André Braga wrote:
On the other hand, there's no need to over-engineer something either. I love the advantages of node monitoring and attributes, but they have their limitations as well. Some thing I came across in developing the plug-in support for Themis years ago was that the node monitoring system (in BeOS R5 at least) doesn't always send the notifications you'd expect. This was particularly true for deleting files, sometimes the notifications weren't sent because the file wasn't technically deleted, it was just moved to the trash directory, and the notification wasn't sent until the trash was emptied or some other similar events.Now, we all know how this story ends: a hundred thousand hacks to handle dependencies and conflicts between services.Why exactly would that happen? In the case of sshd, just a link for its executable in this dir would suffice.How exactly do you think the rc.d/init.d kludge was born? See if this rings familiar: a quick hack, to serve a specific purpose; never meant to be a permanent solution. But which sort of works well enough and is also deemed simple enough so that it will be trivial to replace by the eventual successor. But that suddenly gets used by more and more people -- despite the clear hackish nature of it. And then becomes sort of a de-facto way of doing similar stuff on the system (here, starting services). Which then gets the committee treatment and gets ratified into a standard.
Whatever solution is chosen, certainly as a permanent solution, needs to be well thought out, and its shelf life if temporary needs to be chiseled in stone.
Just my thoughts, Raymond