[openstack-dev] [Fuel] [ironic] [inspector] Rewriting nailgun agent on Python proposal

Jim Rollenhagen jim at jimrollenhagen.com
Fri Mar 18 15:03:46 UTC 2016


On Fri, Mar 18, 2016 at 05:26:13PM +0300, Evgeniy L wrote:
> On Thu, Mar 17, 2016 at 3:16 PM, Dmitry Tantsur <dtantsur at redhat.com> wrote:
> 
> > On 03/16/2016 01:39 PM, Evgeniy L wrote:
> >
> >> Hi Dmitry,
> >>
> >> I can try to provide you description on what current Nailgun agent is,
> >> and what are potential requirements we may need from HW discovery system.
> >>
> >> Nailgun agent is a one-file Ruby script [0] which is periodically run
> >> under cron. It collects information about HW using ohai [1], plus it
> >> does custom parsing, filtration, retrieval of HW information. After the
> >> information is collected, it is sent to Nailgun, that is how node gets
> >> discovered in Fuel.
> >>
> >
> > Quick clarification: does it run on user instances? or does it run on
> > hardware while it's still not deployed to? The former is something that
> > Ironic tries not to do. There is an interest in the latter.
> 
> 
> Both, on user instances (with deployed OpenStack) and on instances which
> are not deployed and in bootstrap.
> What are the reasons Ironic tries not to do that (running HW discovery on
> deployed node)?

Well, there's a number of reasons. Ironic is not meant only for an
"undercloud" (deploying OpenStack on ironic instances). There are both
public and private cloud deployments of ironic in production today, that
make bare metal instances available to users of the cloud. Those users
may not want an agent running inside their instance, and more
importantly, the operators of those clouds may not want to expose the
ironic or inspector APIs to their users.

I'm not sure ironic should say "no, that isn't allowed" but at a minimum
it would need to be opt-in behavior.

> 
> 
> >
> >
> >> To summarise entire process:
> >> 1. After Fuel master node is installed, user restarts the nodes and they
> >> get booted via PXE with bootstrap image.
> >> 2. Inside of bootstrap image Nailgun agent is configured and installed.
> >> 3. Cron runs Nailgun agent.
> >> 3. Information is collected by Nailgun agent.
> >> 4. Information is sent to Nailgun.
> >> 5. Nailgun creates new node, for which user using UI can define
> >> partitioning schema and networks allocation.
> >> 6. After that provisioning/deployment can be run.
> >>
> >
> > So it looks quite similar to ironic-inspector + IPA, except introspection
> > runs once. Rerunning it would not be impossible to implement, though it
> > will require some changes to inspector, so that it can accept "updates" to
> > a node after the introspection is finished.
> >
> >
> >> Every time Nailgun agent sends a request, we submit information about
> >> the time last request from agent was done, if there was no request for
> >> time N, we mark the node as offline.
> >>
> >
> > This is similar to IPA heartbeating, I guess.
> >
> >
> >> With current implementation we have several problems (not all of them
> >> should be addressed by HW discovery system only):
> >>
> >> 1. A lot of things are hardcoded on the agent's side, which does
> >> additional filtration based on pre-hardcoded parameters [2], the less
> >> hardcoded logic in agent we have the easier it's to do upgrades and
> >> deliver fixes (upgrade one service is simpler than hundreds of agents).
> >>
> >
> > Oh, I hear it. In the inspector world we are moving away from processing
> > things on the ramdisk side exactly for this reason: it's too hard to change.
> >
> >
> >> 2. In order to get additional HW information user has to continue
> >> hardcoding it right in Ruby code, as the result, there is no way for
> >> Fuel plugin [3], to get additional hardware specific information, we
> >> need data-driven mechanism to be able to describe, what/how/where
> >> information to get from the node.
> >>
> >
> > Hmm, interesting. Right now we have a plugin mechanism for the ramdisk. We
> > also have a plugin (extra-hardware) trying to collect as much information
> > as it's feasible (based on https://github.com/redhat-cip/hardware).
> >
> 
> Could you please provide a link where I can learn more on plugin mechanism
> for ramdisk?

When IPA does inspection, it sends the inventory as reported by the
hardware managers. When building a ramdisk, you can include out-of-tree
hardware managers, and each hardware manager is called to fetch
inventory.

Docs: http://docs.openstack.org/developer/ironic-python-agent/#hardware-inventory
Example out-of-tree hardware managers:
https://github.com/openstack/proliantutils/tree/master/proliantutils/ipa_hw_manager
https://github.com/rackerlabs/onmetal-ironic-hardware-manager

> 
> 
> >
> > On the other side, there is ongoing work to have an ansible-based deploy
> > ramdisk in Ironic, maybe inspector could benefit from it too. Didn't think
> > about it yet, would be interesting to discuss on the summit.
> 
> 
> And here, I would appreciate if you have any link to get more context (I
> was able to find only playbook for Ironic installation).
> In Fuel we had an idea to implement tasks (abstract from specific
> deployment tool) to make configuration and get information about specific
> hardware.

The spec is in review, from some Mirantis folks in fact:
https://review.openstack.org/#/c/241946/

> 
> 
> >
> >
> >
> >> 3. Hardware gets changed, we have cases when NICs, HDDs, Motherboards
> >> are replaced/removed/added, as the result we should have a tool which
> >> would allow us to see these changes and when they were performed, based
> >> on that we want to be able to notify the user and provide suggestions
> >> how to proceed with these changes.
> >>
> >
> > This could probably done with a new ironic-inspector plugin.
> >
> >
> >> 4. With 3rd real-world cases, we have a problem of node identification,
> >> when HW gets changed and automatic matching doesn't happen (when we
> >> cannot say for sure that this is the node which we've already had), user
> >> should be able to say, that new node X is in fact offline node Y.
> >>
> >
> > Very good question. Right now Inspector is using either BMC IP address or
> > MAC's.
> >
> >
> >> 5. Different source of HW information, we want to have a system which
> >> would allow us to have hardware discovery from IPMI, CSV file, Cobbler,
> >> CMDB, etc at the same time.
> >>
> >
> > Not sure something like that should live within Ironic to be honest. Also
> > worth discussing in details.
> >
> >
> >> 6. Not only hardware gets changed, but operating system (with kernel
> >> versions), for example when we used CentOS as a bootstrap (in bootstrap
> >> we do provisioning/partitioning + initial configuration) and Ubuntu for
> >> running OpenStack, we were getting wide range of weird problems, from
> >> NICs renaming to Disks' ids duplication and deduplication. There should
> >> be a way to track these problems (3rd item), and we should be able to
> >> add operating system specific system to get HW information.
> >>
> >> 7. Cron + Agent based mechanism to define if node is offline is not the
> >> best, since it adds race conditions and in fact it only says if there is
> >> connectivity between Nailgun and Nailgun agent.
> >>
> >
> > We are thinking about using some DLM for that.. No specific plans though,
> > again a topic for the summit.
> >
> >
> >> Will be glad to answer any questions you have, if there are any.

Is the fuel team having a summit session of some sort about integrating
with ironic better? I'd be happy to come to that if it can be scheduled
at a time that ironic doesn't have a session. Otherwise maybe we can
catch up on Friday or something.

I'm glad to see Fuel wanting to integrate better with Ironic.

// jim

> >>
> >> Thanks,
> >>
> >> [0] https://github.com/openstack/fuel-nailgun-agent/blob/master/agent
> >> [1] https://docs.chef.io/ohai.html
> >> [2]
> >> https://github.com/openstack/fuel-nailgun-agent/blob/master/agent#L46-L51
> >> [3] https://wiki.openstack.org/wiki/Fuel/Plugins
> >>
> >>
> >> On Wed, Mar 16, 2016 at 1:39 PM, Dmitry Tantsur <dtantsur at redhat.com
> >> <mailto:dtantsur at redhat.com>> wrote:
> >>
> >>     On 03/15/2016 01:53 PM, Serge Kovaleff wrote:
> >>
> >>         Dear All,
> >>
> >>         Let's compare functional abilities of both solutions.
> >>
> >>         Till the recent Mitaka release Ironic-inspector had only
> >>         Introspection
> >>         ability.
> >>
> >>         Discovery part is proposed and implemented by Anton Arefiev. We
> >>         should
> >>         align expectations and current and future functionality.
> >>
> >>         Adding Tags to attract the Inspector community.
> >>
> >>
> >>     Hi!
> >>
> >>     It would be great to see what we can do to fit the nailgun use case.
> >>     Unfortunately, I don't know much about it right now. What are you
> >>     missing?
> >>
> >>
> >>         Cheers,
> >>         Serge Kovaleff
> >>         http://www.mirantis.com <http://www.mirantis.com/>
> >>         cell: +38 (063) 83-155-70
> >>
> >>         On Tue, Mar 15, 2016 at 2:07 PM, Alexander Saprykin
> >>         <asaprykin at mirantis.com <mailto:asaprykin at mirantis.com>
> >>         <mailto:asaprykin at mirantis.com <mailto:asaprykin at mirantis.com>>>
> >>         wrote:
> >>
> >>              Dear all,
> >>
> >>              Thank you for the opinions about this problem.
> >>
> >>              I would agree with Roman, that it is always better to reuse
> >>              solutions than re-inventing the wheel. We should investigate
> >>              possibility of using ironic-inspector and integrating it
> >>         into fuel.
> >>
> >>              Best regards,
> >>              Alexander Saprykin
> >>
> >>              2016-03-15 13:03 GMT+01:00 Sergii Golovatiuk
> >>              <sgolovatiuk at mirantis.com <mailto:sgolovatiuk at mirantis.com>
> >>         <mailto:sgolovatiuk at mirantis.com
> >>         <mailto:sgolovatiuk at mirantis.com>>>:
> >>
> >>                  My strong +1 to drop off nailgun-agent completely in
> >>         favour of
> >>                  ironic-inspector. Even taking into consideration we'lll
> >>         need to
> >>                  extend  ironic-inspector for fuel needs.
> >>
> >>                  --
> >>                  Best regards,
> >>                  Sergii Golovatiuk,
> >>                  Skype #golserge
> >>                  IRC #holser
> >>
> >>                  On Tue, Mar 15, 2016 at 11:06 AM, Roman Prykhodchenko
> >>                  <me at romcheg.me <mailto:me at romcheg.me>
> >>         <mailto:me at romcheg.me <mailto:me at romcheg.me>>> wrote:
> >>
> >>                      My opition on this is that we have too many
> >> re-invented
> >>                      wheels in Fuel and it’s better think about
> >>         replacing them
> >>                      with something we can re-use than re-inventing them
> >>         one more
> >>                      time.
> >>
> >>                      Let’s take a look at Ironic and try to figure out
> >>         how we can
> >>                      use its features for the same purpose.
> >>
> >>
> >>                      - romcheg
> >>                       > 15 бер. 2016 р. о 10:38 Neil Jerram
> >>                      <Neil.Jerram at metaswitch.com
> >>         <mailto:Neil.Jerram at metaswitch.com>
> >>                      <mailto:Neil.Jerram at metaswitch.com
> >>
> >>         <mailto:Neil.Jerram at metaswitch.com>>> написав(ла):
> >>
> >>                       >
> >>                       > On 15/03/16 07:11, Vladimir Kozhukalov wrote:
> >>                       >> Alexander,
> >>                       >>
> >>                       >> We have many other places where use Ruby
> >>         (astute, puppet
> >>                      custom types,
> >>                       >> etc.). I don't think it is a good reason to
> >>         re-write
> >>                      something just
> >>                       >> because it is written in Ruby. You are right
> >> about
> >>                      tests, about plugins,
> >>                       >> but let's look around. Ironic community has
> >> already
> >>                      invented discovery
> >>                       >> component (btw written in python) and I can't
> >>         see any
> >>                      reason why we
> >>                       >> should continue putting efforts in nailgun
> >>         agent and not
> >>                      try to switch
> >>                       >> to ironic-inspector.
> >>                       >
> >>                       > +1 in general terms.  It's strange to me that
> >>         there are
> >>                      so many
> >>                       > OpenStack deployment systems that each do each
> >>         piece of
> >>                      the puzzle in
> >>                       > their own way (Fuel, Foreman, MAAS/Juju etc.) -
> >>         and which
> >>                      also means
> >>                       > that I need substantial separate learning in
> >>         order to use
> >>                      all these
> >>                       > systems.  It would be great to see some
> >>         consolidation.
> >>                       >
> >>                       > Regards,
> >>                       >       Neil
> >>                       >
> >>                       >
> >>                       >
> >>
> >>
> >> __________________________________________________________________________
> >>                       > OpenStack Development Mailing List (not for
> >>         usage questions)
> >>                       > Unsubscribe:
> >>         OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> >>         <
> >> http://OpenStack-dev-request@lists.openstack.org?subject:unsubscribe>
> >>         <
> >> http://OpenStack-dev-request@lists.openstack.org?subject:unsubscribe>
> >>                       >
> >>         http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
> >>
> >>
> >>
> >>
> >> __________________________________________________________________________
> >>                      OpenStack Development Mailing List (not for usage
> >>         questions)
> >>                      Unsubscribe:
> >>         OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> >>         <
> >> http://OpenStack-dev-request@lists.openstack.org?subject:unsubscribe>
> >>         <
> >> http://OpenStack-dev-request@lists.openstack.org?subject:unsubscribe>
> >>         http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
> >>
> >>
> >>
> >>
> >>
> >> __________________________________________________________________________
> >>                  OpenStack Development Mailing List (not for usage
> >>         questions)
> >>                  Unsubscribe:
> >>         OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> >>         <
> >> http://OpenStack-dev-request@lists.openstack.org?subject:unsubscribe>
> >>
> >>         <
> >> http://OpenStack-dev-request@lists.openstack.org?subject:unsubscribe>
> >>         http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
> >>
> >>
> >>
> >>
> >>
> >> __________________________________________________________________________
> >>              OpenStack Development Mailing List (not for usage questions)
> >>              Unsubscribe:
> >>         OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> >>         <
> >> http://OpenStack-dev-request@lists.openstack.org?subject:unsubscribe>
> >>
> >>         <
> >> http://OpenStack-dev-request@lists.openstack.org?subject:unsubscribe>
> >>         http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
> >>
> >>
> >>
> >>
> >>
> >> __________________________________________________________________________
> >>         OpenStack Development Mailing List (not for usage questions)
> >>         Unsubscribe:
> >>         OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> >>         <
> >> http://OpenStack-dev-request@lists.openstack.org?subject:unsubscribe>
> >>         http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
> >>
> >>
> >>
> >>
> >> __________________________________________________________________________
> >>     OpenStack Development Mailing List (not for usage questions)
> >>     Unsubscribe:
> >>     OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> >>     <http://OpenStack-dev-request@lists.openstack.org?subject:unsubscribe
> >> >
> >>     http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
> >>
> >>
> >>
> >>
> >> __________________________________________________________________________
> >> OpenStack Development Mailing List (not for usage questions)
> >> Unsubscribe:
> >> OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> >> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
> >>
> >>
> >
> > __________________________________________________________________________
> > OpenStack Development Mailing List (not for usage questions)
> > Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> > http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
> >

> __________________________________________________________________________
> OpenStack Development Mailing List (not for usage questions)
> Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev




More information about the OpenStack-dev mailing list