Rate limits

note
For GitLab.com, see GitLab.com-specific rate limits.

Rate limiting is a common technique used to improve the security and durability of a web application.

For example, a simple script can make thousands of web requests per second. The requests could be:

  • Malicious.
  • Apathetic.
  • Just a bug.

Your application and infrastructure may not be able to cope with the load. For more details, see Denial-of-service attack. Most cases can be mitigated by limiting the rate of requests from a single IP address.

Most brute-force attacks are similarly mitigated by a rate limit.

Configurable limits

You can set these rate limits in the Admin Area of your instance:

You can set these rate limits using the Rails console:

Failed authentication ban for Git and container registry

GitLab returns HTTP status code 403 for 1 hour, if 30 failed authentication requests were received in a 3-minute period from a single IP address. This applies only to combined:

  • Git requests.
  • Container registry (/jwt/auth) requests.

This limit:

  • Is reset by requests that authenticate successfully. For example, 29 failed authentication requests followed by 1 successful request, followed by 29 more failed authentication requests would not trigger a ban.
  • Does not apply to JWT requests authenticated by gitlab-ci-token.
  • Is disabled by default.

No response headers are provided.

For configuration information, see Omnibus GitLab configuration options.

Non-configurable limits

Git operations using SSH

Introduced in GitLab 14.7 with a flag named rate_limit_gitlab_shell. Available by default without a feature flag from 15.8.

GitLab applies rate limits to Git operations that use SSH by user account and project. When the rate limit is exceeded, GitLab rejects further connection requests from that user for the project.

The rate limit applies at the Git command (plumbing) level. Each command has a rate limit of 600 per minute. For example:

  • git push has a rate limit of 600 per minute.
  • git pull has its own rate limit of 600 per minute.

Because the same commands are shared by git-upload-pack, git pull, and git clone, they share a rate limit.

The requests per minute threshold for this rate limit is not configurable. Self-managed customers can disable this rate limit.

Repository archives

Introduced in GitLab 12.9.

A rate limit for downloading repository archives is available. The limit applies to the project and to the user initiating the download either through the UI or the API.

The rate limit is 5 requests per minute per user.

Webhook Testing

Introduced in GitLab 13.4.

There is a rate limit for testing webhooks, which prevents abuse of the webhook functionality.

The rate limit is 5 requests per minute per user.

Users sign up

Introduced in GitLab 14.7.

There is a rate limit per IP address on the /users/sign_up endpoint. This is to mitigate attempts to misuse the endpoint. For example, to mass discover usernames or email addresses in use.

The rate limit is 20 calls per minute per IP address.

Update username

Introduced in GitLab 14.7.

There is a rate limit on how frequently a username can be changed. This is enforced to mitigate misuse of the feature. For example, to mass discover which usernames are in use.

The rate limit is 10 calls per minute per authenticated user.

Username exists

Introduced in GitLab 14.7.

There is a rate limit for the internal endpoint /users/:username/exists, used upon sign up to check if a chosen username has already been taken. This is to mitigate the risk of misuses, such as mass discovery of usernames in use.

The rate limit is 20 calls per minute per IP address.

Project Jobs API endpoint

Introduced in GitLab 15.7 with a flag named ci_enforce_rate_limits_jobs_api. Disabled by default.

On self-managed GitLab, by default this feature is not available. To make it available, ask an administrator to enable the feature flag named ci_enforce_rate_limits_jobs_api. The feature is not ready for production use.

There is a rate limit for the endpoint project/:id/jobs, which is enforced to reduce timeouts when retrieving jobs.

The rate limit is 600 calls per minute per authenticated user.

Troubleshooting

Rack Attack is denylisting the load balancer

Rack Attack may block your load balancer if all traffic appears to come from the load balancer. In that case, you must:

  1. Configure nginx[real_ip_trusted_addresses]. This keeps users’ IPs from being listed as the load balancer IPs.
  2. Allowlist the load balancer’s IP addresses.

  3. Reconfigure GitLab: 

    sudo gitlab-ctl reconfigure

Remove blocked IPs from Rack Attack with Redis

To remove a blocked IP:

  1. Find the IPs that have been blocked in the production log: 

    grep "Rack_Attack" /var/log/gitlab/gitlab-rails/auth.log

  2. Since the denylist is stored in Redis, you must open up redis-cli: 

    /opt/gitlab/embedded/bin/redis-cli -s /var/opt/gitlab/redis/redis.socket

  3. You can remove the block using the following syntax, replacing <ip> with the actual IP that is denylisted: 

    del cache:gitlab:rack::attack:allow2ban:ban:<ip>

  4. Confirm that the key with the IP no longer shows up: 

    keys *rack::attack*

By default, the keys command is disabled.

  1. Optionally, add the IP to the allowlist to prevent it being denylisted again.