dimitarvdimitrov / sporkfs Goto Github PK

The VFS receives a context on all calls. We ignore that context's deadline. spork.Spork should take this context into account.

Currently adding new members to the cluster is not supported. It should be. Raft supports this, so we should be able to do it as well.

Proposal

./sporkfs node add -ip 192.168.0.4 -name node4

Right now we rely on restoring state by replaying the logs. This is very inefficient.

Problem

There is a possible bug where the node that proposed a change to a file goes down while a peer is fetching the file after the change has been committed to the local log. In that case the peer will have the command committed, without having the file. This is most concerning for other owner peers of the file. If the proposing node does not rejoin the cluster with the data committed, this will lead to data loss.

Possible solutions

• make file fetching synchronous - leader and owning peers fetch the file sync before confirming the commit
  • this could slow down the progress of the whole system on large file commits

Implement this https://github.com/grpc/grpc/blob/master/doc/health-checking.md and use that in tests instead of relying on rate of logs to infer a node is ready.