Replication
Managing Replication
Dragonfly supports a primary-replica high-availability model, similarly to Redis's replication. When using replication, Dragonfly creates exact copies of the primary instance. Once configured properly, replicas reconnect to the primary instance any time their connections break and will always aim to remain an exact copy of the primary.
The Dragonfly replication management API is compatible with the Redis API and consists of two main commands:
ROLE
and REPLICAOF
.
If you're not sure whether the Dragonfly instance you're currently connected to is a primary instance or a replica, you can check by running the ROLE
command:
dragonfly$> ROLE
1) "master"
2) 1) 1) "172.19.0.4"
2) "6379"
3) "stable_sync"
This command will return either master
or replica
with additional information, depending on the role of the instance.
Redis-Dragonfly Replication
Dragonfly supports Redis-Dragonfly replication to allow a convenient migration of Redis workloads to Dragonfly. We currently support the data structures and replication protocol of Redis OSS up to version 6.2. The instructions below apply to this type of replication as well, with the only difference being that the primary instance is a running Redis server.
Dragonfly-Dragonfly Replication
This replication process internally is vastly different from the original Redis replication algorithm, but from the outside, the API is kept the same to make it compatible with the current ecosystem.
To designate a Dragonfly instance as a replica of another instance on the fly, run the REPLICAOF
command.
This command takes the intended primary instance's hostname or IP address and port as arguments:
dragonfly$> REPLICAOF hostname_or_IP port
If the server is already a replica of another primary, it will stop replicating the old primary and immediately start synchronizing with the new one. It will also discard the entire old dataset.
To promote a replica back to being a primary, run the following REPLICAOF
command:
dragonfly$> REPLICAOF NO ONE
This will stop the instance from replicating the primary instance but will not discard the dataset it has already replicated.
This syntax is useful in cases where the original primary fails.
After running REPLICAOF NO ONE
on a replica of the failed primary, the former replica can be used as the new primary and have its own replicas as a failsafe.
Secure Replication with TLS
Secure Replication Prerequisites
Dragonfly supports replication over TLS. To make it work, you need:
- To configure your Dragonfly instance to use TLS. That includes getting or generating a server certificate.
- To choose a method of authentication between the replica and the primary instance. This can be either with a password or with client certificates.
Also note that we only support .pem
files.
Any other format won't work with Dragonfly.
You can use OpenSSL
to generate the required certificates.
For example, the process of generating a server certificate and signing it with a self-managed CA is as follows:
Generate Dragonfly's private key and certificate signing request (CSR):
openssl req -newkey rsa:4096 -nodes -keyout server-key.pem -out server-req.pem
This will prompt you to add additional details for the subject which contains information related to the certificate. The attributes of the subject are key-value pairs as follows:
CN
: CommonNameOU
: OrganizationalUnitO
: OrganizationL
: LocalityS
: StateOrProvinceNameC
: CountryName
The -nodes
argument means no DES
, that is, do not encrypt the private key.
For production use cases, you should always consider encrypting private keys.
Finally, you should use the CA's private key to sign Dragonfly's CSR and get back the signed certificate:
openssl x509 -req -in server-req.pem -days 365 \
-CA ca-cert.pem -CAkey ca-key.pem \
-CAcreateserial -out server-cert.pem
The ca-cert.pem
and ca-key.pem
can be generated like in the first step described above.
The same process applies for generating client keys.
When generating private keys, always be mindful about storing and provisioning them securely!
Secure Replication Setup
To enable TLS for Dragonfly, you should supply the following arguments:
dragonfly --tls --tls_key_file=server-key.pem \
--tls_cert_file=server-cert.pem \
--tls_ca_cert_file=ca-cert.pem
Start the secondary Dragonfly instance on a different port (i.e., 6380
) with the following arguments:
dragonfly --tls --tls_key_file=replica-client-key.pem \
--tls_cert_file=replica-client-cert.pem \
--tls_ca_cert_file=ca-cert.pem \
--tls_replication=true --port 6380
Finally, connect to the secondary Dragonfly instance and issue the REPLICAOF
command:
redis-cli --tls --key ./client-key.pem --cert ./client-cert.pem --cacert ./ca-cert.pem -p 6380
dragonfly$> REPLICAOF PRIMARY_HOST 6379
Now, the replication works over TLS and is secure.
This example assumes that ca-cert.pem
is a root certificate used to authenticate both the server and the client.
For example, if the server is using a certificate signed by a public CA, this flag should not be passed as an argument to the replica.
Non-TLS Connections over Admin Ports
In some scenarios, both the primary instance and its replica may be located on the same private network. Under these circumstances, you might prefer to use TLS for connections that are external to the private network while allowing plaintext communication within the secure confines of the private network. For situations like these, it's possible to specifically disable TLS for replication on the admin port, as described below.
Start the primary instance with:
dragonfly --tls --tls_key_file=server-key.pem \
--tls_cert_file=server-cert.pem \
--tls_ca_cert_file=ca-cert.pem \
--port 6379 -admin_port=6380 --no_tls_on_admin_port=true
Start the secondary Dragonfly instance with:
dragonfly --tls --tls_key_file=replica-client-key.pem \
--tls_cert_file=replica-client-cert.pem \
--tls_ca_cert_file=ca-cert.pem \
--port 6381 -admin_port=6382
Then, connect to the secondary Dragonfly instance and issue the REPLICAOF
command via the admin port:
redis-cli -p 6382
dragonfly$> REPLICAOF PRIMARY_HOST 6380
From now on, the replica does not communicate with the primary instance over TLS.
But incoming connections on ports 6379
and 6381
still require TLS and are considered secure.
Monitoring Replication Lag
Dragonfly defines the replication lag as the maximal amount of unacknowledged database among all shards.
This metric is calculated by the primary instance and is visible both as the dragonfly_connected_replica_lag_records
field in the Prometheus metrics
as well as in the INFO REPLICATION
command output as the lag
field.
# On the primary instance:
dragonfly$> INFO REPLICATION
# Replication
role:master
connected_slaves:1
slave0:ip=127.0.0.1,port=6380,state=stable_sync,lag=0 # <- The replication lag is reported here.
master_replid:3ad69228e6973c1fc24805efb46d3f6eeeee4fde