The ClusterResourceManager PolicyEngine

The primary goal of the PolicyEngine is to encapsulate the potential complex computation of the TransitionGraph[1] from the CurrentClusterState[2].

The PolicyEngine does not do any network IO, it only runs this algorithm.

Main success scenario

1. The ClusterResourceManager[3] starts the PolicyEngine on demand, or when becomes DesignatedCoordinator[4].

2. The PolicyEngine starts up and signs in with the ClusterResourceManager[3]. This stage verifies the version numbers et cetera.

3. The PolicyEngine waits for orders from the ClusterResourceManager[3].

4. An order is received:

   1. Either a ClusterInformationBase[5] dump requesting a computation of a transition graph.

   2. Or a request to shutdown.
5. The requested operation is performed.

   1. PolicyEngineComputation[6]

6. The answer is sent back to the ClusterResourceManager[3].

   1. In case of the shutdown request, this message signs us out from the ClusterResourceManager[3].

   2. Otherwise, we sent back the TransitionGraph[1].

7. If applicable, we go back to waiting for orders, unless this was the last iteration.
8. We shutdown cleanly with exit code 100.

Extensions

• To ease testing, the PolicyEngine shall have a commandline mode which only reads the ClusterInformationBase[5] dump from stdin and outputs the TransitionGraph[1] on stdout.

Open issues

AndrewBeekhof[7]: It would be nice if we could give you a ClusterInformationBase[5] dump (possibly minus StateInformation[8]) and the PolicyEngine gave a yes or no answer as to whether it was solvable/valid. We could then decide if an AdministrationUpdate[9] was ok to apply to the ClusterInformationBase[5].

---

This information provided courtesy of the Linux-HA project at http://linux-ha.org/