Integrity check Rake task

DETAILS: Tier: Free, Premium, Ultimate Offering: Self-managed

GitLab provides Rake tasks to check the integrity of various components. See also the check GitLab configuration Rake task.

Repository integrity

Even though Git is very resilient and tries to prevent data integrity issues, there are times when things go wrong. The following Rake tasks intend to help GitLab administrators diagnose problem repositories so they can be fixed.

These Rake tasks use three different methods to determine the integrity of Git repositories.

Git repository file system check (git fsck). This step verifies the connectivity and validity of objects in the repository.
Check for config.lock in the repository directory.
Check for any branch/references lock files in refs/heads.

The existence of config.lock or reference locks alone do not necessarily indicate a problem. Lock files are routinely created and removed as Git and GitLab perform operations on the repository. They serve to prevent data integrity issues. However, if a Git operation is interrupted these locks may not be cleaned up properly.

The following symptoms may indicate a problem with repository integrity. If users experience these symptoms you may use the Rake tasks described below to determine exactly which repositories are causing the trouble.

Receiving an error when trying to push code - remote: error: cannot lock ref
A 500 error when viewing the GitLab dashboard or when accessing a specific project.

Check project code repositories

This task loops through the project code repositories and runs the integrity check described previously. If a project uses a pool repository, that is also checked. Other types of Git repositories are not checked.

Linux package installations:
```
sudo gitlab-rake gitlab:git:fsck
```

Self-compiled installations:

sudo -u git -H bundle exec rake gitlab:git:fsck RAILS_ENV=production

Checksum of repository refs

One Git repository can be compared to another by checksumming all refs of each repository. If both repositories have the same refs, and if both repositories pass an integrity check, then we can be confident that both repositories are the same.

For example, this can be used to compare a backup of a repository against the source repository.

Check all GitLab repositories

This task loops through all repositories on the GitLab server and outputs checksums in the format <PROJECT ID>,<CHECKSUM>.

If a repository doesn't exist, the project ID is a blank checksum.
If a repository exists but is empty, the output checksum is 0000000000000000000000000000000000000000.
Projects which don't exist are skipped.

Linux package installations:

sudo gitlab-rake gitlab:git:checksum_projects

Self-compiled installations:

sudo -u git -H bundle exec rake gitlab:git:checksum_projects RAILS_ENV=production

For example, if:

Project with ID#2 doesn't exist, it is skipped.
Project with ID#4 doesn't have a repository, its checksum is blank.
Project with ID#5 has an empty repository, its checksum is 0000000000000000000000000000000000000000.

The output would then look something like:

1,cfa3f06ba235c13df0bb28e079bcea62c5848af2
3,3f3fb58a8106230e3a6c6b48adc2712fb3b6ef87
4,
5,0000000000000000000000000000000000000000
6,6c6b48adc2712fb3b6ef87cfa3f06ba235c13df0

Check specific GitLab repositories

Optionally, specific project IDs can be checksummed by setting an environment variable CHECKSUM_PROJECT_IDS with a list of comma-separated integers, for example:

sudo CHECKSUM_PROJECT_IDS="1,3" gitlab-rake gitlab:git:checksum_projects

Uploaded files integrity

Various types of files can be uploaded to a GitLab installation by users. These integrity checks can detect missing files. Additionally, for locally stored files, checksums are generated and stored in the database upon upload, and these checks verify them against current files.

Integrity checks are supported for the following types of file:

CI artifacts
LFS objects
Project-level Secure Files (introduced in GitLab 16.1.0)
User uploads

Linux package installations:

sudo gitlab-rake gitlab:artifacts:check
sudo gitlab-rake gitlab:ci_secure_files:check
sudo gitlab-rake gitlab:lfs:check
sudo gitlab-rake gitlab:uploads:check

Self-compiled installations:

sudo -u git -H bundle exec rake gitlab:artifacts:check RAILS_ENV=production
sudo -u git -H bundle exec rake gitlab:ci_secure_files:check RAILS_ENV=production
sudo -u git -H bundle exec rake gitlab:lfs:check RAILS_ENV=production
sudo -u git -H bundle exec rake gitlab:uploads:check RAILS_ENV=production

These tasks also accept some environment variables which you can use to override certain values:

Variable	Type	Description
`BATCH`	integer	Specifies the size of the batch. Defaults to 200.
`ID_FROM`	integer	Specifies the ID to start from, inclusive of the value.
`ID_TO`	integer	Specifies the ID value to end at, inclusive of the value.
`VERBOSE`	boolean	Causes failures to be listed individually, rather than being summarized.

sudo gitlab-rake gitlab:artifacts:check BATCH=100 ID_FROM=50 ID_TO=250
sudo gitlab-rake gitlab:ci_secure_files:check BATCH=100 ID_FROM=50 ID_TO=250
sudo gitlab-rake gitlab:lfs:check BATCH=100 ID_FROM=50 ID_TO=250
sudo gitlab-rake gitlab:uploads:check BATCH=100 ID_FROM=50 ID_TO=250

Example output:

$ sudo gitlab-rake gitlab:uploads:check
Checking integrity of Uploads
- 1..1350: Failures: 0
- 1351..2743: Failures: 0
- 2745..4349: Failures: 2
- 4357..5762: Failures: 1
- 5764..7140: Failures: 2
- 7142..8651: Failures: 0
- 8653..10134: Failures: 0
- 10135..11773: Failures: 0
- 11777..13315: Failures: 0
Done!

Example verbose output:

$ sudo gitlab-rake gitlab:uploads:check VERBOSE=1
Checking integrity of Uploads
- 1..1350: Failures: 0
- 1351..2743: Failures: 0
- 2745..4349: Failures: 2
  - Upload: 3573: #<Errno::ENOENT: No such file or directory @ rb_sysopen - /opt/gitlab/embedded/service/gitlab-rails/public/uploads/user-foo/project-bar/7a77cc52947bfe188adeff42f890bb77/image.png>
  - Upload: 3580: #<Errno::ENOENT: No such file or directory @ rb_sysopen - /opt/gitlab/embedded/service/gitlab-rails/public/uploads/user-foo/project-bar/2840ba1ba3b2ecfa3478a7b161375f8a/pug.png>
- 4357..5762: Failures: 1
  - Upload: 4636: #<Google::Apis::ServerError: Server error>
- 5764..7140: Failures: 2
  - Upload: 5812: #<NoMethodError: undefined method `hashed_storage?' for nil:NilClass>
  - Upload: 5837: #<NoMethodError: undefined method `hashed_storage?' for nil:NilClass>
- 7142..8651: Failures: 0
- 8653..10134: Failures: 0
- 10135..11773: Failures: 0
- 11777..13315: Failures: 0
Done!

LDAP check

The LDAP check Rake task tests the bind DN and password credentials (if configured) and lists a sample of LDAP users. This task is also executed as part of the gitlab:check task, but can run independently. See LDAP Rake Tasks - LDAP Check for details.

Verify database values can be decrypted using the current secrets

This task runs through all possible encrypted values in the database, verifying that they are decryptable using the current secrets file (gitlab-secrets.json).

Automatic resolution is not yet implemented. If you have values that cannot be decrypted, you can follow steps to reset them, see our documentation on what to do when the secrets file is lost.

This can take a very long time, depending on the size of your database, as it checks all rows in all tables.

Linux package installations:
```
sudo gitlab-rake gitlab:doctor:secrets
```

Self-compiled installations:

bundle exec rake gitlab:doctor:secrets RAILS_ENV=production

Example output

I, [2020-06-11T17:17:54.951815 #27148]  INFO -- : Checking encrypted values in the database
I, [2020-06-11T17:18:12.677708 #27148]  INFO -- : - ApplicationSetting failures: 0
I, [2020-06-11T17:18:12.823692 #27148]  INFO -- : - User failures: 0
[...] other models possibly containing encrypted data
I, [2020-06-11T17:18:14.938335 #27148]  INFO -- : - Group failures: 1
I, [2020-06-11T17:18:15.559162 #27148]  INFO -- : - Operations::FeatureFlagsClient failures: 0
I, [2020-06-11T17:18:15.575533 #27148]  INFO -- : - ScimOauthAccessToken failures: 0
I, [2020-06-11T17:18:15.575678 #27148]  INFO -- : Total: 1 row(s) affected
I, [2020-06-11T17:18:15.575711 #27148]  INFO -- : Done!

Verbose mode

To get more detailed information about which rows and columns can't be decrypted, you can pass a VERBOSE environment variable:

Linux package installations:

sudo gitlab-rake gitlab:doctor:secrets VERBOSE=1

Self-compiled installations:

bundle exec rake gitlab:doctor:secrets RAILS_ENV=production VERBOSE=1

Example verbose output

I, [2020-06-11T17:17:54.951815 #27148]  INFO -- : Checking encrypted values in the database
I, [2020-06-11T17:18:12.677708 #27148]  INFO -- : - ApplicationSetting failures: 0
I, [2020-06-11T17:18:12.823692 #27148]  INFO -- : - User failures: 0
[...] other models possibly containing encrypted data
D, [2020-06-11T17:19:53.224344 #27351] DEBUG -- : > Something went wrong for Group[10].runners_token: Validation failed: Route can't be blank
I, [2020-06-11T17:19:53.225178 #27351]  INFO -- : - Group failures: 1
D, [2020-06-11T17:19:53.225267 #27351] DEBUG -- :   - Group[10]: runners_token
I, [2020-06-11T17:18:15.559162 #27148]  INFO -- : - Operations::FeatureFlagsClient failures: 0
I, [2020-06-11T17:18:15.575533 #27148]  INFO -- : - ScimOauthAccessToken failures: 0
I, [2020-06-11T17:18:15.575678 #27148]  INFO -- : Total: 1 row(s) affected
I, [2020-06-11T17:18:15.575711 #27148]  INFO -- : Done!

Reset encrypted tokens when they can't be recovered

Introduced in GitLab 16.6.

WARNING: This operation is dangerous and can result in data-loss. Proceed with extreme caution. You must have knowledge about GitLab internals before you perform this operation.

In some cases, encrypted tokens can no longer be recovered and cause issues. Most often, runner registration tokens for groups and projects might be broken on very large instances.

To reset broken tokens:

Identify the database models that have broken encrypted tokens. For example, it can be Group and Project.
Identify the broken tokens. For example runners_token.

To reset broken tokens, run gitlab:doctor:reset_encrypted_tokens with VERBOSE=true MODEL_NAMES=Model1,Model2 TOKEN_NAMES=broken_token1,broken_token2. For example:

VERBOSE=true MODEL_NAMES=Project,Group TOKEN_NAMES=runners_token bundle exec rake gitlab:doctor:reset_encrypted_tokens

You will see every action this task would try to perform:

I, [2023-09-26T16:20:23.230942 #88920]  INFO -- : Resetting runners_token on Project, Group if they can not be read
I, [2023-09-26T16:20:23.230975 #88920]  INFO -- : Executing in DRY RUN mode, no records will actually be updated
D, [2023-09-26T16:20:30.151585 #88920] DEBUG -- : > Fix Project[1].runners_token
I, [2023-09-26T16:20:30.151617 #88920]  INFO -- : Checked 1/9 Projects
D, [2023-09-26T16:20:30.151873 #88920] DEBUG -- : > Fix Project[3].runners_token
D, [2023-09-26T16:20:30.152975 #88920] DEBUG -- : > Fix Project[10].runners_token
I, [2023-09-26T16:20:30.152992 #88920]  INFO -- : Checked 11/29 Projects
I, [2023-09-26T16:20:30.153230 #88920]  INFO -- : Checked 21/29 Projects
I, [2023-09-26T16:20:30.153882 #88920]  INFO -- : Checked 29 Projects
D, [2023-09-26T16:20:30.195929 #88920] DEBUG -- : > Fix Group[22].runners_token
I, [2023-09-26T16:20:30.196125 #88920]  INFO -- : Checked 1/19 Groups
D, [2023-09-26T16:20:30.196192 #88920] DEBUG -- : > Fix Group[25].runners_token
D, [2023-09-26T16:20:30.197557 #88920] DEBUG -- : > Fix Group[82].runners_token
I, [2023-09-26T16:20:30.197581 #88920]  INFO -- : Checked 11/19 Groups
I, [2023-09-26T16:20:30.198455 #88920]  INFO -- : Checked 19 Groups
I, [2023-09-26T16:20:30.198462 #88920]  INFO -- : Done!

If you are confident that this operation resets the correct tokens, disable dry-run mode and run the operation again:

DRY_RUN=false VERBOSE=true MODEL_NAMES=Project,Group TOKEN_NAMES=runners_token bundle exec rake gitlab:doctor:reset_encrypted_tokens

Troubleshooting

The following are solutions to problems you might discover using the Rake tasks documented above.

Dangling objects

The gitlab-rake gitlab:git:fsck task can find dangling objects such as:

dangling blob a12...
dangling commit b34...
dangling tag c56...
dangling tree d78...

To delete them, try running housekeeping.

If the issue persists, try triggering garbage collection via the Rails Console:

p = Project.find_by_path("project-name")
Repositories::HousekeepingService.new(p, :gc).execute

If the dangling objects are younger than the 2 weeks default grace period, and you don't want to wait until they expire automatically, run:

Repositories::HousekeepingService.new(p, :prune).execute

Delete references to missing remote uploads

gitlab-rake gitlab:uploads:check VERBOSE=1 detects remote objects that do not exist because they were deleted externally but their references still exist in the GitLab database.

Example output with error message:

$ sudo gitlab-rake gitlab:uploads:check VERBOSE=1
Checking integrity of Uploads
- 100..434: Failures: 2
- Upload: 100: Remote object does not exist
- Upload: 101: Remote object does not exist
Done!

To delete these references to remote uploads that were deleted externally, open the GitLab Rails Console and run:

uploads_deleted=0
Upload.find_each do |upload|
  next if upload.retrieve_uploader.file.exists?
  uploads_deleted=uploads_deleted + 1
  p upload                            ### allow verification before destroy
  # p upload.destroy!                 ### uncomment to actually destroy
end
p "#{uploads_deleted} remote objects were destroyed."

Delete references to missing artifacts

gitlab-rake gitlab:artifacts:check VERBOSE=1 detects when artifacts (or job.log files):

Are deleted outside of GitLab.
Have references still in the GitLab database.

When this scenario is detected, the Rake task displays an error message. For example:

Checking integrity of Job artifacts
- 1..15: Failures: 2
  - Job artifact: 9: #<Errno::ENOENT: No such file or directory @ rb_sysopen - /var/opt/gitlab/gitlab-rails/shared/artifacts/4b/22/4b227777d4dd1fc61c6f884f48641d02b4d121d3fd328cb08b5531fcacdabf8a/2022_06_30/8/9/job.log>
  - Job artifact: 15: Remote object does not exist
Done!

To delete these references to missing local and/or remote artifacts (job.log files):

Open the GitLab Rails Console.

Run the following Ruby code:

artifacts_deleted = 0
::Ci::JobArtifact.find_each do |artifact|                      ### Iterate artifacts
#  next if artifact.file.filename != "job.log"                 ### Uncomment if only `job.log` files' references are to be processed
  next if artifact.file.file.exists?                           ### Skip if the file reference is valid
  artifacts_deleted += 1
  puts "#{artifact.id}  #{artifact.file.path} is missing."     ### Allow verification before destroy
#  artifact.destroy!                                           ### Uncomment to actually destroy
end
puts "Count of identified/destroyed invalid references: #{artifacts_deleted}"

Delete references to missing LFS objects

If gitlab-rake gitlab:lfs:check VERBOSE=1 detects LFS objects that exist in the database but not on disk, follow the procedure in the LFS documentation to remove the database entries.

Update dangling object storage references

If you have migrated from object storage to local storage and files were missing, then dangling database references remain.

This is visible in the migration logs with errors like the following:

W, [2022-11-28T13:14:09.283833 #10025]  WARN -- : Failed to transfer Ci::JobArtifact ID 11 with error: undefined method `body' for nil:NilClass
W, [2022-11-28T13:14:09.296911 #10025]  WARN -- : Failed to transfer Ci::JobArtifact ID 12 with error: undefined method `body' for nil:NilClass

Attempting to delete references to missing artifacts after you have disabled object storage, results in the following error:

RuntimeError (Object Storage is not enabled for JobArtifactUploader)

To update these references to point to local storage:

Open the GitLab Rails Console.

Run the following Ruby code:

artifacts_updated = 0
::Ci::JobArtifact.find_each do |artifact|                    ### Iterate artifacts
  next if artifact.file_store != 2                           ### Skip if file_store already points to local storage
  artifacts_updated += 1
  # artifact.update(file_store: 1)                           ### Uncomment to actually update
end
puts "Updated file_store count: #{artifacts_updated}"

The script to delete references to missing artifacts now functions correctly and cleans up the database.

Delete references to missing secure files

VERBOSE=1 gitlab-rake gitlab:ci_secure_files:check detects when secure files:

Are deleted outside of GitLab.
Have references still in the GitLab database.

When this scenario is detected, the Rake task displays an error message. For example:

Checking integrity of CI Secure Files
- 1..15: Failures: 2
  - Job SecureFile: 9: #<Errno::ENOENT: No such file or directory @ rb_sysopen - /var/opt/gitlab/gitlab-rails/shared/ci_secure_files/4b/22/4b227777d4dd1fc61c6f884f48641d02b4d121d3fd328cb08b5531fcacdabf8a/2022_06_30/8/9/distribution.cer>
  - Job SecureFile: 15: Remote object does not exist
Done!

To delete these references to missing local or remote secure files:

Open the GitLab Rails Console.

Run the following Ruby code:

secure_files_deleted = 0
::Ci::SecureFile.find_each do |secure_file|                    ### Iterate secure files
  next if secure_file.file.file.exists?                        ### Skip if the file reference is valid
  secure_files_deleted += 1
  puts "#{secure_file.id}  #{secure_file.file.path} is missing."     ### Allow verification before destroy
#  secure_file.destroy!                                           ### Uncomment to actually destroy
end
puts "Count of identified/destroyed invalid references: #{secure_files_deleted}"