[23998] in Source-Commits
/svn/athena r23607 - trunk/debathena/config/reactivate/debian
daemon@ATHENA.MIT.EDU (Evan Broder)
Thu Mar 12 20:07:30 2009
Date: Thu, 12 Mar 2009 20:07:20 -0400
From: Evan Broder <broder@MIT.EDU>
Message-Id: <200903130007.n2D07KUf011406@drugstore.mit.edu>
To: source-commits@mit.edu
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Author: broder
Date: 2009-03-12 20:07:20 -0400 (Thu, 12 Mar 2009)
New Revision: 23607
Added:
trunk/debathena/config/reactivate/debian/debathena-reactivate.logrotate
Modified:
trunk/debathena/config/reactivate/debian/athena-login-snapshot
trunk/debathena/config/reactivate/debian/changelog
Log:
In reactivate:
* Add instrumentation to track why reactivations are failing.
rotate 6
monthly
compress
missingok
notifempty
Modified: trunk/debathena/config/reactivate/debian/athena-login-snapshot
===================================================================
--- trunk/debathena/config/reactivate/debian/athena-login-snapshot 2009-03-12 19:53:05 UTC (rev 23606)
+++ trunk/debathena/config/reactivate/debian/athena-login-snapshot 2009-03-13 00:07:20 UTC (rev 23607)
@@ -21,6 +21,7 @@
# that circumstance should be fairly rare.
set -e
+exec >>/var/log/athena-reactivate 2>&1
updflag=/var/run/athena-update-in-progress
bootflag=/var/run/athena-reboot-after-update
@@ -40,6 +41,22 @@
uloginlvname=login-update
uloginlvpath=/dev/$vgname/login-update
+echo "-----"
+echo "** Beginning Athena reactivation ($event) session at $(date)"
+
+finish() {
+ echo "** Finishing Athena reactivation ($event) at $(date)"
+ echo "-----"
+ echo
+ exit
+}
+trap finish EXIT
+
+v() {
+ echo "** Running:" "$@"
+ echo "$@"
+}
+
(
flock -x 9
case $event in
@@ -49,7 +66,7 @@
if [ -e "$loginlvpath" ]; then
# A login snapshot already exists; perhaps the machine rebooted
# during a login. Clean it up.
- lvremove -f "$loginlvpath"
+ v lvremove -f "$loginlvpath"
fi
if [ -e "$updflag" ]; then
@@ -58,31 +75,31 @@
# (If we already used it up, /etc/nologin should prevent us from
# getting here until the update ends.) Rename it to login.
[ -e "$uloginlvpath" ]
- lvrename "$vgname" "$uloginlvname" "$loginlvname"
+ v lvrename "$vgname" "$uloginlvname" "$loginlvname"
else
# No update is in progress. Create our own snapshot of the root.
sync
- lvcreate --snapshot --size "$snapshotsize" --name "$loginlvname" \
+ v lvcreate --snapshot --size "$snapshotsize" --name "$loginlvname" \
"$rootlvpath"
fi
# Mount the login snapshot.
mkdir -p /login
- mount "$loginlvpath" /login
+ v mount "$loginlvpath" /login
# Enable subtree operations on /media by making it a mount point,
# then share it.
- mount --bind /media /media
- mount --make-shared /media
+ v mount --bind /media /media
+ v mount --make-shared /media
# Bind-mount a bunch of stuff from the real root into the chroot.
for dir in $binddirs; do
- mount --bind "$dir" "/login$dir"
+ v mount --bind "$dir" "/login$dir"
done
# Add the user to a bunch of groups in the chroot.
for group in $addgroups; do
- chroot /login gpasswd -a "$USER" "$group"
+ v chroot /login gpasswd -a "$USER" "$group"
done
# There are some daemons that should be running inside the
@@ -95,12 +112,12 @@
# some point, so we have to try both names and catch the error for
# the one that doesn't exist
for daemon in $daemons; do
- invoke-rc.d $daemon stop || [ $? = 100 ]
- chroot /login invoke-rc.d $daemon start || [ $? = 100 ]
+ v invoke-rc.d $daemon stop || [ $? = 100 ]
+ v chroot /login invoke-rc.d $daemon start || [ $? = 100 ]
done
- touch /login/ClusterLogin
- touch /var/run/athena-login
+ v touch /login/ClusterLogin
+ v touch /var/run/athena-login
# Add an schroot.conf entry for the chroot.
conf=/etc/schroot/schroot.conf
@@ -119,39 +136,39 @@
login-end)
# Clean-up the temporary file to indicate the logged in state
- rm /var/run/athena-login
+ v rm /var/run/athena-login
# Stop any daemons that were specifically started inside the
# chroot
for daemon in $daemons; do
- chroot /login invoke-rc.d $daemon stop || [ $? = 100 ]
- invoke-rc.d $daemon start || [ $? = 100 ]
+ v chroot /login invoke-rc.d $daemon stop || [ $? = 100 ]
+ v invoke-rc.d $daemon start || [ $? = 100 ]
done
# Clean up any remaining user processes using the bind mounts.
if [ -n "$USER" -a "$USER" != root ]; then
for dir in $binddirs; do
- su -s /bin/sh "$USER" -c "fuser -km /login$dir" > /dev/null || true
+ v su -s /bin/sh "$USER" -c "fuser -km /login$dir" > /dev/null || true
done
fi
# Clean up any processes using the chroot mountpoint.
- fuser -km /login > /dev/null || true
+ v fuser -km /login > /dev/null || true
sleep 2
# Clean up the bind mounts we made earlier.
# If any of these fail, the umount of /login will fail below,
# and we will reboot.
for dir in $(echo $binddirs|tac -s\ ); do
- umount "/login$dir" || true
+ v umount "/login$dir" || true
done
# Unmount /media, which we bind-mounted to itself earlier so it
# could be shared and then bind-mounted.
- umount /media || true
+ v umount /media || true
# Attempt to unmount /login.
- if ! umount /login; then
+ if ! v umount /login; then
# There may be an unkillable process in I/O wait keeping the
# mountpoint busy. We need to reboot the machine.
if [ -e "$updflag" ]; then
@@ -160,15 +177,16 @@
if [ ! -e /etc/nologin ]; then
echo "An update and reboot is in progress, please try again later." \
> /etc/nologin.update
- ln /etc/nologin.update /etc/nologin
+ v ln /etc/nologin.update /etc/nologin
fi
- touch "$bootflag"
+ v touch "$bootflag"
else
# We can just reboot now.
- reboot
+ echo "** Rebooting because of umount /login failure"
+ v reboot
fi
fi
- lvremove -f "$loginlvpath"
+ v lvremove -f "$loginlvpath"
if [ -e "$updflag" -a ! -e "$uloginlvpath" ]; then
# An update is in progress and we just used up its snapshot. We
@@ -188,37 +206,37 @@
if [ -e "$uloginlvpath" ]; then
# It already exists; perhaps the machine rebooted during an
# update. Clean it up.
- lvremove -f "$uloginlvpath"
+ v lvremove -f "$uloginlvpath"
fi
sync
- lvcreate --snapshot --size "$snapshotsize" --name "$uloginlvname" \
+ v lvcreate --snapshot --size "$snapshotsize" --name "$uloginlvname" \
"$rootlvpath"
# Touch the flag file signifying an update in progress.
- touch "$updflag"
+ v touch "$updflag"
;;
update-end)
if [ -e "$uloginlvpath" ]; then
# It appears our login snapshot was never used. Clean it up.
- lvremove -f "$uloginlvpath"
+ v lvremove -f "$uloginlvpath"
fi
if [ -e /etc/nologin.update ]; then
# Our login snapshot was used and that login ended before we
# did, causing further logins to block. Now that the update has
# ended, we can unblock logins.
- rm -f /etc/nologin.update /etc/nologin
+ v rm -f /etc/nologin.update /etc/nologin
fi
if [ -e "$bootflag" ]; then
# We need to reboot in order to unmount /login.
echo "Rebooting in order to unmount /login."
- reboot
+ v reboot
fi
# Remove the flag file signifying an update in progress.
- rm -f "$updflag"
+ v rm -f "$updflag"
;;
esac
) 9> $lockfile
Modified: trunk/debathena/config/reactivate/debian/changelog
===================================================================
--- trunk/debathena/config/reactivate/debian/changelog 2009-03-12 19:53:05 UTC (rev 23606)
+++ trunk/debathena/config/reactivate/debian/changelog 2009-03-13 00:07:20 UTC (rev 23607)
@@ -1,3 +1,9 @@
+debathena-reactivate (1.14) unstable; urgency=low
+
+ * Add instrumentation to track why reactivations are failing.
+
+ -- Evan Broder <broder@mit.edu> Thu, 12 Mar 2009 20:06:10 -0400
+
debathena-reactivate (1.13) unstable; urgency=low
* Instead of generating policy-rc.d at login time, have a general
Added: trunk/debathena/config/reactivate/debian/debathena-reactivate.logrotate
===================================================================
--- trunk/debathena/config/reactivate/debian/debathena-reactivate.logrotate 2009-03-12 19:53:05 UTC (rev 23606)
+++ trunk/debathena/config/reactivate/debian/debathena-reactivate.logrotate 2009-03-13 00:07:20 UTC (rev 23607)
@@ -0,0 +1,7 @@
+/var/log/athena-reactivate {
+ rotate 6
+ monthly
+ compress
+ missingok
+ notifempty
+}
