On Sat, Sep 02, 2006 at 06:46:45PM -0700, Russ Allbery wrote: > > I just confirmed that when I install the 1.4.4~beta1-1 krb5kdc, > > whatever existing kdc is stopped and the new kdc binary ends up > > running. > > Hm. > > I should have tested that more thoroughly; I went from the documentation > of start-stop-daemon, which does indeed seem to imply that a daemon > running from a removed binary might not be stopped, but I didn't verify > that it actually failed. > > Ryan, could you give more details about how you encountered this problem?
I upgraded the packages in stable to the security versions, the daemons said that they restarted, but didn't. Note that start-stop-daemon in unstable will work with this combination of options, but it will not in stable. lsof is useful here, as is a /usr that is normally mounted read-only. You can't remount it back to read-only until all the old binaries are no longer in use. > > I'm all for LSB-style initscripts, so I don't mind the change to > > unstable. But I want to actually understand what issue we're fixing > > before issuing an update for stable. The binary is updated by dpkg, and then restart is called in the postinst. --stop with --exec checks that the binary running matches that on disk, I believe by comparing inode numbers. This fails, and the process is not stopped. The process then complains that it can't start, because it can't bind the port (as the old process is still running). -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
