Читать книгу From Traditional Fault Tolerance to Blockchain - Wenbing Zhao - Страница 51

The finite state machine diagram for the Chandy and Lamport distributed snapshot protocol is given in Figure 2.7. A process will be in the Normal state between two rounds of global checkpointing, and in the Checkpointing state during a global checkpointing round. A process may encounter a number of events:

 ◾ The global checkpointing can be initiated by any of the processes in the distributed system. Once a process decides to initiate a global checkpointing round, it takes a local checkpoint and sends a Marker message to each of its outgoing channels. The state of the process changes from Normal to Checkpointing as a result.

 ◾ A process undergoes the same state transition (from Normal to Checkpointing) and take the same actions upon receiving the Marker message for the first time, except that it logs the Maker in a data structure referred to as the Marker Certificate in the finite state machine diagram. The Marker Certificate data structure keeps track of which incoming channel has received a Marker and whether or not all incoming channels have received the Marker. The Marker Certificate is called complete when every incoming channel has received a Marker.

 ◾ When a process receives the Marker message from a channel when it is in the Checkpointing state, it adds the Marker message to the Marker Certificate and checks whether or not the Marker Certificate is complete. If the Marker Certificate is now complete, the process transits to the Normal state (and possibly reports the completion of the global checkpointing to some predefined server). Otherwise, the process will remain in the Checkpointing state.

 ◾ In either the Normal or Checkpointing state, the process may receive a regular message. The regular message is always executed immediately. This is drastically different from the Tamir and Sequin global checkpointing protocol. The regular message will be appended to the channel state from which it is received only when the process is in the Checkpointing state and it has not received the Marker message in this channel.