[25337] in Source-Commits
Re: /svn/athena r24910 - trunk/debathena/config/zephyr-config/debian
daemon@ATHENA.MIT.EDU (Jonathan Reed)
Fri Dec 3 20:38:39 2010
Mime-Version: 1.0 (Apple Message framework v1082)
Content-Type: text/plain; charset=us-ascii
From: Jonathan Reed <jdreed@MIT.EDU>
In-Reply-To: <201012040039.oB40djWp021970@drugstore.mit.edu>
Date: Fri, 3 Dec 2010 20:38:32 -0500
Cc: source-commits@mit.edu
Message-Id: <F8BCEED0-F335-4FE7-9BB3-0A612966B4E4@mit.edu>
To: Geoffrey Thomas <geofft@mit.edu>
Content-Transfer-Encoding: 8bit
Have you verified that this actually works, keeping in mind that given the order of the race condition, we'll see a zhm restart (from ifup.d) followed by an (attempted) start (from the initscript).
If so, then "ACK".
-Jon
On Dec 3, 2010, at 7:39 PM, Geoffrey Thomas wrote:
> Author: geofft
> Date: 2010-12-03 19:39:45 -0500 (Fri, 03 Dec 2010)
> New Revision: 24910
>
> Modified:
> trunk/debathena/config/zephyr-config/debian/changelog
> trunk/debathena/config/zephyr-config/debian/debathena-zephyr-config.if-up
> Log:
> In zephyr-config:
> * In the if-up.d script, restart instead of start, as the zhm
> initscript passes -N (don't send an HM_BOOT flushing subscriptions)
> on the restart command. This works around losing subscriptions if
> you manage to start multiple zhms (which is possible due to a Zephyr
> 3 bug involving SO_REUSEADDR) by the initscript and if-up.d script
> racing; see discussion around Trac #133. Upstream will fix the
> actual bug allowing multiple zhms to run.
> We should possibly actually do as Trac #133 says and kill off the
> initscript entirely in this package, possibly by replacing it with
> an Upstart job that starts on runlevel or network, or alternatively
> let upstream do so eventually, but this fixes the immediate issue.
>
>
> Modified: trunk/debathena/config/zephyr-config/debian/changelog
> ===================================================================
> --- trunk/debathena/config/zephyr-config/debian/changelog 2010-11-27 05:36:52 UTC (rev 24909)
> +++ trunk/debathena/config/zephyr-config/debian/changelog 2010-12-04 00:39:45 UTC (rev 24910)
> @@ -1,3 +1,20 @@
> +debathena-zephyr-config (1.10) unstable; urgency=low
> +
> + * In the if-up.d script, restart instead of start, as the zhm
> + initscript passes -N (don't send an HM_BOOT flushing subscriptions)
> + on the restart command. This works around losing subscriptions if
> + you manage to start multiple zhms (which is possible due to a Zephyr
> + 3 bug involving SO_REUSEADDR) by the initscript and if-up.d script
> + racing; see discussion around Trac #133. Upstream will fix the
> + actual bug allowing multiple zhms to run.
> +
> + We should possibly actually do as Trac #133 says and kill off the
> + initscript entirely in this package, possibly by replacing it with
> + an Upstart job that starts on runlevel or network, or alternatively
> + let upstream do so eventually, but this fixes the immediate issue.
> +
> + -- Geoffrey Thomas <geofft@mit.edu> Fri, 03 Dec 2010 19:18:09 -0500
> +
> debathena-zephyr-config (1.9.2) unstable; urgency=low
>
> * Ensure that zephyr-clients and libzephyr-krb5 stay in sync (Trac #586)
>
> Modified: trunk/debathena/config/zephyr-config/debian/debathena-zephyr-config.if-up
> ===================================================================
> --- trunk/debathena/config/zephyr-config/debian/debathena-zephyr-config.if-up 2010-11-27 05:36:52 UTC (rev 24909)
> +++ trunk/debathena/config/zephyr-config/debian/debathena-zephyr-config.if-up 2010-12-04 00:39:45 UTC (rev 24910)
> @@ -2,6 +2,6 @@
>
> case "$ADDRFAM" in
> inet|NetworkManager)
> - /etc/init.d/zhm start >/dev/null 2>&1 || :
> + /etc/init.d/zhm restart >/dev/null 2>&1 || :
> ;;
> esac
>
