We shall add possibility for replication to read the logs directly from file system instead of asking them from the Master. The main reason for this is automatic log forward recovery from backup+logs, on server level instead of using external tool, which is hard to keep in sync featurewise. The next reason is - this would allow to setup "shared disk" based replication, which has its own benefits: -Distributed file system takes care of log files distribution over network caching etc automatically. This removes master as bottleneck and allows to use ore simple setup with large amount of slave -In case of master crash recovery is simplified as all slaves are still in sync with each other. -Fallback implementation is more trivial - you can just broughtup new master which will continue writing logs to the same location without hassle with slaves positions resync.