Github user sudheeshkatkam commented on a diff in the pull request:
https://github.com/apache/drill/pull/921#discussion_r141981097
--- Diff:
exec/java-exec/src/main/java/org/apache/drill/exec/coord/zk/ZKClusterCoordinator.java
---
@@ -200,11 +206,47 @@ public void unregister(RegistrationHandle handle) {
}
}
+ /**
+ * Update drillbit endpoint state. Drillbit advertises its
+ * state in Zookeeper when a shutdown request of drillbit is
+ * triggered. State information is used during planning and
+ * initial client connection phases.
+ */
+ public RegistrationHandle update(RegistrationHandle handle, State state)
{
+ ZKRegistrationHandle h = (ZKRegistrationHandle) handle;
+ try {
+ endpoint = h.endpoint.toBuilder().setState(state).build();
+ ServiceInstance<DrillbitEndpoint> serviceInstance =
ServiceInstance.<DrillbitEndpoint>builder().name(serviceName).id(h.id).payload(endpoint).build();
+ discovery.updateService(serviceInstance);
+ } catch (Exception e) {
+ e.printStackTrace();
+ }
+ return handle;
+ }
+
@Override
public Collection<DrillbitEndpoint> getAvailableEndpoints() {
return this.endpoints;
}
+ /*
+ * Get a collection of ONLINE Drillbit endpoints by excluding the
drillbits
+ * that are in QUIESCENT state (drillbits shutting down). Primarily used
by the planner
+ * to plan queries only on ONLINE drillbits and used by the client
during initial connection
+ * phase to connect to a drillbit (foreman)
+ * @return A collection of ONLINE endpoints
+ */
+ @Override
+ public Collection<DrillbitEndpoint> getOnlineEndPoints() {
+ Collection<DrillbitEndpoint> runningEndPoints = new ArrayList<>();
+ for (DrillbitEndpoint endpoint: endpoints){
+ if(endpoint.getState().equals(State.ONLINE)) {
--- End diff --
(more relevant in this case)
This check should be:
if(!endpoint.hasState() || endpoint.getState().equals(State.ONLINE)) {
to not break backward compatibility (new client and old cluster of bits).
So the assumption is that old server is ONLINE.
---
