Sean Wilcox wrote: > Same as if ServiceA -- Depends On --> ServiceB is delivered into > /var/svc/manifest/system. > > By directory order, serviceA will import before serviceB but the graph > engine will sort out the > dependency. Same here when serviceA resides in /etc and serviceB in > /var serviceA will import > before serviceB but on initialization/startup of the services the > graph will straighten that out. > > But.. what if serviceA starts before LMI runs... to pick up the one in > /var :( It will be stuck offline > until serviceB is imported in LMI, and then serviceB will start > allowing serviceA to start correct? > This problem happens regardless of whether or not we support importing /var/svc/ manifests during EMI. If we support /var/svc manifests during EMI, users can potentially configure /var to be a separate dataset/filesystem that is mounted at a later point thus /var is inaccessible during EMI. This configuration will also result in the above scenario. This is probably rare but could happen.
The advantage of limiting EMI to process only /etc/svc is that developers can catch the above situation before deployment and correct their services. -tony
