Election Protocols

[Mike Brown_3]

Can anyone tell me;
a) how the election protocol within MC/SG actually works to elect a cluster co-ordinator,
OR
b) if this info is not disclosed to the average punter like me because we don't need to know.
Thanks.

[Carsten Krege · Accepted solution]

The election protocol is both: pretty complex and company confidential. ;) We cannot give away all the details, but it involves the following steps:

- each node starts sending UDP messages to find a coordinator
- one node becomes "candidate" and answers all incoming UDP messages and initiates heartbeat connections (hacl-hb) to the other nodes
- the other nodes become "voters", also open a heartbeat connection to the "candidate" node and send their vote
- if the candidate receives enough votes it becomes a "locker" node and "general service" (hacl-gs) connections between all nodes are setup
- When all of these messages are received (and all general service connections are ready), the candidate/locker/coordinator node enters a new state and sends a "go" message to all other nodes once they are ready
- the cluster formed

Crucial is the exact timing used for each of the steps. During election many of the various SG subsystems (package manager, network manager, SLVM etc) are reconfigured.

As I said the procedure is very complex. The algorithm has to ensure that a "split brain" (see manual) situation cannot occur at any time.

To administer a SG cluster it is not necessary to understand all the details of the election protocol though.

Carsten

[Mike Brown_3]

Carsten - thanks for your reply. That's perfect.
Mike.