[Discussion] CentOS 7 KVM binaries

[Daniel Salvador <gu...@apache.org>]

Hi all, hope you are doing fine.

I started a thread (
https://lists.apache.org/thread/z7s0774n72v4o9dnl140wvm030bxovjd) on dev ML
regarding KVM binaries in CentOS7. However, I forgot to CC the user mailing
list. I think that it would be interesting if you (the users/operators of
ACS) could share your experiences with CentOS. Right below there is a copy
of the email:


---
Hi all, hope you are doing fine.

The following discussion emerged from PR #5297[¹].

It is a known fact that, regarding KVM functionalities, CentOS7 default's
QEMU binary is quite limited. This is due to the removal of some features
of KVM default binary in CentOS (like VM's volume live migration,
memory/CPU hotplug/hot-unplug, live disk snapshot, and so on); more
information can be found in CentOS forum's thread[²].

In my point of view, such limitations in the default QEMU binary in CentOS
make it unfeasible to build a cloud with CentOS and the default QEMU
binary, as operators lose a lot of useful/important operations or have to
go through workarounds, which cause VM's disruption (e.g. having to stop a
VM to migrate the volume between different storage pools, which triggers a
secondary storage usage). There is an alternative binary, "qemu-kvm-ev",
which supports more features than the default one. Probably, most people
using KVM with CentOS are using the "ev" binary (I might be wrong though,
however, that seems to be the case when looking at the users' list).

PR #5297[¹] ran into one of the CentOS7 default's QEMU binary limitations
(live disk snapshot). The easiest solution (and, IHMO, is the best option)
is to guide users to upgrade CentOS7 QEMU binary to "qemu-kvm-ev".

Further, it is important to mention that in our experience, it is not
possible to run a highly available cloud environment with CentOS7 and the
default binaries. In a cloud environment with thousands of VMs, sooner or
later the need to hotplug (increase) CPU/RAM, migrate volumes across
different storage pool tyles (such as iSCSI <> NFS), or some other type of
operation appear, and operators/final customers do not want service
disruption. Our customers, for instance, never shut down VMs for these
kinds of operations, and that is only possible because they are all using
KVM with Ubuntu now.

Moreover, CentOS7 is getting close to its EOL. Therefore, We do not think
that CloudStack should limit its features due to a dying operating system
that presents very limited features by default.

With that said, it would be interesting if dev/users that use CentOS7 could
share their experiences with the "qemu-kvm-ev" in this thread, so we can
decide which way to go. Or, users that only use the default binary, if they
are satisfied with it.

If almost no one is relying on default CentOS7 binaries, we could define as
a step in the documentation, that when using CentOS7 people must use the
"ev" binary. This would free us to evolve ACS more freely and avoid
headaches with workarounds to a limited operating system when there are
alternatives out there.


[¹] https://github.com/apache/cloudstack/pull/5297
[²] https://forums.centos.org/viewtopic.php?t=65618
---


Best regards,
Daniel Salvador