Bug 475204 – enhancement: remove major slow-down in starting xenstored

• Status: RESOLVED FIXED
• Product: Gentoo Linux

[Description Another Mortal 2013-06-29 12:26:11 UTC]

There's an unconditional *15* second sleep in /etc/init.d/xenstored
slowing down the boot process for no good reason.  The attached patch
replaces this with a periodic check/sleep cycle.  On the handful of
systems I tested this on xenstored was available after only a few
seconds (nowhere near 15).  Please consider a more flexible approach!

Reproducible: Always




--- /usr/portage/app-emulation/xen-tools/files/xenstored.initd  2013-05-15 11:01:03.000000000 +0200
+++ /etc/init.d/xenstored       2013-06-29 14:19:24.842794505 +0200
@@ -14,7 +14,11 @@
                --pidfile /var/run/xenstored.pid \
                -- --pid-file=/var/run/xenstored.pid \
                $XENSTORED_OPTS
-       sleep 15
+       until xenstore-exists /
+       do
+               echo -n .
+               sleep .5
+       done
        einfo "Setting domain0 name record"
        /usr/bin/xenstore-write "/local/domain/0/name" "Domain-0"
        eend $*

[Comment 1 Ian Delaney (RETIRED) 2013-06-29 16:20:05 UTC]

nice to 'hear' from you again a.m  I'll look over it later today,

[Comment 2 Ian Delaney (RETIRED) 2013-07-01 05:43:26 UTC]

  01 Jul 2013; Ian Delaney <[e-mail]> files/xenstored.initd,
  xen-tools-4.2.1-r3.ebuild, xen-tools-4.2.1-r4.ebuild,
  xen-tools-4.2.2-r1.ebuild:
  remove major slow-down in starting xenstored, put forward by 'a.m.' in Bug
  #475204

I'll make it a test request for now and allow for comments saying 'good here'

[Comment 3 Andreas Kinzler 2013-07-01 11:05:56 UTC]

CVS still contains the "sleep 15"

[Comment 4 Ian Delaney (RETIRED) 2013-07-04 09:43:07 UTC]

og. Not any more

[Comment 5 Another Mortal 2013-07-23 08:49:40 UTC]

The fix in portage has 'sleep 5' instead of 'sleep .5'; this is still _way_ too long.  On my systems (with .5) I get _3_ dots max, which is still less than 2 seconds.  Why force a 5 second sleep?  Is this really needed?

If the portability of sleeping fractional seconds is at issue, could 'sleep 1' be a good compromise?

[Comment 6 Ian Delaney (RETIRED) 2013-07-23 13:30:08 UTC]

See Bug 476572 and comment

[Comment 7 Ian Delaney (RETIRED) 2013-07-30 14:42:16 UTC]

This can go fixed