Discussion:
how disable fsck on swap+lvm
(too old to reply)
Alexey Shabalin
2011-02-16 14:37:39 UTC
Permalink
You can see in log:
Feb 16 16:04:51 host-29 kernel: [ 11.636564] Adding 524284k swap on
/dev/mapper/VG1-Swap. Priority:-1 extents:1 across:524284k
Feb 16 16:04:51 host-29 kernel: [ 11.669141] clock[1954]: Setting
system clock (localtime): succeeded
Feb 16 16:04:51 host-29 kernel: [ 11.770936] clock[1965]: Today's
date: Wed Feb 16 16:03:48 MSK 2011 succeeded
Feb 16 16:04:51 host-29 kernel: [ 68.503001] systemd[1]: Job
dev-disk-by\x2duuid-bf5d2d3e\x2d174c\x2d44ea\x2dbf89\x2dcd6ecead5f81.device/start
timed out.

swap on lvm(VG1-Swap), and systemd run fsck on this partition.
I know that swap on lvm no good practics, but need foolproof.
--
Alexey Shabalin
Alexey Shabalin
2011-02-16 15:33:28 UTC
Permalink
Feb 16 16:04:51 host-29 kernel: [   11.636564] Adding 524284k swap on
/dev/mapper/VG1-Swap.  Priority:-1 extents:1 across:524284k
Feb 16 16:04:51 host-29 kernel: [   11.669141] clock[1954]: Setting
system clock (localtime): succeeded
Feb 16 16:04:51 host-29 kernel: [   11.770936] clock[1965]: Today's
date: Wed Feb 16 16:03:48 MSK 2011 succeeded
Feb 16 16:04:51 host-29 kernel: [   68.503001] systemd[1]: Job
dev-disk-by\x2duuid-bf5d2d3e\x2d174c\x2d44ea\x2dbf89\x2dcd6ecead5f81.device/start
timed out.
swap on lvm(VG1-Swap), and systemd run fsck on this partition.
I know that swap on lvm no good practics, but need foolproof.
PS: lvm and swap enabled by external script. systemd can't run
vgchange and enable swap on lvm.
--
Alexey Shabalin
Lennart Poettering
2011-02-16 17:57:30 UTC
Permalink
Post by Alexey Shabalin
Feb 16 16:04:51 host-29 kernel: [ 11.636564] Adding 524284k swap on
/dev/mapper/VG1-Swap. Priority:-1 extents:1 across:524284k
Feb 16 16:04:51 host-29 kernel: [ 11.669141] clock[1954]: Setting
system clock (localtime): succeeded
Feb 16 16:04:51 host-29 kernel: [ 11.770936] clock[1965]: Today's
date: Wed Feb 16 16:03:48 MSK 2011 succeeded
Feb 16 16:04:51 host-29 kernel: [ 68.503001] systemd[1]: Job
dev-disk-by\x2duuid-bf5d2d3e\x2d174c\x2d44ea\x2dbf89\x2dcd6ecead5f81.device/start
timed out.
swap on lvm(VG1-Swap), and systemd run fsck on this partition.
This does not show anything related to fsck?

Also, systemd will invoke hwclock for you anyway. You need no additional
clock utilities.
Post by Alexey Shabalin
I know that swap on lvm no good practics, but need foolproof.
If you use LVM make sure to use a new version that properly installs its
udev rules.

Lennart
--
Lennart Poettering - Red Hat, Inc.
Alexey Shabalin
2011-02-17 15:25:43 UTC
Permalink
Post by Lennart Poettering
Feb 16 16:04:51 host-29 kernel: [   11.636564] Adding 524284k swap on
/dev/mapper/VG1-Swap.  Priority:-1 extents:1 across:524284k
Feb 16 16:04:51 host-29 kernel: [   11.669141] clock[1954]: Setting
system clock (localtime): succeeded
Feb 16 16:04:51 host-29 kernel: [   11.770936] clock[1965]: Today's
date: Wed Feb 16 16:03:48 MSK 2011 succeeded
Feb 16 16:04:51 host-29 kernel: [   68.503001] systemd[1]: Job
dev-disk-by\x2duuid-bf5d2d3e\x2d174c\x2d44ea\x2dbf89\x2dcd6ecead5f81.device/start
timed out.
swap on lvm(VG1-Swap), and systemd run fsck on this partition.
This does not show anything related to fsck?
Also, systemd will invoke hwclock for you anyway. You need no additional
clock utilities.
hwclock - very easy and cann't read config /etc/sysconfig/clock:
HWCLOCK_SET_TIME_AT_START=true
HWCLOCK_SET_AT_HALT=true
HWCLOCK_ADJUST=true
UTC=true
ZONE="Europe/Moscow" (that /etc/localtime is a copy of)
Post by Lennart Poettering
I know that swap on lvm no good practics, but need foolproof.
If you use LVM make sure to use a new version that properly installs its
udev rules.
I used lvm-2.02.84. I don't see udev rules that run "vgchange -a y".
--
Alexey Shabalin
Lennart Poettering
2011-02-17 15:36:34 UTC
Permalink
Post by Alexey Shabalin
Post by Lennart Poettering
Feb 16 16:04:51 host-29 kernel: [   11.636564] Adding 524284k swap on
/dev/mapper/VG1-Swap.  Priority:-1 extents:1 across:524284k
Feb 16 16:04:51 host-29 kernel: [   11.669141] clock[1954]: Setting
system clock (localtime): succeeded
Feb 16 16:04:51 host-29 kernel: [   11.770936] clock[1965]: Today's
date: Wed Feb 16 16:03:48 MSK 2011 succeeded
Feb 16 16:04:51 host-29 kernel: [   68.503001] systemd[1]: Job
dev-disk-by\x2duuid-bf5d2d3e\x2d174c\x2d44ea\x2dbf89\x2dcd6ecead5f81.device/start
timed out.
swap on lvm(VG1-Swap), and systemd run fsck on this partition.
This does not show anything related to fsck?
Also, systemd will invoke hwclock for you anyway. You need no additional
clock utilities.
HWCLOCK_SET_TIME_AT_START=true
HWCLOCK_SET_AT_HALT=true
HWCLOCK_ADJUST=true
UTC=true
ZONE="Europe/Moscow" (that /etc/localtime is a copy of)
Well, the fact whether the RTC is in UTC is stored in line 3 of
/etc/adjtime, and that should be the only place to configure it. There
is no point in duplicating this information in a non-standard way
elsewhere.

And the zone is stored in /etc/localtime and something similar applies
too.

And I am not sure it really makes sense to make it configurable whether
the RTC should be set on shutdown or not. It should be. Full stop. If
somebody really wants to disable it, he should just mask the service by
doing this:

# ln -sf /dev/null /etc/systemd/system/hwclock-save.service

That's much nicer and simpler, and cross-distro.
Post by Alexey Shabalin
Post by Lennart Poettering
I know that swap on lvm no good practics, but need foolproof.
If you use LVM make sure to use a new version that properly installs its
udev rules.
I used lvm-2.02.84. I don't see udev rules that run "vgchange -a y".
Well, ensure you have /lib/udev/rules.d/11-dm-lvm.rules installed.

Lennart
--
Lennart Poettering - Red Hat, Inc.
Alexey Shabalin
2011-02-17 16:07:26 UTC
Permalink
Post by Lennart Poettering
Post by Alexey Shabalin
Post by Lennart Poettering
Post by Alexey Shabalin
I know that swap on lvm no good practics, but need foolproof.
If you use LVM make sure to use a new version that properly installs its
udev rules.
I used lvm-2.02.84. I don't see udev rules that run "vgchange -a y".
Well, ensure you have /lib/udev/rules.d/11-dm-lvm.rules installed.
I have this rule, of course.
But i have not some rule with "vgchange -a y", and have lv-disk as "LV
Status not available"
--
Alexey Shabalin
Miklos Vajna
2011-02-18 00:37:20 UTC
Permalink
Post by Alexey Shabalin
But i have not some rule with "vgchange -a y", and have lv-disk as "LV
Status not available"
Here is what we use at the moment:

[Unit]
Description=LVM activation
DefaultDependencies=no
After=udev-settle.service
Before=local-fs.target
ConditionPathExists=/sbin/vgchange

[Service]
ExecStart=/sbin/vgchange --sysinit -a y
Type=oneshot

[Install]
WantedBy=sysinit.target

Would be nice to have this in upstream lvm2, though. (Mea culpa, I
didn't try it, either. :) )
Andrey Borzenkov
2011-02-18 03:39:57 UTC
Permalink
Post by Miklos Vajna
Post by Alexey Shabalin
But i have not some rule with "vgchange -a y", and have lv-disk as "LV
Status not available"
[Unit]
Description=LVM activation
DefaultDependencies=no
After=udev-settle.service
Before=local-fs.target
ConditionPathExists=/sbin/vgchange
Why don't you just ship it with LVM so this condition becomes unnecessary?
Post by Miklos Vajna
[Service]
ExecStart=/sbin/vgchange --sysinit -a y
Type=oneshot
[Install]
WantedBy=sysinit.target
Would be nice to have this in upstream lvm2, though. (Mea culpa, I
didn't try it, either. :) )
_______________________________________________
systemd-devel mailing list
http://lists.freedesktop.org/mailman/listinfo/systemd-devel
Lennart Poettering
2011-02-18 10:31:48 UTC
Permalink
Post by Miklos Vajna
Post by Alexey Shabalin
But i have not some rule with "vgchange -a y", and have lv-disk as "LV
Status not available"
[Unit]
Description=LVM activation
DefaultDependencies=no
After=udev-settle.service
Before=local-fs.target
ConditionPathExists=/sbin/vgchange
Please don't litter unit files with unnecssary ConditionPathExists=
please. .service files should be installed along side their binaries and
shipped in the same packages. So that if the .service file exists the
binary exists too, and if the binary doesn't then the .servie file
doesn't exist either.

Otherwise looks good. And yupp, e should have that in upstream LVM.

Lennart
--
Lennart Poettering - Red Hat, Inc.
Tom Gundersen
2011-02-18 11:02:15 UTC
Permalink
Post by Miklos Vajna
[Unit]
Description=LVM activation
DefaultDependencies=no
After=udev-settle.service
Before=local-fs.target
Is the Before=local-fs.target really necessary?

Any mounts that will reside on the lvm will already be
Before=local-fs.target, and in the case where automounts are used
instead, I guess we do not want to wait for lvm before local-fs.target
becomes available?

I have the same question about the different fsck services. They
should clearly be Before=*.mount, but wouldn't it be better if they
were not Before=local-fs.service?

Cheers,

Tom
Andrey Borzenkov
2011-02-18 11:24:06 UTC
Permalink
Post by Tom Gundersen
Post by Miklos Vajna
[Unit]
Description=LVM activation
DefaultDependencies=no
After=udev-settle.service
Before=local-fs.target
Is the Before=local-fs.target really necessary?
Any mounts that will reside on the lvm will already be
Before=local-fs.target, and in the case where automounts are used
instead, I guess we do not want to wait for lvm before local-fs.target
becomes available?
I think it makes sense. As long as no file system is using LVM (at
least, no filesystem, activated by default) there is no point to wait
for it.
Post by Tom Gundersen
I have the same question about the different fsck services. They
should clearly be Before=*.mount, but wouldn't it be better if they
were not Before=local-fs.service?
Makes sense for the same reason. local-fs should wait only for
filesystems, activated by default; they should handle own dependencies
themselves.

As long as we re on theme ...

How to express interdependency between different virtual devices? I.e.
we have at least mdadm and device-mapper; mdadm is quite probably
assembled incrementally so there is no clear checkpoint we can wait
for; and LVM can use arrays managed by mdraid so it has to be started
after it. How folks are handling it in different distributions?
Lennart Poettering
2011-02-18 11:48:29 UTC
Permalink
Post by Andrey Borzenkov
Post by Tom Gundersen
I have the same question about the different fsck services. They
should clearly be Before=*.mount, but wouldn't it be better if they
were not Before=local-fs.service?
Makes sense for the same reason. local-fs should wait only for
filesystems, activated by default; they should handle own dependencies
themselves.
As long as we re on theme ...
How to express interdependency between different virtual devices? I.e.
we have at least mdadm and device-mapper; mdadm is quite probably
assembled incrementally so there is no clear checkpoint we can wait
for; and LVM can use arrays managed by mdraid so it has to be started
after it. How folks are handling it in different distributions?
The way for now is to do things how they always have been done,
i.e. just initialize one after the other with lots of wasted time in
between. On Fedora we still use a shell script for that copied out of
the old rc.sysinit.

In the long run all these things need to become dynamic and actively
watch for devices popping up. But this requires bigger changes in both
LVM and MD and we'll have to see when LVM and MD actually catch up with
this and make the change.

Lennart
--
Lennart Poettering - Red Hat, Inc.
Lennart Poettering
2011-02-18 11:39:07 UTC
Permalink
Post by Tom Gundersen
Post by Miklos Vajna
[Unit]
Description=LVM activation
DefaultDependencies=no
After=udev-settle.service
Before=local-fs.target
Is the Before=local-fs.target really necessary?
Any mounts that will reside on the lvm will already be
Before=local-fs.target, and in the case where automounts are used
instead, I guess we do not want to wait for lvm before local-fs.target
becomes available?
Well, maybe syncing it to local-fs is not necessary, but we probably
want to sync it to basic.target at least. Mounts are not the only users
of block devices, so a later service might want to use it and hence we
should make sure that normal services (which are started after
basic.target) can rely that LVM stuff is set up. Some database servers
are able to use block devices directly, without any FS on them.
Post by Tom Gundersen
I have the same question about the different fsck services. They
should clearly be Before=*.mount, but wouldn't it be better if they
were not Before=local-fs.service?
You have a point there. I have now relaxed this to base.target. (Just to
avoid confusion by stuff like udisks and other disk daemon software,
which we might not want to start at a time where the fsck is still going on).

Lennart
--
Lennart Poettering - Red Hat, Inc.
Tom Gundersen
2011-02-18 13:28:08 UTC
Permalink
On Fri, Feb 18, 2011 at 12:39 PM, Lennart Poettering
Post by Lennart Poettering
Well, maybe syncing it to local-fs is not necessary, but we probably
want to sync it to basic.target at least. Mounts are not the only users
of block devices, so a later service might want to use it and hence we
should make sure that normal services (which are started after
basic.target) can rely that LVM stuff is set up. Some database servers
are able to use block devices directly, without any FS on them.
Post by Tom Gundersen
I have the same question about the different fsck services. They
should clearly be Before=*.mount, but wouldn't it be better if they
were not Before=local-fs.service?
You have a point there. I have now relaxed this to base.target. (Just to
avoid confusion by stuff like udisks and other disk daemon software,
which we might not want to start at a time where the fsck is still going on).
Regarding syncing things with base.target:

I agree that this is a good idea for the time being, to avoid a flood of bugs.

However, in the long term would it not be better to avoid syncing
stuff (like udev-settle.service, lvm.service, etc.) with base.target?

If services cannot deal with this then 1) they are buggy as they would
not be able to deal with devices being plugged/appearing after boot,
so they should be fixed 2) in the interim they should get
After=lvm.service (or whatever they rely on) to make it work like
before and also make it clear that they are buggy.

My apologies for going a bit off-topic...

Cheers,

Tom

PS
Sorry to Lennart for sending twice, I forgot to cc the ML
Alexey Shabalin
2011-02-21 15:47:43 UTC
Permalink
Post by Lennart Poettering
Post by Alexey Shabalin
Post by Lennart Poettering
Feb 16 16:04:51 host-29 kernel: [   11.636564] Adding 524284k swap on
/dev/mapper/VG1-Swap.  Priority:-1 extents:1 across:524284k
Feb 16 16:04:51 host-29 kernel: [   11.669141] clock[1954]: Setting
system clock (localtime): succeeded
Feb 16 16:04:51 host-29 kernel: [   11.770936] clock[1965]: Today's
date: Wed Feb 16 16:03:48 MSK 2011 succeeded
Feb 16 16:04:51 host-29 kernel: [   68.503001] systemd[1]: Job
dev-disk-by\x2duuid-bf5d2d3e\x2d174c\x2d44ea\x2dbf89\x2dcd6ecead5f81.device/start
timed out.
swap on lvm(VG1-Swap), and systemd run fsck on this partition.
This does not show anything related to fsck?
I know that swap on lvm no good practics, but need foolproof.
If you use LVM make sure to use a new version that properly installs its
udev rules.
I used lvm-2.02.84. I don't see udev rules that run "vgchange -a y".
Well, ensure you have /lib/udev/rules.d/11-dm-lvm.rules installed.
I was test two variant.
- systemd-18, udev-166, lvm2-2.02.84,mdadm-3.1.4
- dont't use rc.sysinit or any other external script (like
fedora-storage-init or fedora-swap). Only systemd services and
upstream udev rules.

First variant:
/ - on LVM, runned by initrd
swap - on LVM


Feb 21 16:21:35 host-29 kernel: [ 0.000000] Kernel command line:
BOOT_IMAGE=/boot/vmlinuz root=/dev/mapper/VG1-root ro panic=30 splash
init=/bin/systemd systemd.log_level=debug
-- skip --
Feb 21 16:21:35 host-29 kernel: [ 10.418096] systemd[1]: Installed
new job swap.target/start as 20
Feb 21 16:21:35 host-29 kernel: [ 10.419570] systemd[1]: Installed
new job dev-disk-by\x2duuid-bf5d2d3e\x2d174c\x2d44ea\x2dbf89\x2dcd6ecead5f81.swap/start
as 21
Feb 21 16:21:35 host-29 kernel: [ 10.422231] systemd[1]: Installed
new job dev-disk-by\x2duuid-bf5d2d3e\x2d174c\x2d44ea\x2dbf89\x2dcd6ecead5f81.device/start
as 22
-- skip --
Feb 21 16:21:35 host-29 kernel: [ 12.995562] systemd[1]:
udev.service: control process exited, code=exited status=0
Feb 21 16:21:35 host-29 kernel: [ 13.004173] systemd[1]:
udev.service got final SIGCHLD for state start-post
Feb 21 16:21:35 host-29 kernel: [ 13.005934] systemd[1]:
udev.service changed start-post -> running
Feb 21 16:21:35 host-29 kernel: [ 13.007450] systemd[1]: Job
udev.service/start finished, success=yes
Feb 21 16:21:35 host-29 kernel: [ 13.014888] systemd[1]: About to
execute: /sbin/udevadm settle
Feb 21 16:21:35 host-29 kernel: [ 13.029302] systemd[1]: Forked
/sbin/udevadm as 2053
Feb 21 16:21:35 host-29 kernel: [ 13.031287] systemd[1]:
udev-settle.service changed dead -> start
Feb 21 16:21:35 host-29 kernel: [ 13.037557] systemd[1]: About to
execute: /sbin/udevadm trigger --type=failed --action=add
Feb 21 16:21:35 host-29 kernel: [ 13.044068] systemd[1]: Forked
/sbin/udevadm as 2055
Feb 21 16:21:35 host-29 kernel: [ 13.045777] systemd[1]:
udev-retry.service changed dead -> start
-- skip --
Feb 21 16:21:35 host-29 kernel: [ 13.803160] systemd[1]:
dev-disk-by\x2dpath-virtio\x2dpci\x2dvirtio2.device changed dead ->
plugged
Feb 21 16:21:35 host-29 kernel: [ 13.805666] systemd[1]:
dev-vda.device changed dead -> plugged
Feb 21 16:21:35 host-29 kernel: [ 13.807133] systemd[1]:
sys-devices-virtio\x2dpci-virtio2-block-vda.device changed dead ->
plugged
Feb 21 16:21:35 host-29 kernel: [ 14.475926] systemd[1]:
dev-disk-by\x2dpath-virtio\x2dpci\x2dvirtio2\x2dpart1.device changed
dead -> plugged
Feb 21 16:21:35 host-29 kernel: [ 14.478240] systemd[1]:
dev-vda1.device changed dead -> plugged
Feb 21 16:21:35 host-29 kernel: [ 14.479750] systemd[1]:
sys-devices-virtio\x2dpci-virtio2-block-vda-vda1.device changed dead
-> plugged
-- skip --
Feb 21 16:21:35 host-29 kernel: [ 27.344511] systemd[1]: Got D-Bus
request: org.freedesktop.DBus.Local.Disconnected() on
/org/freedesktop/DBus/Local
Feb 21 16:21:35 host-29 kernel: [ 70.422401] systemd[1]: Job
dev-disk-by\x2duuid-bf5d2d3e\x2d174c\x2d44ea\x2dbf89\x2dcd6ecead5f81.device/start
timed out.
Feb 21 16:21:35 host-29 kernel: [ 70.424801] systemd[1]: Job
dev-disk-by\x2duuid-bf5d2d3e\x2d174c\x2d44ea\x2dbf89\x2dcd6ecead5f81.device/start
finished, success=no
Feb 21 16:21:35 host-29 kernel: [ 70.427203] systemd[1]: Job
dev-disk-by\x2duuid-bf5d2d3e\x2d174c\x2d44ea\x2dbf89\x2dcd6ecead5f81.swap/start
finished, success=no
Feb 21 16:21:35 host-29 kernel: [ 70.432647] systemd[1]: swap.target
changed dead -> active
Feb 21 16:21:35 host-29 kernel: [ 70.434125] systemd[1]: Job
swap.target/start finished, success=yes

"Job swap.target/start finished, success=yes", but "device/start
finished, success=no" and swap not enabled.

Second variant:
/ - on md0 (raid1 mdadm), runned by initrd
/home - on LVM (raid1+lvm)

Feb 21 18:14:26 shabalin2 kernel: [ 3.818788] md: Autodetecting RAID arrays.
--skip--
Feb 21 18:14:26 shabalin2 kernel: [ 3.825068] md/raid1:md1: active
with 2 out of 2 mirrors
Feb 21 18:14:26 shabalin2 kernel: [ 3.829903] md/raid1:md0: active
with 2 out of 2 mirrors
Feb 21 18:14:26 shabalin2 kernel: [ 4.513273] EXT4-fs (md0):
mounted filesystem with ordered data mode. Opts: (null)
--skip--

Feb 21 18:14:26 shabalin2 kernel: [ 23.213353] systemd[1]: Got D-Bus
request: org.freedesktop.DBus.Local.Disconnected() on
/org/freedesktop/DBus/Local
Feb 21 18:14:26 shabalin2 kernel: [ 67.938433] systemd[1]: Job
dev-disk-by\x2duuid-a1d69a60\x2d3654\x2d4c22\x2da307\x2d13af8d94732e.device/start
timed out.
Feb 21 18:14:26 shabalin2 kernel: [ 67.938819] systemd[1]: Job
dev-disk-by\x2duuid-a1d69a60\x2d3654\x2d4c22\x2da307\x2d13af8d94732e.device/start
finished,success=no
Feb 21 18:14:26 shabalin2 kernel: [ 67.939158] systemd[1]: Job
home.mount/start finished, success=no
Feb 21 18:14:26 shabalin2 kernel: [ 67.939892] systemd[1]: Job
***@dev-disk-by\x2duuid-a1d69a60\x2d3654\x2d4c22\x2da307\x2d13af8d94732e.service/start
finished, success=no


"Job home.mount/start finished, success=no" and i have not mounted /home
I need manual run:
vgchange -a y
mount /home

systemd+udev can't replace rc.sysinit(or external script like
fedora-storage-init) now. :(
--
Alexey Shabalin
Miklos Vajna
2011-02-21 23:10:11 UTC
Permalink
Post by Alexey Shabalin
systemd+udev can't replace rc.sysinit(or external script like
fedora-storage-init) now. :(
For the second case, can you see if this patch:

http://article.gmane.org/gmane.linux.lvm.devel/6138

helps for you?

It replaces the last part of fedora-storage-init by a native service
file. Looks like upstream does not want to see systemd stuff in the
codebase yet.
Alexey Shabalin
2011-02-21 23:49:40 UTC
Permalink
Post by Miklos Vajna
Post by Alexey Shabalin
systemd+udev can't replace rc.sysinit(or external script like
fedora-storage-init) now. :(
http://article.gmane.org/gmane.linux.lvm.devel/6138
helps for you?
Ok, I will test this. Thank you.
Post by Miklos Vajna
It replaces the last part of fedora-storage-init by a native service
file. Looks like upstream does not want to see systemd stuff in the
codebase yet.
For replace rc.sysinit, may be make like rc.sysinit:
first run multipath, second mdadm, and last lvm services.
Andrey Borzenkov
2011-02-22 04:33:03 UTC
Permalink
Post by Alexey Shabalin
Post by Miklos Vajna
Post by Alexey Shabalin
systemd+udev can't replace rc.sysinit(or external script like
fedora-storage-init) now. :(
http://article.gmane.org/gmane.linux.lvm.devel/6138
helps for you?
Ok, I will test this. Thank you.
Post by Miklos Vajna
It replaces the last part of fedora-storage-init by a native service
file. Looks like upstream does not want to see systemd stuff in the
codebase yet.
first run multipath, second mdadm, and last lvm services.
That is exactly what fedora-storage-init does. Why not simply use it then? :)
Alexey Shabalin
2011-02-22 11:17:02 UTC
Permalink
Post by Andrey Borzenkov
Post by Alexey Shabalin
Post by Miklos Vajna
Post by Alexey Shabalin
systemd+udev can't replace rc.sysinit(or external script like
fedora-storage-init) now. :(
http://article.gmane.org/gmane.linux.lvm.devel/6138
helps for you?
Ok, I will test this. Thank you.
Post by Miklos Vajna
It replaces the last part of fedora-storage-init by a native service
file. Looks like upstream does not want to see systemd stuff in the
codebase yet.
first run multipath, second mdadm, and last lvm services.
That is exactly what fedora-storage-init does. Why not simply use it then? :)
Yes, did so.
http://git.altlinux.org/people/shaba/packages/systemd.git?p=systemd.git;a=commit;h=c46365ee7dfdf480dbc86d5bd15d5368126854d8
But Lennart say: "Shell is evil" :)
--
Alexey Shabalin
Miklos Vajna
2011-02-23 16:49:39 UTC
Permalink
Post by Alexey Shabalin
Post by Andrey Borzenkov
Post by Alexey Shabalin
first run multipath, second mdadm, and last lvm services.
That is exactly what fedora-storage-init does. Why not simply use it then? :)
Yes, did so.
http://git.altlinux.org/people/shaba/packages/systemd.git?p=systemd.git;a=commit;h=c46365ee7dfdf480dbc86d5bd15d5368126854d8
But Lennart say: "Shell is evil" :)
And he is right. :)

It would be nice to have multipath.service, mdadm.service and
lvm.service for that. I posted a patch for the last as a first step, but
stopped working on that, seeing upstream is not interested.
Andrey Borzenkov
2011-02-25 12:58:45 UTC
Permalink
Post by Miklos Vajna
Post by Alexey Shabalin
Post by Andrey Borzenkov
Post by Alexey Shabalin
first run multipath, second mdadm, and last lvm services.
That is exactly what fedora-storage-init does. Why not simply use it then? :)
Yes, did so.
http://git.altlinux.org/people/shaba/packages/systemd.git?p=systemd.git;a=commit;h=c46365ee7dfdf480dbc86d5bd15d5368126854d8
But Lennart say: "Shell is evil" :)
And he is right. :)
It would be nice to have multipath.service, mdadm.service and
lvm.service for that. I posted a patch for the last as a first step, but
stopped working on that, seeing upstream is not interested.
It would actually be much better ti have incremental assembly for
multipath and LVM. This already exists for mdadm (although may be
disabled by distribution).

I wonder if multipathd already can accept udev events and reconfigure
itself. At least, I remember messages in log every time new device
comes. In this case it would be enough to just start it early and let
it configure itself.

I do not think it is possible for LVM right now.
Lennart Poettering
2011-03-17 00:11:35 UTC
Permalink
Post by Alexey Shabalin
Post by Miklos Vajna
Post by Alexey Shabalin
systemd+udev can't replace rc.sysinit(or external script like
fedora-storage-init) now. :(
http://article.gmane.org/gmane.linux.lvm.devel/6138
helps for you?
Ok, I will test this. Thank you.
Post by Miklos Vajna
It replaces the last part of fedora-storage-init by a native service
file. Looks like upstream does not want to see systemd stuff in the
codebase yet.
first run multipath, second mdadm, and last lvm services.
Sorry for the later reply to this, somehow this mail got lost in my
mbox.

Anyway: please be aware that on fedora we start most storage things for
lvm, md and multipath with a traditional script for now. These
subsystems haven't yet adopted the new dynamic way how things are done.

Lennart
--
Lennart Poettering - Red Hat, Inc.
Continue reading on narkive:
Loading...