Open edsko opened 4 years ago
In particular the chain sync would be in terms of (offset, length) pairs into files. To manage the lifetimes of files in the volatile DB reliably we might also need to take advantage of the ability of unix sockets to send open FDs of files over the socket. On unix files are only really deleted when the last reference to them is closed. Deleting isn't blocked by open FDs.
@coot This also seems like a proper Networking Team Issue. So I've removed the Consensus label.
What do you think?
(Obviously this idea would require some cooperation from the Consensus and Node teams to actually implement it.)
cc @karknu @dcoutts
When serving many clients, it might be beneficial to do so from several OS processes, rather than a single large one. We could do this by having a single main node, and then several proxy nodes that follow the main node using a (local) chain sync protocol. ideally, the proxies and the main node should then share (almost?) all of the on-disk storage, just keeping their own in-memory data structures.