Replication

    ArangoDB offers synchronous and asynchronous replication.

    Synchronous replication is used between the DB-Servers of an ArangoDB Cluster.

    Asynchronous replication is used:

    • between the master and the slave of an ArangoDB Master/Slave setup
    • between multiple ArangoDB (inside the same Data Center replication is synchronous)

    Synchronous replication only works within an ArangoDB Cluster and is typically used for mission critical data which must be accessible at all times. Synchronous replication generally stores a copy of a shard’s data on another DB-Server and keeps it in sync. Essentially, when storing data after enabling synchronous replication the Cluster will wait for all replicas to write all the data before greenlighting the write operation to the client. This will naturally increase the latency a bit, since one more network hop is needed for each write. However, it will enable the cluster to immediately fail over to a replica whenever an outage has been detected, without losing any committed data, and mostly without even signaling an error condition to the client.

    Synchronous replication is organized such that every shard has a leader and followers, where r denoted the replication factor. The number of followers can be controlled using the parameter whenever you create a collection, the replicationFactor parameter is the total number of copies being kept, that is, it is one plus the number of followers.

    Asynchronous replication

    In ArangoDB any write operation is logged in the write-ahead log.

    When using asynchronous replication slaves (or followers)
    connect to a master (or leader) and apply locally all the events from the master log in the same order. As a result the slaves (followers) will have the same state of data as the master (leader).

    Slaves (followers) are only eventually consistent with the master (leader).

    Transactions are honored in replication, i.e. transactional write operations will become visible on slaves atomically.

    As all write operations will be logged to a master database’s write-ahead log, the replication in ArangoDB currently cannot be used for write-scaling. The main purposes of the replication in current ArangoDB are to provide read-scalability and “hot backups” of specific databases.

    In an asynchronous replication scenario slaves will pull changes from the master. Slaves need to know to which master they should connect to, but a master is not aware of the slaves that replicate from it. When the network connection between the master and a slave goes down, write operations on the master can continue normally. When the network is up again, slaves can reconnect to the master and transfer the remaining changes. This will happen automatically provided slaves are configured appropriately.

    Before 3.3.0 asynchronous replication was per database. Starting with 3.3.0 it is possible to setup global replication.

    As described above, write operations are applied first in the master, and then applied in the slaves.

    For example, let’s assume a write operation is executed in the master at point in time t0. To make a slave apply the same operation, it must first fetch the write operation’s data from master’s write-ahead log, then parse it and apply it locally. This will happen at some point in time after t0, let’s say t1.

    The difference between t1 and t0 is called the replication lag, and it is unavoidable in asynchronous replication. The amount of replication lag depends on many factors, a few of which are:

    • the network capacity between the slaves and the master
    • the frequency in which slaves poll the master for updates

    Replication overhead

    As the master servers are logging any write operation in the write-ahead-log anyway replication doesn’t cause any extra overhead on the master. However it will of course cause some overhead for the master to serve incoming read requests of the slaves. Returning the requested data is however a trivial task for the master and should not result in a notable performance degradation in production.