Arangodump has the following limitations:
- In a cluster, arangodump does not guarantee to dump a consistent snapshot if write operations happen while the dump is in progress (see Hot Backups for an alternative). It is therefore recommended not to perform any data-modification operations on the cluster while arangodump is running. This is in contrast to what happens on a single instance, a master/slave, or active failover setup, where even if write operations are ongoing, the created dump is consistent, as a snapshot is taken when the dump starts.
- If the MMFiles engine is in use, on a single instance, a master/slave, or active failover setup, even if the write operations are suspended, it is not guaranteed that the dump includes all the data that has been previously written as arangodump will only dump the data included in the datafiles but not the data that has not been transferred from the WAL to the datafiles. A WAL flush can be forced as documented in the WAL flush section.