<html>
  <head>
    <meta content="text/html; charset=ISO-8859-1"
      http-equiv="Content-Type">
  </head>
  <body text="#000000" bgcolor="#FFFFFF">
    <div class="moz-cite-prefix">Hi Mike,<br>
      <br>
      Le 08/10/2013 08:51, Mike Spreitzer a écrit :<br>
    </div>
    <blockquote
cite="mid:OF3EF1E2D0.9D3F9F95-ON85257BFE.0024C2D5-85257BFE.0025B310@us.ibm.com"
      type="cite">
      <meta http-equiv="Content-Type" content="text/html;
        charset=ISO-8859-1">
      <font face="sans-serif" size="2">On second thought, I should
        revise and
        extend my remarks.  This message supersedes my previous two
        replies.</font>
      <br>
      <br>
      <font face="sans-serif" size="2">Thanks.  I have a few questions.
         First, I am a bit stymied by the style of API documentation
        used
        in that document and many others: it shows the first line of an
        HTTP request
        but says nothing about all the other details.  I am sure some of
        those
        requests must have interesting bodies, but I am not always sure
        which ones
        have a body at all, let alone what goes in it.  I suspect there
        may
        be some headers that are important too.  Am I missing something?</font><font
        size="3">
        <br>
      </font><font face="sans-serif" size="2"><br>
        That draft says the VMs are created before the group.  Is there
        a
        way today to create a VM without scheduling it?  Is there a way
        to
        activate a resource that has already been scheduled but not
        activated?</font><font size="3">
      </font><font face="sans-serif" size="2">By "activate" I mean, for
        a VM instance for example, to start running it.  </font><font
        size="3"><br>
      </font><font face="sans-serif" size="2"><br>
        As I understand your draft, it lays out a three phase process
        for a client
        to follow: create resources without scheduling or activating
        them, then
        present the groups and policies to the service for joint
        scheduling, then
        activate the resources.  With regard to a given resource, things
        must
        happen in that order; between resources there is a little more
        flexibility.
         Activations are invoked by the client in an order that is
        consistent
        with (a) runtime dependencies that are mediated directly by the
        client
        (e.g., string slinging in the heat engine) and (b) the nature of
        the resources
        (for example, you  can not attach a volume to a VM instance
        until
        after both have been created).  Other than those considerations,
        the
        ordering and/or parallelism is a degree of freedom available to
        the client.
         Have I got this right?</font><font size="3"> <br>
      </font><font face="sans-serif" size="2"><br>
        Couldn't we simplify this into a two phase process: create
        groups and resources
        with scheduling, then activate the resources in an acceptable
        order?</font><font size="3">
        <br>
      </font></blockquote>
    <br>
    <font size="3"><font size="3"><font size="3">I c<font size="3">an't
            answ<font size="3">er for <font size="3">it, but as<font
                  size="3"> far as I can understand the dra<font
                    size="3">ft, there is no clear understand<font
                      size="3">ing that we <font size="3">have to<font
                          size="3"> po<font size="3">stpone the VM boot
                            *after* creating the <font size="3">Groups.<br>
                              <font size="3">As s<font size="3">ta<font
                                    size="3">ted in the document, <font
                                      size="3">there is a strong prere<font
                                        size="3">q<font size="3"> which
                                          is that all the resources
                                          mapped <font size="3">to the
                                            Group m<font size="3">ust <font
                                                size="3">have their o<font
                                                  size="3">wn uuids<font
                                                    size="3">, <font
                                                      size="3">but t<font
                                                        size="3">he<font
                                                          size="3">re is
                                                          no clear outs<font
                                                          size="3">tanding
                                                          tha<font
                                                          size="3">t it
                                                          should prevent
                                                          the VMs to
                                                          actual<font
                                                          size="3">ly
                                                          boot.<br>
                                                          <br>
                                                          <font size="3">At
                                                          the moment, <font
                                                          size="3">deferring
                                                          <font size="3">a
                                                          boot<font
                                                          size="3">able
                                                          state in Nova
                                                          is not <font
                                                          size="3">yet
                                                          impleme<font
                                                          size="3">nted
                                                          and that's
                                                          part of
                                                          Climate folks
                                                          to i<font
                                                          size="3">mplement
                                                          it<font
                                                          size="3">, <font
                                                          size="3">s<font
                                                          size="3">o I
                                                          can<font
                                                          size="3">'t
                                                          get your
                                                          point.<br>
                                                          <br>
                                                          <font size="3">-Sylvain<br>
                                                          <br>
                                                          </font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font>
    <blockquote
cite="mid:OF3EF1E2D0.9D3F9F95-ON85257BFE.0024C2D5-85257BFE.0025B310@us.ibm.com"
      type="cite"><font size="3">
      </font><font face="sans-serif" size="2"><br>
        FYI: my group is using Weaver as the software orchestration
        technique,
        so there are no runtime dependencies that are mediated directly
        by the
        client.  The client sees a very simple API: the client presents
        a
        definition of all the groups and resources, and the service
        first schedules
        it all then activates in an acceptable order.  (We already have
        something
        in OpenStack that can do resource invocations in an acceptable
        order, right?)
         Weaver is not the only software orchestration technique with
        this
        property.  The simplicity of this API is one reason I recommend
        software
        orchestration techniques that take dependency mediation out of
        the client's
        hands.  I hope that with coming work on HOT we can get OpenStack
        to
        this level of API simplicity.  But that struggle lies farther
        down
        the roadmap...</font><font size="3"> <br>
      </font>
      <br>
      <font face="sans-serif" size="2">I was wondering if you could
        explain
        why you included all those integer IDs; aren't the UUIDs
        sufficient?  Do
        you intend that clients will see/manipulate the integer IDs?</font>
      <br>
      <br>
      <font face="sans-serif" size="2">If I understand your UML
        correctly,
        an InstanceGroup owns its metadata but none of the other
        subsidiary objects
        introduced.  Why not?  If an InstanceGroup is deleted, shouldn't
        all those other subsidiary objects be deleted too?</font>
      <br>
      <font face="sans-serif" size="2"><br>
        Thanks,</font><font size="3"> </font><font face="sans-serif"
        size="2"><br>
        Mike</font><font size="3"> </font>
      <br>
      <br>
      <tt><font size="2">"Yathiraj Udupi (yudupi)"
          <a class="moz-txt-link-rfc2396E" href="mailto:yudupi@cisco.com"><yudupi@cisco.com></a>
          wrote on 10/07/2013 11:10:20 PM:<br>
          > Hi, </font></tt>
      <br>
      <tt><font size="2">> <br>
          > Based on the discussions we have had in the past few
          scheduler sub-<br>
          > team meetings,  I am sharing a document that proposes an
          updated
          <br>
          > Instance Group Model and API extension model. </font></tt>
      <br>
      <tt><font size="2">> This is a work-in-progress draft version,
          but
          sharing it for early feedback. </font></tt>
      <br>
      <tt><font size="2">> </font></tt><a moz-do-not-send="true"
href="https://docs.google.com/document/d/17OIiBoIavih-1y4zzK0oXyI66529f-7JTCVj-BcXURA/edit?usp=sharing"><tt><font
            size="2">https://docs.google.com/document/d/17OIiBoIavih-1y4zzK0oXyI66529f-7JTCVj-BcXURA/edit?usp=sharing</font></tt></a><tt><font
          size="2">
        </font></tt>
      <br>
      <tt><font size="2">> <br>
          > This model support generic instance types, where an
          instance can <br>
          > represent a virtual node of any resource type.  But in
          the context
          <br>
          > of Nova, an instance refers to the VM instance. </font></tt>
      <br>
      <tt><font size="2">> <br>
          > This builds on the existing proposal for Instance Group
          Extension
          as<br>
          > documented here in this blueprint:  <a class="moz-txt-link-freetext" href="https://">https://</a><br>
          >
          blueprints.launchpad.net/nova/+spec/instance-group-api-extension
        </font></tt>
      <br>
      <tt><font size="2">> <br>
          > Thanks,</font></tt>
      <br>
      <tt><font size="2">> Yathi. </font></tt>
      <br>
      <tt><font size="2">> <br>
          >   </font></tt>
      <br>
      <fieldset class="mimeAttachmentHeader"></fieldset>
      <br>
      <pre wrap="">_______________________________________________
OpenStack-dev mailing list
<a class="moz-txt-link-abbreviated" href="mailto:OpenStack-dev@lists.openstack.org">OpenStack-dev@lists.openstack.org</a>
<a class="moz-txt-link-freetext" href="http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev">http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev</a>
</pre>
    </blockquote>
    <br>
  </body>
</html>