1:57 p.m.
Kernel booting used to be deterministic. That is, if you booted the
kernel 10 times, every one of them would be exactly the same. Then,
several years ago, the kernel boot process was made threaded to speed
it up. This means that every boot is slightly different.
Is there a way to make the kernel boot in deterministic fashion again?
Everything will be done sequentially, and no task will kick off if its
predecessors are not done.
Kernel boot parameter, dracut option, systemd option?
Thanks.
3:08 p.m.
New subject: Is there a way to specify that boot should be deterministic
instead of using threads?
On 01/06/2017 11:57 AM, stan wrote:
Is there a way to make the kernel boot in deterministic fashion
again?
Everything will be done sequentially, and no task will kick off if its
predecessors are not done.
I think I know the answer, but I'm going to ask anyway: why? I'm asking
because there's probably more than one way to do it (There usually is in
Linux.) and the best way will depend on what you need.
3:17 p.m.
New subject: Is there a way to specify that boot should be deterministic
instead of using threads?
On Fri, 6 Jan 2017 13:08:50 -0800
Joe Zeff <joe(a)zeff.us> wrote:
On 01/06/2017 11:57 AM, stan wrote:
> Is there a way to make the kernel boot in deterministic fashion
> again? Everything will be done sequentially, and no task will kick
> off if its predecessors are not done.
I think I know the answer, but I'm going to ask anyway: why? I'm
asking because there's probably more than one way to do it (There
usually is in Linux.) and the best way will depend on what you need.
So I can be assured that variation in the boot process is eliminated as
a factor in any problem I'm seeing.
To use the NetworkManager case as an example, I boot with multi-thread,
see the error, save the boot record. Then, boot with single thread /
process, don't get the error, and save the boot record. Compare them,
and the clue to what is going wrong is highlighted.
When I looked at the options, no obvious way to do this popped out, so
I thought I would tap the collective wisdom. If you know more than
one, I'd be happy to hear them all.
10:02 a.m.
New subject: Is there a way to specify that boot should be deterministic
instead of using threads?
> > Is there a way to make the kernel boot in deterministic fashion
> > again? Everything will be done sequentially, and no task will kick
> > off if its predecessors are not done.
>
> I think I know the answer, but I'm going to ask anyway: why? I'm
> asking because there's probably more than one way to do it (There
> usually is in Linux.) and the best way will depend on what you need.
So I can be assured that variation in the boot process is eliminated as
a factor in any problem I'm seeing.
To use the NetworkManager case as an example, I boot with multi-thread,
see the error, save the boot record. Then, boot with single thread /
process, don't get the error, and save the boot record. Compare them,
and the clue to what is going wrong is highlighted.
Just FYI, if you're trying to deal with a networking dependency (e.g., you
have multiple interfaces and you want to make sure they're all up before
proceeding) that is broken by design and has been marked as "will not fix"
by systemd, from what I can tell. System startup will proceed as soon as
*any* network interfaces are online. If that's not your problem, then ...
at least there's a chance there might be a fix for it.
-jdm
When I looked at the options, no obvious way to do this popped out, so
I thought I would tap the collective wisdom. If you know more than
one, I'd be happy to hear them all.
_______________________________________________
users mailing list -- users(a)lists.fedoraproject.org
To unsubscribe send an email to users-leave(a)lists.fedoraproject.org
1:10 p.m.
New subject: Is there a way to specify that boot should be deterministic
instead of using threads?
On Sun, 8 Jan 2017 11:02:50 -0500
Justin Moore <justin.nonwork(a)gmail.com> wrote:
Just FYI, if you're trying to deal with a networking dependency
(e.g., you have multiple interfaces and you want to make sure they're
all up before proceeding) that is broken by design and has been
marked as "will not fix" by systemd, from what I can tell. System
startup will proceed as soon as *any* network interfaces are online.
If that's not your problem, then ... at least there's a chance there
might be a fix for it.
No, I only have a single interface, but I notice in the logs that the
wait-on-online service fails during boot. I guess it doesn't matter for
my use case, or that is normal. There must be some complication on
waiting for multiple interfaces that makes wait-on-online service
difficult to provide, or why not just wait on *all* interfaces.
My problem seems to be a timeout in a function of NetworkManager called
dhcp4. When it is called, I presume it tries to allocate a dhcp
address, and times out after 45 seconds when it can't. Haven't tracked
it down completely.
The behavior is now affecting rawhide as well, so it must have been in
a recent (NetworkManager?) update, that I just applied to rawhide (I get
a little behind there, since I don't run it that often).
I'll probably try downloading a previous version of NetworkManager from
koji, and downgrade to see if it fixes the issue.
Thanks for your response.
2:51 a.m.
New subject: Is there a way to specify that boot should be deterministic
instead of using threads?
On 01/08/2017 11:10 AM, stan wrote:
No, I only have a single interface, but I notice in the logs that
the
wait-on-online service fails during boot. I guess it doesn't matter for
my use case, or that is normal. There must be some complication on
waiting for multiple interfaces that makes wait-on-online service
difficult to provide, or why not just wait on*all* interfaces.
Have you considered masking it?
7:01 a.m.
New subject: Is there a way to specify that boot should be deterministic
instead of using threads?
Allegedly, on or about 08 January 2017, Justin Moore sent:
Just FYI, if you're trying to deal with a networking dependency
(e.g.,
you have multiple interfaces and you want to make sure they're all up
before proceeding) that is broken by design and has been marked as
"will not fix" by systemd, from what I can tell. System startup will
proceed as soon as *any* network interfaces are online.
Going off on a tangent, does the system allow for you to insert your own
extra wait-for-all step as a prerequisite for the supplied
wait-for-online (whatever it's actually named) service?
Surely there's a need for users to be able to put holds into some of the
various /wait-for-it/ targets.
--
[tim@localhost ~]$ uname -rsvp
Linux 3.9.10-100.fc17.x86_64 #1 SMP Sun Jul 14 01:31:27 UTC 2013 x86_64
Boilerplate: All mail to my mailbox is automatically deleted, there is
no point trying to privately email me, I only get to see the messages
posted to the mailing list.
Lucky for you I typed this, you'd never be able to read my handwriting.
3:20 p.m.
New subject: Is there a way to specify that boot should be deterministic
instead of using threads?
On Fri, Jan 06, 2017 at 12:57:00PM -0700, stan wrote:
Kernel booting used to be deterministic. That is, if you booted the
kernel 10 times, every one of them would be exactly the same. Then,
several years ago, the kernel boot process was made threaded to speed
it up. This means that every boot is slightly different.
Do you mean the kernel, or do you mean the *system* boot process? I
have no idea about the former, but for the later, the answer is
no. It might be *possible* to add a "can you find a serialized ordering
for this target?" feature to systemd, but no one has.
--
Matthew Miller
<mattdm(a)fedoraproject.org>
Fedora Project Leader
5:05 p.m.
New subject: Is there a way to specify that boot should be deterministic
instead of using threads?
On Fri, 6 Jan 2017 16:20:50 -0500
Matthew Miller <mattdm(a)fedoraproject.org> wrote:
On Fri, Jan 06, 2017 at 12:57:00PM -0700, stan wrote:
> Kernel booting used to be deterministic. That is, if you booted the
> kernel 10 times, every one of them would be exactly the same. Then,
> several years ago, the kernel boot process was made threaded to
> speed it up. This means that every boot is slightly different.
Do you mean the kernel, or do you mean the *system* boot process? I
have no idea about the former, but for the later, the answer is
no. It might be *possible* to add a "can you find a serialized
ordering for this target?" feature to systemd, but no one has.
I think I meant system. But both the kernel and systemd seem to use
multi-threading intrinsically.
I think systemd has to have some sort of task ordering, or it couldn't
function, since there *are* ordering dependencies during boot. They
are probably just fuzzy, though. And there are things like the wait
on network task to try to protect against out of order start.
Thanks.
6:02 p.m.
New subject: Is there a way to specify that boot should be deterministic
instead of using threads?
On Fri, 2017-01-06 at 16:05 -0700, stan wrote:
I think systemd has to have some sort of task ordering, or it
couldn't
function, since there *are* ordering dependencies during boot. They
are probably just fuzzy, though. And there are things like the wait
on network task to try to protect against out of order start.
I doubt that the dependencies are "fuzzy". What would that even mean?
AFAIK they define a partial order and each process just executes as
soon as it can. That's why the physical order can vary.
You want a way for the partial ordering to be converted to a fixed
total ordering under control of some option, but I don't know if that's
possible without rewriting a lot of systemd rules.
poc
6:40 p.m.
New subject: Is there a way to specify that boot should be deterministic
instead of using threads?
On Sat, 07 Jan 2017 00:02:32 +0000
Patrick O'Callaghan <pocallaghan(a)gmail.com> wrote:
I doubt that the dependencies are "fuzzy". What would that
even mean?
AFAIK they define a partial order and each process just executes as
soon as it can. That's why the physical order can vary.
You just defined what I meant by 'fuzzy', much better than I did.
You want a way for the partial ordering to be converted to a fixed
total ordering under control of some option, but I don't know if
that's possible without rewriting a lot of systemd rules.
I don't want to go there. So, I'll find an alternative.
Thanks.
5:26 a.m.
New subject: Is there a way to specify that boot should be deterministic
instead of using threads?
On Fri, 2017-01-06 at 17:40 -0700, stan wrote:
> I doubt that the dependencies are "fuzzy". What would
that even mean?
> AFAIK they define a partial order and each process just executes as
> soon as it can. That's why the physical order can vary.
You just defined what I meant by 'fuzzy', much better than I did.
I simply meant that the dependencies are not fuzzy (they don't say e.g.
"wait till foo is nearly ready"), but that the order of execution is
not completely defined. No matter.
poc
2670
days inactive
2673
days old
11 comments
6 participants
participants (6)
-
Joe Zeff -
Justin Moore -
Matthew Miller -
Patrick O'Callaghan -
stan -
Tim