gr-replication.txt

GRAPHD REPLICATION
==================

1. Creating a Replica Database

To create a replica, one must specify the location of the
database to be replicated in the replica section of a config
file, for example:

   ...
   replica
   {
      host myhost
      port 8100
   }
   ....

When started with this config file, graphd will try and establish
a replication connection with the graphd listening on port 8100
at myhost.

The replication conection will first restore the replica from
the master.  Once the restore has caught up, new primitives
will be sent to the replica as they are committed on the master.
Primitives will be sent in transactional chunks corresponding
to the original writes.

If you have a copy of the master's database files, it is faster
to use these to "seed" the replica rather than to create an
empty replica which will restore itself from the master.

You may use

   gstatus -r

to check the replication status of a graph.  At the protocol
level, "status (rep)", provides the same information.  For
example:

status (rep)
ok ((("127.0.0.1:1234" "tcp://localhost:1234") ("" "tcp://localhost:8100") ()))

The result indicates that we are connected to the
replica server at 127.0.0.1:1234, which we found
at the address tcp://localhost:1234.  We have no
connection to the write master, but the address to
use when we need one is tcp://localhost:8100.

The last empty list indicates that we are not serving
as master to any replicas.

2. Writing to a Replica

When a replica connects to a master, the master returns
the address of the "write master," that is, the graphd
to forward write commands to.  Subsequent write commands
to the replica are transparently forwarded to the write master.

3. Access Modes

Replication is an access mode.  If you set the access mode
to "replica" graphd attempts to connect to the master
specified in the configuration file.  If it was already
in replica mode, it will drop the current connection
to both the write master and the replica server and
attempt to reconnect.

A common transition would be to move graphd from mode
"replica" to mode "read-only."

4. Failures

A graphd in replica mode will attempt to connect to the
specified master every ten seconds until it succeeds.