In the backup/restore RFC there is a brief discussion on the alternative approach of using RocksDB’s SSTables for ingestion, which may render better performance as compared to the current WriteBatch data format.
What’s the main concerns that stop you from using the alternative approach? Or is it that you guys just haven’t got to it yet?
Hi zzkk,
We just haven’t gotten to it yet. Preliminary tests have proven the performance benefit is significant (and with much less disk usage). The distributed consensus part is tricky, so it’s a decent sized project. Currently slated for the 1.1 release, but I’m not guaranteeing that yet.
I’d love to hear your use case if you’re willing to share it.
Dan