URL: https://github.com/SSSD/sssd/pull/142 Author: fidencio Title: #142: Fix breakage caused by having the NSS responder socket-activated Action: opened
PR body: """ **MONITOR: Don't return an error in case we fail to register a service** This patch is a pre-requisite for the next one and both patches could be squashed. Although, I've decided to keep this patch separated as it brings back a behaviour that has been used before the socket-activation series.
**MONITOR: Avoid starting an already started service** This patch basically prevents that we end up socket-activating (or systemd enabling) a service that has been explicitly enabled in the services' line.
**NSS: Make sssd-nss.service a hard requirement** This patch takes a step back and do not allow the NSS responder to be socket-activated. Instead, it's added as a hard requirement for SSSD (in sssd.service) and will be running whenever SSSD is running. Thankfully to our previous patch we ensure that the only one instance of the responder will be running. So, if nss is not present in the services' line, it will be started as a dep. In case it's part of the services' line, it will be started by the monitor and later on started by systemd. The second instance will be killed during the services' registration part. """
To pull the PR as Git branch: git remote add ghsssd https://github.com/SSSD/sssd git fetch ghsssd pull/142/head:pr142 git checkout pr142
URL: https://github.com/SSSD/sssd/pull/142 Title: #142: Fix breakage caused by having the NSS responder socket-activated
fidencio commented: """ CI: http://sssd-ci.duckdns.org/logs/job/61/67/summary.html """
See the full comment at https://github.com/SSSD/sssd/pull/142#issuecomment-277394350
URL: https://github.com/SSSD/sssd/pull/142 Title: #142: Fix breakage caused by having the NSS responder socket-activated
Label: +Changes requested
URL: https://github.com/SSSD/sssd/pull/142 Title: #142: Fix breakage caused by having the NSS responder socket-activated
fidencio commented: """ The second patch is not good to go. When a responder is started by the monitor and then the same responder is socket-activated or manually started by systemctl it will lead us in a bad state as the socket(s) used by the first will be unlinked.
So, we need a more robust way to avoid this to happen. It's not going to happen ever when systemd is managing the services (systemd takes care of this case), but it is likely to happen (and will happen if you consider the 3rd patch) when we mix up who is managing the services. """
See the full comment at https://github.com/SSSD/sssd/pull/142#issuecomment-277434802
URL: https://github.com/SSSD/sssd/pull/142 Title: #142: Fix breakage caused by having the NSS responder socket-activated
fidencio commented: """ And I've been told that disabling the NSS Responder socket is a no-go. Closing the PR, so, closing the PR. """
See the full comment at https://github.com/SSSD/sssd/pull/142#issuecomment-277450764
URL: https://github.com/SSSD/sssd/pull/142 Title: #142: Fix breakage caused by having the NSS responder socket-activated
Label: +Rejected
URL: https://github.com/SSSD/sssd/pull/142 Author: fidencio Title: #142: Fix breakage caused by having the NSS responder socket-activated Action: closed
To pull the PR as Git branch: git remote add ghsssd https://github.com/SSSD/sssd git fetch ghsssd pull/142/head:pr142 git checkout pr142
sssd-devel@lists.fedorahosted.org
-
fidencio