Configuring a repository cache

You can configure a repository cache for GitHub Enterprise Server by creating a new instance, connecting the repository cache to your primary instance, and configuring replication of repository networks to the repository cache.

In this article

About configuration for repository caching

You can configure repository caching by creating a special type of replica called a repository cache. Then, you can set data location policies that govern which repository networks are replicated to the repository cache.

Repository caching is not supported with clustering.

DNS for repository caches

The primary instance and repository cache should have different DNS names. For example, if your primary instance is at github.example.com, you might decide to name a cache europe-ci.github.example.com or github.asia.example.com.

To have your CI machines fetch from the repository cache instead of the primary instance, you can use Git's url.<base>.insteadOf configuration setting. For more information, see git-config in the Git documentation.

For example, the global .gitconfig for the CI machine would include these lines.

[url "https://europe-ci.github.example.com/"]
	insteadOf = https://github.example.com/

Then, when told to fetch https://github.example.com/myorg/myrepo, Git will instead fetch from https://europe-ci.github.example.com/myorg/myrepo.

Configuring a repository cache

  1. Set up a new GitHub Enterprise Server instance on your desired platform. This instance will be your repository cache. For more information, see Setting up a GitHub Enterprise Server instance.

  2. Set an admin password that matches the password on the primary appliance and continue.

  3. Click Create a replica node.

  4. Under "Add new SSH key", type your SSH key.

  5. Click Add key.

  6. Connect to the repository cache's IP address using SSH.

    ssh -p 122 admin@REPLICA-IP

  7. To generate a key pair for replication, use the ghe-repl-setup command with the primary appliance's IP address and copy the public key that it returns.

    ghe-repl-setup PRIMARY_IP

  8. To add the public key to the list of authorized keys on the primary appliance, browse to https://PRIMARY-HOSTNAME/setup/settings and add the key you copied from the replica to the list.

  9. To verify the connection to the primary and enable replica mode for the repository cache, run ghe-repl-setup again.

    • If the repository cache is your only additional node, no arguments are required.

      ghe-repl-setup PRIMARY-IP

    • If you're configuring a repository cache in addition to one or more existing replicas, use the -a or --add argument.

      ghe-repl-setup -a PRIMARY-IP

  10. To configure the repository cache, use the ghe-repl-node command and include the necessary parameters.

    • Set a cache-location for the repository cache, replacing CACHE-LOCATION with an alphanumeric identifier, such as the region where the cache is deployed. The CACHE-LOCATION value must not be any of the subdomains reserved for use with subdomain isolation, such as assets or media. For a list of reserved names, see Enabling subdomain isolation.

    • Set a cache-domain for the repository cache, replacing EXTERNAL-CACHE-DOMAIN with the hostname Git clients will use to access the repository cache. If you do not specify a cache-domain, GitHub Enterprise Server will prepend the CACHE-LOCATION value as a subdomain to the hostname configured for your instance. For more information, see Configuring the hostname for your instance.

    • If you haven't already, set the datacenter name on the primary and any replica appliances, replacing DC-NAME with a datacenter name.

      ghe-repl-node --datacenter DC-NAME

    • New caches will attempt to seed from another cache in the same datacenter. Set a datacenter for the repository cache, replacing REPLICA-DC-NAME with the name of the datacenter where you're deploying the node.

    ghe-repl-node --cache CACHE-LOCATION --cache-domain EXTERNAL-CACHE-DOMAIN --datacenter REPLICA-DC-NAME

  11. To start replication of the datastores, use the ghe-repl-start command.

    ghe-repl-start

    Warning

    ghe-repl-start causes a brief outage on the primary server, during which users may see internal server errors. To provide a friendlier message, run ghe-maintenance -s on the primary node before running ghe-repl-start on the replica node to put the appliance in maintenance mode. Once replication starts, disable maintenance mode with ghe-maintenance -u. Git replication will not progress while the primary node is in maintenance mode.

  12. To verify the status of each datastore's replication channel, use the ghe-repl-status command.

    ghe-repl-status

  13. To enable replication of repository networks to the repository cache, set a data location policy. For more information, see Data location policies.

Data location policies

You can control data locality by configuring data location policies for your repositories with the spokesctl cache-policy command. Data location policies determine which repository networks are replicated on which repository caches. By default, no repository networks will be replicated on any repository caches until a data location policy is configured.

Data location policies affect only Git content. Content in the database, such as issues and pull request comments, will be replicated to all nodes regardless of policy.

Note

Data location policies are not the same as access control. You must use repository roles to control which users may access a repository. For more information about repository roles, see Repository roles for an organization.

You can configure a policy to replicate all networks with the --default flag. For example, this command will create a policy to replicate a single copy of every repository network to the set of repository caches whose cache_location is "kansas".

ghe-spokesctl cache-policy set --default 1 kansas

To configure replication for a repository network, specify the repository that is the root of the network. A repository network includes a repository and all of the repository's forks. You cannot replicate part of a network without replicating the whole network.

ghe-spokesctl cache-policy set <owner/repository> 1 kansas

You can override a policy that replicates all networks and exclude specific networks by specifying a replica count of zero for the network. For example, this command specifies that any repository cache in location "kansas" cannot contain any copies of that network.

ghe-spokesctl cache-policy set <owner/repository> 0 kansas

Replica counts greater than one in a given cache location are not supported.