gitlab-org--gitlab-foss/doc/raketasks/backup_restore.md

29 KiB

stage group info
Systems Geo To determine the technical writer assigned to the Stage/Group associated with this page, see https://about.gitlab.com/handbook/engineering/ux/technical-writing/#assignments

Back up and restore GitLab (FREE SELF)

GitLab provides Rake tasks for backing up and restoring GitLab instances.

An application data backup creates an archive file that contains the database, all repositories and all attachments.

You can only restore a backup to exactly the same version and type (CE/EE) of GitLab on which it was created. The best way to migrate your projects from one server to another is through a backup and restore.

WARNING: GitLab doesn't back up items that aren't stored on the file system. If you're using object storage, be sure to enable backups with your object storage provider, if desired.

Requirements

To be able to back up and restore, ensure that Rsync is installed on your system. If you installed GitLab:

  • Using the Omnibus package, Rsync is already installed.

  • From source, check if rsync is installed. If Rsync is not installed, install it. For example:

    # Debian/Ubuntu
    sudo apt-get install rsync
    
    # RHEL/CentOS
    sudo yum install rsync
    

gitaly-backup for repository backup and restore

The gitaly-backup binary is used by the backup Rake task to create and restore repository backups from Gitaly. gitaly-backup replaces the previous backup method that directly calls RPCs on Gitaly from GitLab.

The backup Rake task must be able to find this executable. In most cases, you don't need to change the path to the binary as it should work fine with the default path /opt/gitlab/embedded/bin/gitaly-backup. If you have a specific reason to change the path, it can be configured in Omnibus GitLab packages:

  1. Add the following to /etc/gitlab/gitlab.rb:

    gitlab_rails['backup_gitaly_backup_path'] = '/path/to/gitaly-backup'
    
  2. Reconfigure GitLab for the changes to take effect.

Backup timestamp

The backup archive is saved in backup_path, which is specified in the config/gitlab.yml file. The filename is [TIMESTAMP]_gitlab_backup.tar, where TIMESTAMP identifies the time at which each backup was created, plus the GitLab version. The timestamp is needed if you need to restore GitLab and multiple backups are available.

For example, if the backup name is 1493107454_2018_04_25_10.6.4-ce_gitlab_backup.tar, the timestamp is 1493107454_2018_04_25_10.6.4-ce.

Back up GitLab

For detailed information on backing up GitLab, see Backup GitLab.

Restore GitLab

For detailed information on restoring GitLab, see Restore GitLab.

Alternative backup strategies

If your GitLab instance contains a lot of Git repository data, you may find the GitLab backup script to be too slow. If your GitLab instance has a lot of forked projects, the regular backup task also duplicates the Git data for all of them. In these cases, consider using file system snapshots as part of your backup strategy.

Example: Amazon Elastic Block Store (EBS)

A GitLab server using Omnibus GitLab hosted on Amazon AWS. An EBS drive containing an ext4 file system is mounted at /var/opt/gitlab. In this case you could make an application backup by taking an EBS snapshot. The backup includes all repositories, uploads and PostgreSQL data.

Example: Logical Volume Manager (LVM) snapshots + rsync

A GitLab server using Omnibus GitLab, with an LVM logical volume mounted at /var/opt/gitlab. Replicating the /var/opt/gitlab directory using rsync would not be reliable because too many files would change while rsync is running. Instead of rsync-ing /var/opt/gitlab, we create a temporary LVM snapshot, which we mount as a read-only file system at /mnt/gitlab_backup. Now we can have a longer running rsync job which creates a consistent replica on the remote server. The replica includes all repositories, uploads and PostgreSQL data.

If you're running GitLab on a virtualized server, you can possibly also create VM snapshots of the entire GitLab server. It's not uncommon however for a VM snapshot to require you to power down the server, which limits this solution's practical use.

Back up repository data separately

First, ensure you back up existing GitLab data while skipping repositories:

# for Omnibus GitLab package installations
sudo gitlab-backup create SKIP=repositories

# for installations from source:
sudo -u git -H bundle exec rake gitlab:backup:create SKIP=repositories RAILS_ENV=production

For manually backing up the Git repository data on disk, there are multiple possible strategies:

Prevent writes and copy the Git repository data

Git repositories must be copied in a consistent way. They should not be copied during concurrent write operations, as this can lead to inconsistencies or corruption issues. For more details, issue #270422 has a longer discussion explaining the potential problems.

To prevent writes to the Git repository data, there are two possible approaches:

  • Use maintenance mode to place GitLab in a read-only state.

  • Create explicit downtime by stopping all Gitaly services before backing up the repositories:

    sudo gitlab-ctl stop gitaly
    # execute git data copy step
    sudo gitlab-ctl start gitaly
    

You can copy Git repository data using any method, as long as writes are prevented on the data being copied (to prevent inconsistencies and corruption issues). In order of preference and safety, the recommended methods are:

  1. Use rsync with archive-mode, delete, and checksum options, for example:

    rsync -aR --delete --checksum source destination # be extra safe with the order as it will delete existing data if inverted
    
  2. Use a tar pipe to copy the entire repository's directory to another server or location.

  3. Use sftp, scp, cp, or any other copying method.

Online backup through marking repositories as read-only (experimental)

One way of backing up repositories without requiring instance-wide downtime is to programmatically mark projects as read-only while copying the underlying data.

There are a few possible downsides to this:

  • Repositories are read-only for a period of time that scales with the size of the repository.
  • Backups take a longer time to complete due to marking each project as read-only, potentially leading to inconsistencies. For example, a possible date discrepancy between the last data available for the first project that gets backed up compared to the last project that gets backed up.
  • Fork networks should be entirely read-only while the projects inside get backed up to prevent potential changes to the pool repository.

There is an experimental script that attempts to automate this process in the Geo team Runbooks project.

Back up and restore for installations using PgBouncer

Do not back up or restore GitLab through a PgBouncer connection. These tasks must bypass PgBouncer and connect directly to the PostgreSQL primary database node, or they cause a GitLab outage.

When the GitLab backup or restore task is used with PgBouncer, the following error message is shown:

ActiveRecord::StatementInvalid: PG::UndefinedTable

Each time the GitLab backup runs, GitLab starts generating 500 errors and errors about missing tables will be logged by PostgreSQL:

ERROR: relation "tablename" does not exist at character 123

This happens because the task uses pg_dump, which sets a null search path and explicitly includes the schema in every SQL query to address CVE-2018-1058.

Since connections are reused with PgBouncer in transaction pooling mode, PostgreSQL fails to search the default public schema. As a result, this clearing of the search path causes tables and columns to appear missing.

Bypassing PgBouncer

There are two ways to fix this:

  1. Use environment variables to override the database settings for the backup task.
  2. Reconfigure a node to connect directly to the PostgreSQL primary database node.

Environment variable overrides

By default, GitLab uses the database configuration stored in a configuration file (database.yml). However, you can override the database settings for the backup and restore task by setting environment variables that are prefixed with GITLAB_BACKUP_:

  • GITLAB_BACKUP_PGHOST
  • GITLAB_BACKUP_PGUSER
  • GITLAB_BACKUP_PGPORT
  • GITLAB_BACKUP_PGPASSWORD
  • GITLAB_BACKUP_PGSSLMODE
  • GITLAB_BACKUP_PGSSLKEY
  • GITLAB_BACKUP_PGSSLCERT
  • GITLAB_BACKUP_PGSSLROOTCERT
  • GITLAB_BACKUP_PGSSLCRL
  • GITLAB_BACKUP_PGSSLCOMPRESSION

For example, to override the database host and port to use 192.168.1.10 and port 5432 with the Omnibus package:

sudo GITLAB_BACKUP_PGHOST=192.168.1.10 GITLAB_BACKUP_PGPORT=5432 /opt/gitlab/bin/gitlab-backup create

See the PostgreSQL documentation for more details on what these parameters do.

Migrate to a new server

You can use GitLab backup and restore to migrate your instance to a new server. This section outlines a typical procedure for a GitLab deployment running on a single server. If you're running GitLab Geo, an alternative option is Geo disaster recovery for planned failover.

WARNING: Avoid uncoordinated data processing by both the new and old servers, where multiple servers could connect concurrently and process the same data. For example, when using incoming email, if both GitLab instances are processing email at the same time, then both instances miss some data. This type of problem can occur with other services as well, such as a non-packaged database, a non-packaged Redis instance, or non-packaged Sidekiq.

Prerequisites:

  • Some time before your migration, consider notifying your users of upcoming scheduled maintenance with a broadcast message banner.
  • Ensure your backups are complete and current. Create a complete system-level backup, or take a snapshot of all servers involved in the migration, in case destructive commands (like rm) are run incorrectly.

Prepare the new server

To prepare the new server:

  1. Copy the SSH host keys from the old server to avoid man-in-the-middle attack warnings. See Manually replicate the primary site's SSH host keys for example steps.

  2. Install and configure GitLab except incoming email:

    1. Install GitLab.

    2. Configure by copying /etc/gitlab files from the old server to the new server, and update as necessary. Read the Omnibus configuration backup and restore instructions for more detail.

    3. If applicable, disable incoming email.

    4. Block new CI/CD jobs from starting upon initial startup after the backup and restore. Edit /etc/gitlab/gitlab.rb and set the following:

      nginx['custom_gitlab_server_config'] = "location /api/v4/jobs/request {\n deny all;\n return 503;\n}\n"
      
    5. Reconfigure GitLab:

      sudo gitlab-ctl reconfigure
      
  3. Stop GitLab to avoid any potential unnecessary and unintentional data processing:

    sudo gitlab-ctl stop
    
  4. Configure the new server to allow receiving the Redis database and GitLab backup files:

    sudo rm -f /var/opt/gitlab/redis/dump.rdb
    sudo chown <your-linux-username> /var/opt/gitlab/redis /var/opt/gitlab/backups
    

Prepare and transfer content from the old server

  1. Ensure you have an up-to-date system-level backup or snapshot of the old server.

  2. Enable maintenance mode, if supported by your GitLab edition.

  3. Block new CI/CD jobs from starting:

    1. Edit /etc/gitlab/gitlab.rb, and set the following:

      nginx['custom_gitlab_server_config'] = "location /api/v4/jobs/request {\n deny all;\n return 503;\n}\n"
      
    2. Reconfigure GitLab:

      sudo gitlab-ctl reconfigure
      
  4. Disable periodic background jobs:

    1. On the top bar, select Menu > Admin.
    2. On the left sidebar, select Monitoring > Background Jobs.
    3. Under the Sidekiq dashboard, select Cron tab and then Disable All.
  5. Wait for the currently running CI/CD jobs to finish, or accept that jobs that have not completed may be lost. To view jobs currently running, on the left sidebar, select Overviews > Jobs, and then select Running.

  6. Wait for Sidekiq jobs to finish:

    1. On the left sidebar, select Monitoring > Background Jobs.
    2. Under the Sidekiq dashboard, select Queues and then Live Poll. Wait for Busy and Enqueued to drop to 0. These queues contain work that has been submitted by your users; shutting down before these jobs complete may cause the work to be lost. Make note of the numbers shown in the Sidekiq dashboard for post-migration verification.
  7. Flush the Redis database to disk, and stop GitLab other than the services needed for migration:

    sudo /opt/gitlab/embedded/bin/redis-cli -s /var/opt/gitlab/redis/redis.socket save && sudo gitlab-ctl stop && sudo gitlab-ctl start postgresql && sudo gitlab-ctl start gitaly
    
  8. Create a GitLab backup:

    sudo gitlab-backup create
    
  9. Disable the following GitLab services and prevent unintentional restarts by adding the following to the bottom of /etc/gitlab/gitlab.rb:

    alertmanager['enable'] = false
    gitlab_exporter['enable'] = false
    gitlab_pages['enable'] = false
    gitlab_workhorse['enable'] = false
    grafana['enable'] = false
    logrotate['enable'] = false
    gitlab_rails['incoming_email_enabled'] = false
    nginx['enable'] = false
    node_exporter['enable'] = false
    postgres_exporter['enable'] = false
    postgresql['enable'] = false
    prometheus['enable'] = false
    puma['enable'] = false
    redis['enable'] = false
    redis_exporter['enable'] = false
    registry['enable'] = false
    sidekiq['enable'] = false
    
  10. Reconfigure GitLab:

    sudo gitlab-ctl reconfigure
    
  11. Verify everything is stopped, and confirm no services are running:

    sudo gitlab-ctl status
    
  12. Transfer the Redis database and GitLab backups to the new server:

    sudo scp /var/opt/gitlab/redis/dump.rdb <your-linux-username>@new-server:/var/opt/gitlab/redis
    sudo scp /var/opt/gitlab/backups/your-backup.tar <your-linux-username>@new-server:/var/opt/gitlab/backups
    

Restore data on the new server

  1. Restore appropriate file system permissions:

    sudo chown gitlab-redis /var/opt/gitlab/redis
    sudo chown gitlab-redis:gitlab-redis /var/opt/gitlab/redis/dump.rdb
    sudo chown git:root /var/opt/gitlab/backups
    sudo chown git:git /var/opt/gitlab/backups/your-backup.tar
    
  2. Restore the GitLab backup.

  3. Verify that the Redis database restored correctly:

    1. On the top bar, select Menu > Admin.
    2. On the left sidebar, select Monitoring > Background Jobs.
    3. Under the Sidekiq dashboard, verify that the numbers match with what was shown on the old server.
    4. While still under the Sidekiq dashboard, select Cron and then Enable All to re-enable periodic background jobs.
  4. Test that read-only operations on the GitLab instance work as expected. For example, browse through project repository files, merge requests, and issues.

  5. Disable Maintenance Mode, if previously enabled.

  6. Test that the GitLab instance is working as expected.

  7. If applicable, re-enable incoming email and test it is working as expected.

  8. Update your DNS or load balancer to point at the new server.

  9. Unblock new CI/CD jobs from starting by removing the custom NGINX configuration you added previously:

    # The following line must be removed
    nginx['custom_gitlab_server_config'] = "location /api/v4/jobs/request {\n deny all;\n return 503;\n}\n"
    
  10. Reconfigure GitLab:

    sudo gitlab-ctl reconfigure
    
  11. Remove the scheduled maintenance broadcast message banner.

Additional notes

This documentation is for GitLab Community and Enterprise Edition. We back up GitLab.com and ensure your data is secure. You can't, however, use these methods to export or back up your data yourself from GitLab.com.

Issues are stored in the database, and can't be stored in Git itself.

To migrate your repositories from one server to another with an up-to-date version of GitLab, use the import Rake task to do a mass import of the repository. If you do an import Rake task rather than a backup restore, you get all of your repositories, but no other data.

Troubleshooting

The following are possible problems you might encounter, along with potential solutions.

Restoring database backup using Omnibus packages outputs warnings

If you're using backup restore procedures, you may encounter the following warning messages:

ERROR: must be owner of extension pg_trgm
ERROR: must be owner of extension btree_gist
ERROR: must be owner of extension plpgsql
WARNING:  no privileges could be revoked for "public" (two occurrences)
WARNING:  no privileges were granted for "public" (two occurrences)

Be advised that the backup is successfully restored in spite of these warning messages.

The Rake task runs this as the gitlab user, which doesn't have superuser access to the database. When restore is initiated, it also runs as the gitlab user, but it also tries to alter the objects it doesn't have access to. Those objects have no influence on the database backup or restore, but display a warning message.

For more information, see:

When the secrets file is lost

If you didn't back up the secrets file, you must complete several steps to get GitLab working properly again.

The secrets file is responsible for storing the encryption key for the columns that contain required, sensitive information. If the key is lost, GitLab can't decrypt those columns, preventing access to the following items:

In cases like CI/CD variables and runner authentication, you can experience unexpected behaviors, such as:

  • Stuck jobs.
  • 500 errors.

In this case, you must reset all the tokens for CI/CD variables and runner authentication, which is described in more detail in the following sections. After resetting the tokens, you should be able to visit your project and the jobs begin running again.

Use the information in the following sections at your own risk.

Verify that all values can be decrypted

You can determine if your database contains values that can't be decrypted by using a Rake task.

Take a backup

You must directly modify GitLab data to work around your lost secrets file.

WARNING: Be sure to create a full database backup before attempting any changes.

Disable user two-factor authentication (2FA)

Users with 2FA enabled can't sign in to GitLab. In that case, you must disable 2FA for everyone, after which users must reactivate 2FA.

Reset CI/CD variables

  1. Enter the database console:

    For Omnibus GitLab 14.1 and earlier:

    sudo gitlab-rails dbconsole
    

    For Omnibus GitLab 14.2 and later:

    sudo gitlab-rails dbconsole --database main
    

    For installations from source, GitLab 14.1 and earlier:

    sudo -u git -H bundle exec rails dbconsole -e production
    

    For installations from source, GitLab 14.2 and later:

    sudo -u git -H bundle exec rails dbconsole -e production --database main
    
  2. Examine the ci_group_variables and ci_variables tables:

    SELECT * FROM public."ci_group_variables";
    SELECT * FROM public."ci_variables";
    

    These are the variables that you need to delete.

  3. Drop the table:

    DELETE FROM ci_group_variables;
    DELETE FROM ci_variables;
    
  4. If you know the specific group or project from which you wish to delete variables, you can include a WHERE statement to specify that in your DELETE:

    DELETE FROM ci_group_variables WHERE group_id = <GROUPID>;
    DELETE FROM ci_variables WHERE project_id = <PROJECTID>;
    

You may need to reconfigure or restart GitLab for the changes to take effect.

Reset runner registration tokens

  1. Enter the database console:

    For Omnibus GitLab 14.1 and earlier:

    sudo gitlab-rails dbconsole
    

    For Omnibus GitLab 14.2 and later:

    sudo gitlab-rails dbconsole --database main
    

    For installations from source, GitLab 14.1 and earlier:

    sudo -u git -H bundle exec rails dbconsole -e production
    

    For installations from source, GitLab 14.2 and later:

    sudo -u git -H bundle exec rails dbconsole -e production --database main
    
  2. Clear all tokens for projects, groups, and the entire instance:

    WARNING: The final UPDATE operation stops the runners from being able to pick up new jobs. You must register new runners.

    -- Clear project tokens
    UPDATE projects SET runners_token = null, runners_token_encrypted = null;
    -- Clear group tokens
    UPDATE namespaces SET runners_token = null, runners_token_encrypted = null;
    -- Clear instance tokens
    UPDATE application_settings SET runners_registration_token_encrypted = null;
    -- Clear key used for JWT authentication
    -- This may break the $CI_JWT_TOKEN job variable:
    -- https://gitlab.com/gitlab-org/gitlab/-/issues/325965
    UPDATE application_settings SET encrypted_ci_jwt_signing_key = null;
    -- Clear runner tokens
    UPDATE ci_runners SET token = null, token_encrypted = null;
    

Reset pending pipeline jobs

  1. Enter the database console:

    For Omnibus GitLab 14.1 and earlier:

    sudo gitlab-rails dbconsole
    

    For Omnibus GitLab 14.2 and later:

    sudo gitlab-rails dbconsole --database main
    

    For installations from source, GitLab 14.1 and earlier:

    sudo -u git -H bundle exec rails dbconsole -e production
    

    For installations from source, GitLab 14.2 and later:

    sudo -u git -H bundle exec rails dbconsole -e production --database main
    
  2. Clear all the tokens for pending jobs:

    -- Clear build tokens
    UPDATE ci_builds SET token = null, token_encrypted = null;
    

A similar strategy can be employed for the remaining features. By removing the data that can't be decrypted, GitLab can be returned to operation, and the lost data can be manually replaced.

Fix integrations and webhooks

If you've lost your secrets, the integrations settings pages and webhooks settings pages are probably displaying 500 error messages.

The fix is to truncate the affected tables (those containing encrypted columns). This deletes all your configured integrations, webhooks, and related metadata. You should verify that the secrets are the root cause before deleting any data.

  1. Enter the database console:

    For Omnibus GitLab 14.1 and earlier:

    sudo gitlab-rails dbconsole
    

    For Omnibus GitLab 14.2 and later:

    sudo gitlab-rails dbconsole --database main
    

    For installations from source, GitLab 14.1 and earlier:

    sudo -u git -H bundle exec rails dbconsole -e production
    

    For installations from source, GitLab 14.2 and later:

    sudo -u git -H bundle exec rails dbconsole -e production --database main
    
  2. Truncate the following tables:

    -- truncate web_hooks table
    TRUNCATE integrations, chat_names, issue_tracker_data, jira_tracker_data, slack_integrations, web_hooks, zentao_tracker_data, web_hook_logs;
    

Container Registry push failures after restoring from a backup

If you use the Container Registry, pushes to the registry may fail after restoring your backup on an Omnibus GitLab instance after restoring the registry data.

These failures mention permission issues in the registry logs, similar to:

level=error
msg="response completed with error"
err.code=unknown
err.detail="filesystem: mkdir /var/opt/gitlab/gitlab-rails/shared/registry/docker/registry/v2/repositories/...: permission denied"
err.message="unknown error"

This issue is caused by the restore running as the unprivileged user git, which is unable to assign the correct ownership to the registry files during the restore process (issue #62759).

To get your registry working again:

sudo chown -R registry:registry /var/opt/gitlab/gitlab-rails/shared/registry/docker

If you changed the default file system location for the registry, run chown against your custom location, instead of /var/opt/gitlab/gitlab-rails/shared/registry/docker.

Backup fails to complete with Gzip error

When running the backup, you may receive a Gzip error message:

sudo /opt/gitlab/bin/gitlab-backup create
...
Dumping ...
...
gzip: stdout: Input/output error

Backup failed

If this happens, examine the following:

  • Confirm there is sufficient disk space for the Gzip operation.
  • If NFS is being used, check if the mount option timeout is set. The default is 600, and changing this to smaller values results in this error.