Re-scan is disabled (through the API: scan already in progress).

Files were not scanned​

No secrets found. Warning: N changed file(s) could not be scanned (no diff from the VCS).​

Why: the VCS sent no diff for some files (very large or cut-off files), or listed only part of the changed files (the VCS listed only part of the changed files). Under Files the VCS sends no diff for = Warn, do not hold (the default) the change can still pass.

What to do: review those files by hand; the card lists them under Not everything could be scanned. To hold such changes instead, set Hold the change: the line then reads No secrets found, but N changed file(s) could not be scanned (no diff from the VCS). and the control fails until the change is split. See Secret scan → Settings.

Not scanned: scans are off for watched repositories.​

Why: the repository is watched and the company does not scan watched merge requests. A Re-scan is refused with scans are off for watched repositories (Company → Repositories).

What to do: turn on Scan watched merge requests for secrets, vulnerable dependencies and licenses on the company's Repositories tab (see Watching repositories).

No scan card on the merge request​

A card is shown only once the scan has been started for the merge request. The policy that covers it does not run that scan (its control is Off), so there is nothing to re-scan. Check the policy's controls (see Controls).

Dependency sources​

Lockfile not updated although the dependencies did not change​

lockfile-missing-update is skipped for a package.json or go.mod whose dependency sections are unchanged, but other manifests (pyproject.toml, Pipfile, Cargo.toml, Gemfile, composer.json) count on any change, such as a new project version or description. And a package.json change to dependencies, devDependencies, optionalDependencies, peerDependencies, the bundled dependencies or overrides always counts.

What to do: run the package manager so it writes the lockfile, and commit the lockfile with the manifest; or, when the lockfile really needs no change, accept the finding (in Report mode it does not hold the merge). A repository that keeps no lockfile for the manifest gets no finding.

Registry not allowed for our private registry​

The host is not on the allow-list of that ecosystem. What to do: add it under Private registries for the ecosystem (Company → Scanning → Dependencies → Dependency sources, or the policy's Registries override): the host as the lockfile shows it, with the port when there is one (npm.example.com, nexus.example.com:8443), or with a path to allow only one repository on a shared host (nexus.example.com/repository/npm-internal). Entries are compared exactly, without wildcards. When everything goes through a mirror of the public registry, add the mirror and turn on Only the listed registries. Takes effect at the next scan: select Re-scan dependencies.

A registry-only entry whose lockfile records no host (for example pnpm entries with only an integrity hash, or yarn berry npm: entries) is never a finding: the registry is set in the package manager's configuration, not in the lockfile.

Local path for the project's own packages​

A Cargo.lock package without a source (the project's own crates and path dependencies), an npm link leaving the repository, a uv editable or directory package and a go.mod replace to a local directory count as local paths. They are findings only when they are new in the merge request: a version bump of one that was already there is not, and npm links to folders inside the repository (workspaces) never are. What to do: turn on Allow local path dependencies, or list those packages under Packages left out.

Source unknown (review)​

The lockfile writes this entry's source in a form Merge Pilot does not read (for example a yarn berry file:, portal: or exec: entry, a pnpm local tarball, a pip hg+ or svn+ requirement, or a Gemfile.lock plugin source). It is listed for review and never holds the merge. Check where the entry comes from; when its host is not an allowed registry, the entry is reported as Registry not allowed instead.

A dependency from a git repository or URL is intended​

Turn on Allow git dependencies or Allow URL dependencies for the company or one policy, or list the package under Packages left out.

Too many findings​

Only Max findings per scan findings are stored and listed (50 by default); the count includes all of them. For false positives, use gitleaks:allow on the line, Allowed paths or Allowed rules (see False positives). For the dependency scan, Max vulnerabilities stored keeps the most severe.

Re-scan is refused​

MessageWhy
the merge request is merged: only open merge requests are scanned again; its recorded scans stay as they were (or closed)A merged or closed merge request's scans are the record of what merged. The button is not shown
scans are off for watched repositories (Company → Repositories)See above
the merge request's policy does not run the CI configuration checkThe CI configuration check is Off in the merge request's policy

See Re-scan.

Project analysis​

Run analysis is missing​

Only company admins and platform admins start an analysis. Other roles read the results.

analysis already running​

An analysis of the project is running. One that runs for more than 65 minutes is marked Failed first, and a new one can start.

the last analysis completed less than 5 minutes ago; run it again in N minutes / the last analysis failed less than 5 minutes ago; run it again in N minutes​

Why: another analysis cannot start within 5 minutes after the last one completed, or failed for a reason that is not temporary. The 5 minutes count from any analysis of the project, including one started by a merge or a direct change.

What to do: wait the minutes the message names. After a temporary failure (a timeout, a network error, a VCS server error) the button reads Retry and works at once. See Failures, retry and cooldown.

Failed: The analysis did not finish within 65 minutes and was stopped. or The analysis did not start within 65 minutes (Merge Pilot was busy with other work) and was stopped.​

Why: the analysis was still running after 65 minutes, or it was never picked up within that time. Both messages end with Run it again; it can be run again at once. The analysis did not finish in time and was stopped. means the same, found when another analysis of the project was about to start: the stopped one is ended so the new one can run.

What to do: select Retry. Look at the Jobs page for a project_analysis job waiting in Retry (for example for a VCS rate limit) or failing.

Results may be incomplete​

The analysis completed but could not see everything; each reason is listed (for example Manifest limit reached: …, Dependency limit reached; set it on the company under Scanning → Dependencies, The vulnerability lookup failed; run the analysis again). Nothing it did not see is taken as fixed. What each reason means and what to change is in Results may be incomplete:

  • raise Max dependencies per project analysis or Project analysis timeout (seconds) on the company's Scanning tab → Dependencies;
  • exclude test fixtures and examples with the project's Excluded paths;
  • commit lockfiles, so dependencies have exact versions;
  • run the analysis again after a failed lookup.

The analysis is marked Stale​

The last completed analysis is older than 7 days. Nothing runs it again on its own except a merge or direct change that touches a dependency manifest (when the company's settings say so); select Run analysis.

Exploited (KEV) and EPSS marks are missing​

The Vulnerabilities page says when the feeds were read. Exploitation data (CISA KEV, FIRST EPSS) has not been read yet. means they were not read so far; … is not read on this installation. means the installation does not read them; The last refresh failed, the earlier data is kept: … keeps the previous data. See KEV and EPSS.

SBOM export is refused​

No Export SBOM button​

The button shows once an analysis of the project has completed. While the first analysis runs, or when it failed, there is nothing to export: run an analysis first.

this project has not been analysed yet: run an analysis first​

The project's dependencies have never been recorded. Run an analysis on the Security tab, then export.

format must be cyclonedx or spdx​

API only: format must be cyclonedx or spdx.

Exported, but not every manifest of the branch could be read: the SBOM may leave dependencies out (see the analysis).​

The export worked, but the last analysis could not read every manifest. The document says so too (merge-pilot:incomplete in CycloneDX, the creator comment in SPDX; the API sends X-SBOM-Incomplete: true). Fix what the analysis reports under Results may be incomplete, run it again, and export again. See Incomplete SBOMs.

The SBOM is out of date​

The SBOM is built from the last completed analysis, not read fresh. Run an analysis first for an up-to-date one.