On Tue, Jun 16, 2009 at 7:55 AM, Felix Meschberger<[email protected]> wrote: >> Isn't it possible to avoid that case with start levels? (we are >> talking about the initial startup only, if the configadmin goes away >> later, no problem, the system ready state should IMHO never switch >> from true to false as it is only for the startup) > > This is of course wrong. *If* we have a system status service, this is > used throughout the lifetime of the system. That is it may happen that > the system becomes unavailable if something is going away. > > Just to have a status mechanism for startup makes no sense.
Well, the whole point of the system ready detection is to find out at what point in time everything is up and running to serve requests (and not send a 503 Service Unavailable) after a startup. With "legacy" frameworks such as the old Spring, where components were hardwired, startup was deterministic and quite easily to detect. Now within the context of a highly asynchronous system we need a separate solution. But I don't think that bundles going away or crashing during the lifetime should lead to a 503 immediately. They might indicate a problem and one wants to see this problem rather than a 503. Or the system ready config has to be changed alongside a major redeployment took place and you don't want to restart the system (considering the config is stored in the sling.properties that are only read on restart). If we don't get this right, you see a 503 way too often also things might be running perfectly. Regards, Alex -- Alexander Klimetschek [email protected]
