[gentoo-user] How to find out all openrc dependencies?

Post by J. Roeleveld
Hi all,
For a while I've been seeing the following ERROR-messages when booting 1 of
* ERROR: cannot start multipathd as localmount would not start
* ERROR: cannot start zfs-import as localmount would not start
INIT: Entering runlevel: 3
* Starting multipathd ...
[ ok ]
* Importing ZFS pool(s) ...
[ ok ]
But I am trying to find the cause of these errors as they are preventing
parallel-start from actually working correctly.
# /lib/rc/bin/rc-depend multipathd
sysfs devfs udev udev-trigger modules fsck root localmount multipathd
# /lib/rc/bin/rc-depend localmount
sysfs devfs udev udev-trigger modules fsck root localmount
# /lib/rc/bin/rc-depend zfs-import
multipath sysfs devfs udev udev-trigger modules fsck root localmount
multipathd zfs-import
# /lib/rc/bin/rc-depend multipath
multipath
From how I read these, it should be able to start "localmount" properly
before even trying to start "multipathd" and "zfs-import"
There is also no technical dependency for "localmount" (the root filesystem
is not on ZFS on this system)
Any help/suggestions on how to find the cause would be appreciated.
--
Joost

Check if hwclock is in the boot runlevel:

rc-update -s -v | grep hwclock

J. Roeleveld

2024-04-11 10:00:01 UTC

rc-update -s -v | grep hwclock

What does "hwclock" got to do with this?
It has no dependency with multipathd, zfs-import, localmount or anything else
that is showing an error.

--
Joost

Michael

2024-04-11 10:20:01 UTC

rc-update -s -v | grep hwclock

What does "hwclock" got to do with this?
It has no dependency with multipathd, zfs-import, localmount or anything
else that is showing an error.
--
Joost

Our systems are certainly different, but I noticed this dependency on my
localmount which is missing on yours:

# /lib/rc/bin/rc-depend localmount
sysfs devfs udev udev-trigger hwclock modules fsck root dmcrypt localmount
^^^^^^^
Have you compared your system services which has this problem, with other
systems of yours which can startup properly?

J. Roeleveld

2024-04-14 07:30:01 UTC

Post by J. Roeleveld
Hi all,
For a while I've been seeing the following ERROR-messages when booting
1
of
* ERROR: cannot start multipathd as localmount would not start
* ERROR: cannot start zfs-import as localmount would not start
INIT: Entering runlevel: 3
* Starting multipathd ...
[ ok ]
* Importing ZFS pool(s) ...
[ ok ]
But I am trying to find the cause of these errors as they are preventing
parallel-start from actually working correctly.
# /lib/rc/bin/rc-depend multipathd
sysfs devfs udev udev-trigger modules fsck root localmount multipathd
# /lib/rc/bin/rc-depend localmount
sysfs devfs udev udev-trigger modules fsck root localmount
# /lib/rc/bin/rc-depend zfs-import
multipath sysfs devfs udev udev-trigger modules fsck root localmount
multipathd zfs-import
# /lib/rc/bin/rc-depend multipath
multipath
From how I read these, it should be able to start "localmount" properly
before even trying to start "multipathd" and "zfs-import"
There is also no technical dependency for "localmount" (the root filesystem
is not on ZFS on this system)
Any help/suggestions on how to find the cause would be appreciated.
--
Joost

rc-update -s -v | grep hwclock

What does "hwclock" got to do with this?
It has no dependency with multipathd, zfs-import, localmount or anything
else that is showing an error.
--
Joost

Our systems are certainly different, but I noticed this dependency on my
# /lib/rc/bin/rc-depend localmount
sysfs devfs udev udev-trigger hwclock modules fsck root dmcrypt localmount
^^^^^^^
Have you compared your system services which has this problem, with other
systems of yours which can startup properly?

Adding additional dependencies into the tree is more likely to cause further
issues. I am actually looking for how to quickly find out which dependency is
causing a circular dependency issue as the first time it thinks it needs to
start a service it fails. But the 2nd time it starts, it goes correctly.

I removed hwclock from ALL VMs as they don't actually have a hwclock.

I did find out the actual cause of the problem through a lot of trial and
error, but this is not really useful in actually quickly finding the problem.
Being able to "simulate" the startup sequence for how OpenRC wants to do
things would have simplified and sped up the entire process.

--
Joost

Michael

2024-04-14 08:30:02 UTC

Post by J. Roeleveld
Hi all,
For a while I've been seeing the following ERROR-messages when booting
1
of
* ERROR: cannot start multipathd as localmount would not start
* ERROR: cannot start zfs-import as localmount would not start
INIT: Entering runlevel: 3
* Starting multipathd ...
[ ok ]
* Importing ZFS pool(s) ...
[ ok ]
But I am trying to find the cause of these errors as they are preventing
parallel-start from actually working correctly.
# /lib/rc/bin/rc-depend multipathd
sysfs devfs udev udev-trigger modules fsck root localmount multipathd
# /lib/rc/bin/rc-depend localmount
sysfs devfs udev udev-trigger modules fsck root localmount
# /lib/rc/bin/rc-depend zfs-import
multipath sysfs devfs udev udev-trigger modules fsck root localmount
multipathd zfs-import
# /lib/rc/bin/rc-depend multipath
multipath
From how I read these, it should be able to start "localmount" properly
before even trying to start "multipathd" and "zfs-import"
There is also no technical dependency for "localmount" (the root filesystem
is not on ZFS on this system)
Any help/suggestions on how to find the cause would be appreciated.
--
Joost

rc-update -s -v | grep hwclock

What does "hwclock" got to do with this?
It has no dependency with multipathd, zfs-import, localmount or anything
else that is showing an error.
--
Joost

Out of curiosity - what was the cause of this? I have only come across
hwclock on my installations (not VMs).

Post by J. Roeleveld
but this is not really useful in actually quickly finding the
problem. Being able to "simulate" the startup sequence for how OpenRC wants
to do things would have simplified and sped up the entire process.
--
Joost

I enable the rc log and check how the various services try to start up,
however the information provided is not always useful.

J. Roeleveld

2024-04-14 16:50:01 UTC

Post by Michael
Our systems are certainly different, but I noticed this dependency on my
# /lib/rc/bin/rc-depend localmount
sysfs devfs udev udev-trigger hwclock modules fsck root dmcrypt localmount
^^^^^^^
Have you compared your system services which has this problem, with other
systems of yours which can startup properly?

Out of curiosity - what was the cause of this? I have only come across
hwclock on my installations (not VMs).

The issue is related to how openrc handles the boot process. It first starts
everything in "boot" and then switches to whatever is set as default runlevel.

At some point, Gentoo devs found it useful to add multipath and zfs-import to
the boot-runlevel. This was not originally the case.
I consider it best that the boot-runlevel is kept as empty as possible and
have the user/admin add necessary stuff to it when needed.

I removed them from the boot runlevel and the issue has been resolved.

This was the last item on the list preventing correct parallel boot and my
server now boots in 8 minutes instead of the original 20.
It's still 8 because I need to add some tricks to reduce some of the
simultaneous stuff which will lead to timeout issues which actually make it
worse if left to defaults.