On 04/02/2013 02:47 PM, Mike Kolesnik wrote:
------------------------------------------------------------------------
On 03/27/2013 05:48 PM, Mike Kolesnik wrote:
----- Original Message -----
On 03/20/2013 08:20 PM, Yair Zaslavsky wrote:
----- Original Message -----
From: "Shireesh Anjal"<san...@redhat.com>
To: "Mike Kolesnik"<mkole...@redhat.com>
Cc:engine-devel@ovirt.org
Sent: Wednesday, March 20, 2013 4:47:08 PM
Subject: Re: [Engine-devel] FeatureSupported and
compatibility
versions
On 03/18/2013 01:11 PM, Shireesh Anjal wrote:
On 03/18/2013 12:59 PM, Mike Kolesnik wrote:
----- Original Message -----
Hi all,
The current mechanism in oVirt to check whether
a feature is
supported
in a particular compatibility version is to use
the
FeatureSupported
class. e.g.
FeatureSupported.networkLinking(getVm().getVdsGroupCompatibilityVersion())
Checks whether the "network linking" feature is
supported for
the
the
VM's cluster compatibility version. This
internally checks
whether
the
value of the corresponding config
(NetworkLinkingSupported) for
the
given compatibility version is true/false.
I'm not sure if this is a good idea, since a
feature is
typically
supported "from" a particular version. E.g.
Gluster support was
introduced in 3.1, and it continues to be
available in all
subsequent
versions. So I see no point in adding
configuration for every
version
indicating whether the feature is supported in
that version or
not. I
suggest to use either of the following options:
You can "merge" the configs into a single config
when older
versions
go out of the supported versions for the system.
i.e. in 4.0 you can have upgrade script that merges
all
GlusterFeatureSupported to one entry instead of
several.
Why are we even storing this information in config? Is this
something
that can be "configured" at customer site?
As previously explained (but off list :) ) , Config gives you
the
ability to have a cachable "map" of entry (i.e - "feature name")
per version and value.
I guess it was convinient for the developers to use that.
I also mentioned that customers/oVirt users should config the
entries of vdc_options using engine-config tool only.
Not all entries are exposed via engine-config.properties (and
no,
not just "is feature supported" entries are hidden).
1) Instead of using a boolean config for each
version, use a
single
string config that indicates the "supported
from" version e.g.
GlusterSupportedFrom = 3.1. There could be rare
cases where a
feature,
for some reason, is removed in some release. In
such cases, we
could
use
one additional config for the "supported to"
version.
2) Continue with the boolean approach, but do
not have entries
for
every
version; rather make use of the "default value"
for majority of
cases,
and add the explicit version mapping for the
minority e.g.
GlusterSupported = true by default, and false
in case of 3.0
(only
one
config required for 3.0)
I'm not sure why we would want to complicate this
simple
mechanism?
Is there much to gain?
I think option 1 suggested above is simpler - to
implement as
well
as
to understand.
Let me give you an example of why I don't like current
mechanism.
I
introduce a version check for a feature that was
introduced in
3.1.
I'm being asked now to add three entries in config
3.0 - false
3.1 - true
3.2 - true
It will also mean that when 3.3 goes out, someone has
to make
sure
that another entry is added for 3.3-true. I think it is
not
logical
as
well as scalable if you have more versions. And it
sounds far
more
complex (to maintain) than just having
<Feature>SupportedFrom = 3.1
So I would like to know if there are any objections to
my
proposal.
I
intend to use this for at least the gluster related
features.
I've sent a patch (http://gerrit.ovirt.org/12970) with following
changes:
1) Introduced CompatibilityUtils that provides utility methods for
checking if a given feature is supported in the config. One method
to
check based on boolean values (as is being done today for virt
features), and nother to check based on a range (from, to) which I
would
like to use for gluster features.
2) Renamed FeatureSupported to VirtFeatureSupported, and made it use
the
first utility method from CompatibilityUtils
3) Introduced GlusterFeatureSupported for gluster features, which
uses
the second utility method from CompatibilityUtils
Key advantage here is that
- we don't have to touch any virt specifc source for adding
compatibility checks for gluster features
- virt features continue to use the existing boolean config check
Any comments / suggestions / reviews will be highly appreciated :)
I think splitting to two classes is OK, but the underlying mechanism
IMO should be as Omer suggested:
Use the default value from the java config file, and if in the DB there
is a version specific value then use it for that version only.
Review comments here are on the contrary:
http://gerrit.ovirt.org/#/c/12970/5/backend/manager/dbscripts/upgrade/pre_upgrade/0000_config.sql
The comment in the review simply states that the mechanism is probably
broken, not that that's the way it has to be done.
The comment explicitly asks me to add entries for every version. What
you have looked at is my response to this comment, which suggests that
the current mechanism is not great. In fact, what I had done in
patch-set 5 is exactly what you are suggesting : true as default value
and explicit entries in config for the "false" values. But it was not
accepted.
I don't think "From, To, etc" is a good design, it's not a standard way
and is very restrictive.
Can you please explain in what way is it restrictive?
Also, what is the "etc" you are referring to?
What if for certain version it is not supported, you add "except"? Or
do you specify 2 ranges?
Starting to add from/to creates a limited design of one range, which
would be difficult to tune if necessary.
Really? Does someone really think that there will be a feature that will
be supported in multiple different ranges of versions? I see zero
possibility for this. I would love to see some +1s on this concept
before I can accept this argument.
I think the design generally for config values is very simple and
suits us well - use the default value, unless a specific version is
configured differently.
I think the current design is wrong. A feature gets supported "from" a
particular version, and that's all that is required in most of the
cases. Expecting developers to add version-by-version mapping for
features is bad. The "to" part in my patch is just to handle rare cases,
if at all they come up. I'm willing to even remove that if such a case
doesn't exist today.
Also, even though I have followed it for the sake of consistency, I
don't think these values need to be stored in the config (db) at all.
Only explanation I've got for it is that it was probably 'convenient'
for developers to use the config mechanism. I'm for having this check
purely in code in a central place, and not the config (db).
This way you can specify the feature is supported, and disable it for
specific versions.
So one has to look at both code (FeatureSupported) as well as db
(config) to get an idea of what versions the feature is supported in.
Not great.
I think this direction gives us the flexibility that we would like to
have.
Currently it doesn't work that way, but I think it's not impossible to
change, and more worthwhile than introducing a new mechanism.
I disagree, and would like to use the "supported from" mechanism at
least for gluster features.
Thoughts?
Regards,
Shireesh
_______________________________________________
Engine-devel mailing list
Engine-devel@ovirt.org
http://lists.ovirt.org/mailman/listinfo/engine-devel
