If BLR is replicating from one node in a Galera cluster (e.g. GTID = 0-1-X), and updates are made also via other nodes than that node (i.e. the GTIDs will be different, e.g. GTID = 0-2-X), all events will end up in the same file (as they should) until a rotation.

In the case above, the full path of the file will e.g. be .../0/1/galera000-bin.000001 and it may contain events like 0-1-0, 0-1-1, 0-1-2, 0-2-3, 0-2-4 etc.

However, now when a slave looks for event 0-2-4, the BLR will assume to find it in file 0/2/galera000-bin.000001, which does not exist and the operation will thus fail.

BLR should, one way or another, store the full path information in its internal database.