David Bustos wrote: >Quoth Darren.Reed at sun.com on Wed, Apr 25, 2007 at 06:25:26PM -0700: > > >>With PSARC/2005/334, we made the pfil SMF service redundant, >>and therefore took to removing the manifest file from Solaris. >>This kind of works....except where it causes problems... >> >>The two problems we're trying to resolve are 6543556 and 6543558. >> >>What danger is there in leaving the manifest file present and just >>doing an disable and delete - will anything add it back in later on? >> >> > >It shouldn't if the manifest doesn't change. I think the bigger danger >is the inconsistency between the filesystem and the repository. We want >them to stay synchronized, module the user's customizations. And you >deleting the service doesn't count as a user customization. > >
Hmm. I'm not sure that leaving the file in place makes any difference to us. With 6543556, even if I try doing a manual "svccfg import /var/svc/manifest/network/pfil.xml" after the patchrm, it fails with: svccfg: Could not create "pfil_network" dependency in svc:/network/physical (already exists). svccfg: Import of /var/svc/manifest/network/pfil.xml failed. Progress: svccfg: Service "network/pfil": some properties imported. svccfg: Instance "default": previous snapshot taken. To me this smells like we need to ensure that the pfil manifest is (re)imported before the network/physical one is when the patch is backed out, correct? In addition to manually running svccfg I've tried just touch'ing pfil.xml to see if SMF will notice the timestamp change and pick it automatically (it does, but the end result isn't any better.) At present, patchrm of this patch leaves the box without any functioning network services. >>Or is there something else obvious that we should do that we're not? >> >> > >Is the problem that the patch install script deletes the service, and >then the backout restores the manifest, but it doesn't get imported? If >so, then the patch install script probably also needs to delete the >manifest's hash. > >If your problem is that the service won't be imported until >manifest-import runs, and that's too late, then we'll have to resort to >dirty tricks. > > The current patch we're working with (T120012-08 on sunsoftpatch.sfbay) can be applied but there are warnings about the pfil service at the next boot regardless of whether or not pfil.xml is present. This is 6543558. There is no explicit relationship joining pfil and manifest-import, they appear to join up at single-user, so it would seem quite possible for svc.startd to run pfil before it gets to manifest-import, correct? 6543558 appears just to be a noise generator, as the warning messages disappear with the next boot, so while it's untidy it's not a fatal problem like above. Darren
