checkpoint机制是不是与多个group不好同时采用?因为多个group的话,每个group都有自己的镜像数据,新机器加入的话,难道每个group都要从旧机器接收镜像追赶进度,而每个group学完都会退出进程
jettchen00 opened this issue · 2 comments
jettchen00 commented
checkpoint机制是不是与多个group不好同时采用?因为多个group的话,每个group都有自己的镜像数据,新机器加入的话,难道每个group都要从旧机器接收镜像追赶进度,而每个group学完都会退出进程
lynncui00 commented
For better generality to phxpaxos,we need to consider a situation that different group with different state machine, so every group's logic must be independent. (include checkpoint logic)
For your case, there's a way to fix it, not very difficult.
- Because each group is in the same process, each group can communicate easily.
- Write some logic in
StateMachine::LoadCheckpointState
, to let the group that finishes first wait for other groups. For example, sleep.
jettchen00 commented
thanks