We’re overhauling Dgraph’s docs to make them clearer and more approachable. If you notice any issues during this transition or have suggestions, please let us know.
Dgraph cluster nodes use a range of ports to communicate over gRPC and HTTP. Choose these ports carefully based on your topology and mode of deployment, as this impacts the access security rules or firewall configurations required for each port.

​
Types of ports

Dgraph Alpha and Dgraph Zero nodes use a variety of gRPC and HTTP ports, as follows:
  • gRPC-internal-private: Used between the cluster nodes for internal communication and message exchange. Communication using these ports is TLS-encrypted.
  • gRPC-external-private: Used by Dgraph Live Loader and Dgraph Bulk Loader to access APIs over gRPC.
  • gRPC-external-public: Used by Dgraph clients to access APIs in a session that can persist after a query.
  • HTTP-external-private: Used for monitoring and administrative tasks.
  • HTTP-external-public: Used by clients to access APIs over HTTP.

​
Default ports used by different nodes

Dgraph Node TypegRPC-internal-privategRPC-external-privategRPC-external-publicHTTP-external-privateHTTP-external-public
zero508015080160802
alpha708090808080
ratel8000
1: Dgraph Zero uses port 5080 for internal communication within the cluster, and to support the data import tools. 2: Dgraph Zero uses port 6080 for administrative operations. Dgraph clients can’t access this port. Users must modify security rules or open firewall ports depending upon their underlying network to allow communication between cluster nodes, between the Dgraph instances, and between Dgraph clients. In general, you should configure the gRPC and HTTP external-public ports for open access by Dgraph clients, and configure the gRPC-internal ports for open access by the cluster nodes. Ratel UI accesses Dgraph Alpha on the HTTP-external-public port (which defaults to localhost:8080) and can be configured to talk to a remote Dgraph cluster. This way you can run Ratel on your local machine and point to a remote cluster. But, if you are deploying Ratel along with Dgraph cluster, then you may have to expose port 8000 to the public. Port Offset To make it easier for users to set up a cluster, Dgraph has default values for the ports used by Dgraph nodes. To support multiple nodes running on a single machine or VM, you can set a node to use different ports using an offset (using the command option --port_offset). This command increments the actual ports used by the node by the offset value provided. You can also use port offsets when starting multiple Dgraph Zero nodes in a development environment. For example, when a user runs Dgraph Alpha with the --port_offset 2 setting, then the Alpha node binds to port 7082 (gRPC-internal-private), 8082 (HTTP-external-public) and 9082 (gRPC-external-public), respectively. Ratel UI by default listens on port 8000. You can use the -port flag to configure it to listen on any other port.

​
High availability cluster configuration

In a high availability (HA) cluster configuration, you should run three or five replicas for the Zero node, and three or five replicas for the Alpha node. A Dgraph cluster is divided into Raft groups, where Dgraph Zero is group 0 and each shard of Dgraph Alpha is a subsequent numbered group (group 1, group 2, etc.). The number of replicas in each Raft group must be an odd number for the group to have consensus, which exists when the majority of nodes in a group are available.
If the number of replicas in a Raft group is 2N + 1, up to N nodes can go offline without any impact on reads or writes. So, if there are five replicas, three must be online to avoid an impact to reads or writes.

​
Dgraph Zero

Run three Dgraph Zero instances, assigning a unique integer ID to each using the --raft superflag’s idx option, and passing the address of any healthy Dgraph Zero instance using the --peer flag. To run three replicas for the Alpha nodes, set --replicas=3. Each time a new Alpha node is added, the Zero node checks the existing groups and assign them as appropriate.

​
Dgraph Alpha

You can run as many Dgraph Alpha nodes as you want. You can manually set the --raft superflag’s idx option, or you can leave that flag empty, and the Zero node assigns an id to the Alpha node. This id persists in the write-ahead log, so be careful not to delete it. The new Alpha nodes automatically detect each other by communicating with Dgraph Zero and establish connections to each other. If you don’t have a proxy or load balancer for the Zero nodes, you can provide a list of Zero node addresses for Alpha nodes to use at startup with the --zero flag. The Alpha node tries to connect to one of the Zero nodes starting from the first Zero node address in the list. For example: --zero=zero1,zero2,zero3 where zero1 is the host:port of a Zero instance. Typically, a Zero node would first attempt to replicate a group, by assigning a new Alpha node to run the same group previously assigned to another. After the group has been replicated per the --replicas flag, Dgraph Zero creates a new group. Over time, the data is evenly split across all of the groups. So, it’s important to ensure that the number of Alpha nodes is a multiple of the replication setting. For example, if you set --replicas=3 in for a Zero node, and then run three Alpha nodes for no sharding, but 3x replication. Or, if you run six Alpha nodes, sharding the data into two groups, with 3x replication.