4:26 p.m.
--Apple-Mail=_59A168E2-93D5-4CF3-B121-15F35FE387B0
Content-Transfer-Encoding: quoted-printable
Content-Type: text/plain;
charset=us-ascii
On 23 Jul 2017, at 11:58, Roy Golan <rgolan(a)redhat.com> wrote:
=20
On Wed, Jul 19, 2017 at 12:46 PM Francesco Romani <fromani(a)redhat.com =
<mailto:fromani@redhat.com>> wrote:
Hi all,
=20
=20
With libvirt 3.2.0 and onwards, it seems we have now the tools to =
solve
<https://bugzilla.redhat.com/show_bug.cgi?id=3D1181665>
=20
and eventually get rid of the disk polling we do. This change is
expected to have huge impact on performance, so I'm working on it.
=20
=20
I had plans for a comprehensive refactoring in this area, but looks =
like
a solution backportable for 4.1.z is appealing, so I
=20
started with this first, saving the refactoring (which I still very =
much
want) for later.
=20
=20
So, quick summary: libvirt >=3D 3.2.0 allows to set a threshold to any
node in the backing chain of each drive of a VM
=20
=
(https://libvirt.org/html/libvirt-libvirt-domain.html#virDomainSetBlockThr=
eshold =
<https://libvirt.org/html/libvirt-libvirt-domain.html#virDomainSetBlockThr=
eshold>),
and fire one event exactly once
=20
when that threshold is crossed. The event needs to be explicitely
rearmed after.
=20
This is exactly what we need to get rid of polling in the steady =
state,
so far so good.
=20
=20
The problem is: we can't use this for some important flows we have, =
and
which involve usage of disks not (yet) attached to a given VM.
=20
=20
=20
=20
Possibly affected flows:
=20
- live storage migration:
=20
we use flags =3D (libvirt.VIR_DOMAIN_BLOCK_COPY_SHALLOW |
libvirt.VIR_DOMAIN_BLOCK_COPY_REUSE_EXT |
VIR_DOMAIN_BLOCK_COPY_TRANSIENT_JOB)
=20
meaning that Vdsm is in charge of handling the volume
=20
- snapshots:
=20
we use snapFlags =3D =
(libvirt.VIR_DOMAIN_SNAPSHOT_CREATE_REUSE_EXT |
libvirt.VIR_DOMAIN_SNAPSHOT_CREATE_NO_METADATA)
=20
=20
(same meaning as above)
=20
- live merge: should be OK (according to a glance at the source and a
chat with Adam).
=20
=20
So looks like we will need to bridge this gap.
=20
=20
So we can still use the BLOCK_THRESHOLD event for steady state, and
avoid polling in the vast majority of the cases.
=20
With "steady state" I mean that the VM is running, with no
administration (snapshot, live merge, live storage migration...)
operation in progress.
=20
I think it is fair to assume that VMs are in this state the vast
majority of the time.
For the very important cases on which we cannot depend on events, we =
can
fall back to polling, but in a smarter way:
=20
instead of polling everything every 2s, let's just poll just the =
drives
involved in the ongoing operations.
=20
Those should be far less of the total amount of drives, and for a far
shorter time than today, so polling should be practical.
=20
Since the event fires once, we will need to rearm it only if the
operation is ongoing, and only just before to start it (both =
conditions
easy to check)
We can disable the polling on completion, or on error. This per se is
easy, but we will need a careful review of the flows, and perhaps some
safety nets in place.
=20
=20
Consider fusing polling and events into a single pipeline of events so =
they can
be used together. If a poll triggers an event (with =
distinguished origin)
then it all the handling is done in one place and it should be easy
to =
stop or start polling, or remove them totally.
=20
Anyway, should we miss to disable the polling, we will "just" have =
some
overhead.
=20
=20
On recovery, we will need to make sure to rearm all the relevant =
events,
but we can just plug in the recovery we must do already, so this =
should
be easy as well.
=20
=20
What is needed in order to 'rearm' it? is there an API to get the =
state
of event subscription?
If we lost an event how do we know to rearm it? is it idempotent to =
rearm?
=20
Remind me, do we extend a disk if the VM paused with out of space =
event?
=20
How will we handle 2 subsequent events if we didn't extend between =
them?
(expecting the extend to be async operation)
=20
So it seems to me this could fly and we can actually have the
performance benefits of events.
=20
=20
However, due to the fact that we need to review some existing and
delicate flows, I think we should still keep the current polling code
around for the next release.
=20
+1 =20
=20
I believe the best route is:
=20
1. offer the new event-based code for 4.2, keep the polling around.
Default to events for performance
=20
2. remove the polling completely in 4.3
=20
=20
Still wonder if removing them totally is good. The absence of the =
events should
be supervised somehow - like in today, a failure to poll =
getstats of a domain will result in a VM going unresponsive. Not the =
most accurate state but at least gives some visibility. So polling =
should cover us where events will fail. (similar to engine's vms =
monitoring)
It is a different case. With disk extensions there always is the =
fallback of actually hitting the 100%, pausing the VM, and triggering =
the extend anyway. So I do not think there is a need for another =
mechanism when event is missed (e.g. due to a vdsm restart).
=20
I'm currently working on the patches here:
=
https://gerrit.ovirt.org/#/q/status:open+project:vdsm+branch:master+topic:=
watermark-event-minimal =
<https://gerrit.ovirt.org/#/q/status:open+project:vdsm+branch:master+topic=
:watermark-event-minimal>
=20
=20
Even though the basics are in place, I don't think they are ready for
review yet.
=20
=20
Comments welcome, as usual.
=20
=20
--
Francesco Romani
Senior SW Eng., Virtualization R&D
Red Hat
IRC: fromani github: @fromanirh
=20
_______________________________________________
Devel mailing list
Devel(a)ovirt.org <mailto:Devel@ovirt.org>
http://lists.ovirt.org/mailman/listinfo/devel =
<http://lists.ovirt.org/mailman/listinfo/devel>
_______________________________________________
Devel mailing list
Devel(a)ovirt.org
http://lists.ovirt.org/mailman/listinfo/devel
--Apple-Mail=_59A168E2-93D5-4CF3-B121-15F35FE387B0
Content-Transfer-Encoding: quoted-printable
Content-Type: text/html;
charset=us-ascii
<html><head><meta http-equiv=3D"Content-Type"
content=3D"text/html =
charset=3Dus-ascii"></head><body style=3D"word-wrap: break-word; =
-webkit-nbsp-mode: space; -webkit-line-break: after-white-space;" =
class=3D""><br class=3D""><div><blockquote
type=3D"cite" class=3D""><div =
class=3D"">On 23 Jul 2017, at 11:58, Roy Golan <<a =
href=3D"mailto:rgolan@redhat.com"
class=3D"">rgolan(a)redhat.com</a>&gt; =
wrote:</div><br class=3D"Apple-interchange-newline"><div
class=3D""><div =
dir=3D"ltr" class=3D""><div
class=3D"gmail_quote"><div dir=3D"ltr" =
class=3D"">On Wed, Jul 19, 2017 at 12:46 PM Francesco Romani <<a =
href=3D"mailto:fromani@redhat.com" target=3D"_blank" =
class=3D"">fromani(a)redhat.com</a>&gt; wrote:<br =
class=3D""></div><blockquote class=3D"gmail_quote"
style=3D"margin:0 0 0 =
.8ex;border-left:1px #ccc solid;padding-left:1ex">Hi all,<br
class=3D"">
<br class=3D"">
<br class=3D"">
With libvirt 3.2.0 and onwards, it seems we have now the tools to =
solve<br class=3D"">
<a href=3D"https://bugzilla.redhat.com/show_bug.cgi?id=3D1181665" =
rel=3D"noreferrer" target=3D"_blank" =
class=3D"">https://bugzilla.redhat.com/show_bug.cgi?id=3D118...
=
class=3D"">
<br class=3D"">
and eventually get rid of the disk polling we do. This change is<br =
class=3D"">
expected to have huge impact on performance, so I'm working on it.<br =
class=3D"">
<br class=3D"">
<br class=3D"">
I had plans for a comprehensive refactoring in this area, but looks =
like<br class=3D"">
a solution backportable for 4.1.z is appealing, so I<br class=3D"">
<br class=3D"">
started with this first, saving the refactoring (which I still very =
much<br class=3D"">
want) for later.<br class=3D"">
<br class=3D"">
<br class=3D"">
So, quick summary: libvirt >=3D 3.2.0 allows to set a threshold to =
any<br class=3D"">
node in the backing chain of each drive of a VM<br class=3D"">
<br class=3D"">
(<a =
href=3D"https://libvirt.org/html/libvirt-libvirt-domain.html#virDoma...
lockThreshold" rel=3D"noreferrer" target=3D"_blank" =
class=3D"">https://libvirt.org/html/libvirt-libvirt-domain.h...
etBlockThreshold</a>),<br class=3D"">
and fire one event exactly once<br class=3D"">
<br class=3D"">
when that threshold is crossed. The event needs to be explicitely<br =
class=3D"">
rearmed after.<br class=3D"">
<br class=3D"">
This is exactly what we need to get rid of polling in the steady =
state,<br class=3D"">
so far so good.<br class=3D"">
<br class=3D"">
<br class=3D"">
The problem is: we can't use this for some important flows we have, =
and<br class=3D"">
which involve usage of disks not (yet) attached to a given VM.<br =
class=3D"">
<br class=3D"">
<br class=3D""></blockquote><div class=3D""><br
class=3D""></div><div =
class=3D""> <br
class=3D""></div><blockquote class=3D"gmail_quote" =
style=3D"margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
Possibly affected flows:<br class=3D"">
<br class=3D"">
- live storage migration:<br class=3D"">
<br class=3D"">
we use flags =3D =
(libvirt.VIR_DOMAIN_BLOCK_COPY_SHALLOW |<br class=3D"">
=
libvirt.VIR_DOMAIN_BLOCK_COPY_REUSE_EXT |<br class=3D"">
=
VIR_DOMAIN_BLOCK_COPY_TRANSIENT_JOB)<br class=3D"">
<br class=3D"">
meaning that Vdsm is in charge of handling the volume<br =
class=3D"">
<br class=3D"">
- snapshots:<br class=3D"">
<br class=3D"">
we use snapFlags =3D =
(libvirt.VIR_DOMAIN_SNAPSHOT_CREATE_REUSE_EXT |<br class=3D"">
=
libvirt.VIR_DOMAIN_SNAPSHOT_CREATE_NO_METADATA)<br class=3D"">
<br class=3D"">
<br class=3D"">
(same meaning as above)<br class=3D"">
<br class=3D"">
- live merge: should be OK (according to a glance at the source and a<br =
class=3D"">
chat with Adam).<br class=3D"">
<br class=3D"">
<br class=3D"">
So looks like we will need to bridge this gap.<br class=3D"">
<br class=3D"">
<br class=3D"">
So we can still use the BLOCK_THRESHOLD event for steady state, and<br =
class=3D"">
avoid polling in the vast majority of the cases.<br class=3D"">
<br class=3D"">
With "steady state" I mean that the VM is running, with no<br
class=3D"">
administration (snapshot, live merge, live storage migration...)<br =
class=3D"">
operation in progress.<br class=3D"">
<br class=3D"">
I think it is fair to assume that VMs are in this state the vast<br =
class=3D"">
majority of the time.<br class=3D"">
For the very important cases on which we cannot depend on events, we =
can<br class=3D"">
fall back to polling, but in a smarter way:<br class=3D"">
<br class=3D"">
instead of polling everything every 2s, let's just poll just the =
drives<br class=3D"">
involved in the ongoing operations.<br class=3D"">
<br class=3D"">
Those should be far less of the total amount of drives, and for a far<br =
class=3D"">
shorter time than today, so polling should be practical.<br class=3D"">
<br class=3D"">
Since the event fires once, we will need to rearm it only if the<br =
class=3D"">
operation is ongoing, and only just before to start it (both =
conditions<br class=3D"">
easy to check)<br class=3D"">
We can disable the polling on completion, or on error. This per se is<br =
class=3D"">
easy, but we will need a careful review of the flows, and perhaps =
some<br class=3D"">
safety nets in place.<br class=3D"">
<br class=3D""></blockquote><div class=3D""><br
class=3D""></div><div =
class=3D"">Consider fusing polling and events into a single pipeline of =
events so they can be used together. If a poll triggers an event (with =
distinguished origin)<br class=3D""></div><div
class=3D"">then it all =
the handling is done in one place and it should be easy to stop or start =
polling, or remove them totally.<br class=3D""></div><div =
class=3D""> </div><blockquote
class=3D"gmail_quote" style=3D"margin:0=
0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
Anyway, should we miss to disable the polling, we will "just" have =
some<br class=3D"">
overhead.<br class=3D"">
<br class=3D""></blockquote><div
class=3D""> <br =
class=3D""></div><blockquote class=3D"gmail_quote"
style=3D"margin:0 0 0 =
.8ex;border-left:1px #ccc solid;padding-left:1ex">
On recovery, we will need to make sure to rearm all the relevant =
events,<br class=3D"">
but we can just plug in the recovery we must do already, so this =
should<br class=3D"">
be easy as well.<br class=3D"">
<br class=3D""></blockquote><div
class=3D""> </div></div><div =
dir=3D"ltr" class=3D""><div
class=3D"gmail_quote"><div class=3D"">What =
is needed in order to 'rearm' it? is there an API to get the state of =
event subscription?<br class=3D""></div><div
class=3D"">If we lost an =
event how do we know to rearm it? is it idempotent to rearm?<br =
class=3D""></div><div class=3D""><br
class=3D"">Remind me, do we extend =
a disk if the VM paused with out of space event?<br class=3D""><br =
class=3D""></div><div class=3D"">How will we handle 2
subsequent events =
if we didn't extend between them? (expecting the extend to be async =
operation)<br class=3D""><br
class=3D""></div></div></div><div
dir=3D"ltr"=
class=3D""><div class=3D"gmail_quote"><blockquote
class=3D"gmail_quote" =
style=3D"margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
So it seems to me this could fly and we can actually have the<br =
class=3D"">
performance benefits of events.<br class=3D"">
<br class=3D""></blockquote><div class=3D""><br
=
class=3D""></div><blockquote class=3D"gmail_quote"
style=3D"margin:0 0 0 =
.8ex;border-left:1px #ccc solid;padding-left:1ex">
However, due to the fact that we need to review some existing and<br =
class=3D"">
delicate flows, I think we should still keep the current polling code<br =
class=3D"">
around for the next release.<br class=3D"">
<br class=3D""></blockquote><div
class=3D"">+1 <br =
class=3D""> <br
class=3D""></div><blockquote class=3D"gmail_quote" =
style=3D"margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
I believe the best route is:<br class=3D"">
<br class=3D"">
1. offer the new event-based code for 4.2, keep the polling around.<br =
class=3D"">
Default to events for performance<br class=3D"">
<br class=3D"">
2. remove the polling completely in 4.3<br class=3D"">
<br class=3D"">
<br class=3D""></blockquote><div class=3D"">Still
wonder if removing =
them totally is good. The absence of the events should be supervised =
somehow - like in today, a failure to poll getstats of a domain will =
result in a VM going unresponsive. Not the most accurate state but at =
least gives some visibility. So polling should cover us where events =
will fail. (similar to engine's vms monitoring)<br =
class=3D""></div></div></div></div></div></blockquote><div><br
=
class=3D""></div>It is a different case. With disk extensions there =
always is the fallback of actually hitting the 100%, pausing the VM, and =
triggering the extend anyway. So I do not think there is a need for =
another mechanism when event is missed (e.g. due to a vdsm =
restart).</div><div><br class=3D""><blockquote
type=3D"cite" =
class=3D""><div class=3D""><div dir=3D"ltr"
class=3D""><div dir=3D"ltr" =
class=3D""><div class=3D"gmail_quote"><div
class=3D""><br =
class=3D""></div><blockquote class=3D"gmail_quote"
style=3D"margin:0 0 0 =
.8ex;border-left:1px #ccc solid;padding-left:1ex">
I'm currently working on the patches here:<br class=3D"">
<a =
href=3D"https://gerrit.ovirt.org/#/q/status:open+project:vdsm+branch...
r+topic:watermark-event-minimal" rel=3D"noreferrer"
target=3D"_blank" =
class=3D"">https://gerrit.ovirt.org/#/q/status:open+project:...
ster+topic:watermark-event-minimal</a><br class=3D"">
<br class=3D"">
<br class=3D"">
Even though the basics are in place, I don't think they are ready for<br =
class=3D"">
review yet.<br class=3D"">
<br class=3D"">
<br class=3D"">
Comments welcome, as usual.<br class=3D"">
<br class=3D"">
<br class=3D"">
--<br class=3D"">
Francesco Romani<br class=3D"">
Senior SW Eng., Virtualization R&D<br class=3D"">
Red Hat<br class=3D"">
IRC: fromani github: @fromanirh<br class=3D"">
<br class=3D"">
_______________________________________________<br class=3D"">
Devel mailing list<br class=3D"">
<a href=3D"mailto:Devel@ovirt.org" target=3D"_blank" =
class=3D"">Devel(a)ovirt.org</a><br class=3D"">
<a href=3D"http://lists.ovirt.org/mailman/listinfo/devel" =
rel=3D"noreferrer" target=3D"_blank" =
class=3D"">http://lists.ovirt.org/mailman/listinfo/devel<... =
class=3D"">
</blockquote></div></div></div>
_______________________________________________<br class=3D"">Devel =
mailing list<br class=3D""><a href=3D"mailto:Devel@ovirt.org"
=
class=3D"">Devel(a)ovirt.org</a><br =
class=3D"">http://lists.ovirt.org/mailman/listinfo/devel<...
</div><br class=3D""></body></html>=
--Apple-Mail=_59A168E2-93D5-4CF3-B121-15F35FE387B0--