cloudstack-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Andrija Panic <andrija.pa...@gmail.com>
Subject Re: VPC's VR missing public NIC eth1
Date Fri, 30 May 2014 11:48:13 GMT
Hi Joris,

just to be sure - you want me to capture the log from the moment I reboot
router - or you want me to stop it, then start capturing log, and start it
(and continue capture untill ethnull errors inside VR) ?

Thanks,


On 30 May 2014 13:39, Joris van Lieshout <JvanLieshout@schubergphilis.com>
wrote:

> Hi Andrija,
>
> Thanks for the answers. In deed your situation is different so PV/HVM is
> not the issue.
>
> When reading back the log output you have provided I noted that the VR
> messages log indicates that it's waiting for ethnull to be up. This raises
> the question where null was introduced instead of 1. The ACS management
> log output you send was, what I think, later down the road where ACS gives
> up trying to wait for the VR to come up. If you would capture the
> job-executor in the management log from startCommand till the exception,
> do you see anywhere a mention of ethnull? You might need to reed into the
> DirectAgent executing the startCommand to find a clue. The thing is that I
> only have experience with XS based environment so I cannot point you to
> the exact output to look for. On XS, at least, it is
> "[c.c.h.x.r.CitrixResourceBase] (DirectAgent-351:ctx-4a51bb9e) Created a
> vif e4c362bd-764b-f651-dc9a-1abd5cb33c43 on 1"
>
> Kind regards,
> Joris van Lieshout
>
> Schuberg Philis
>
>
>
>
>
> On 30/05/14 10:48, "Andrija Panic" <andrija.panic@gmail.com> wrote:
>
> >Hi Deen,
> >no, in DB there is field "vlan_id" with value "untagged" - that
> >"vlan://untagged" is shown from ACS gui, and is used in API call (or
> >better
> >said commands that are seen in management server logs).
> >
> >Best,
> >Andrija
> >
> >
> >On 30 May 2014 10:37, Daan Hoogland <daan.hoogland@gmail.com> wrote:
> >
> >> Andrija,
> >>
> >> Do not just assign a second net vlan://500 You have one like that and
> >> you don't want conflicting nets using the same vlan. I am wondering
> >> why 'untagged' comes out as 'vlan://untagged'. I think that is the
> >> bug. Did you find the string 'vlan://untagged' in your db?
> >>
> >> On Fri, May 30, 2014 at 10:20 AM, Andrija Panic
> >><andrija.panic@gmail.com>
> >> wrote:
> >> > Hi Joris,
> >> >
> >> > thank you for taking time to address this issue :)
> >> >
> >> > So...:
> >> >
> >> > - I'm on KVM (stock CentOS 6.2 patched by Inktank for CEPH support),
> >>OS
> >> is
> >> > Centos 6.5, libvirt 1.2.3 compiled.
> >> > - ACS 4.3 having problems, ACS 4.2.1 was fine
> >> > - not XS, so I guess no answers for this part :)
> >> > - guest_os_id is 184 = Debian 7 x64
> >> > - SVM = systemvm-kvm-4.3 = os type 184 = Debian 7 x64
> >> >
> >> > This worked previously on 4.2.1 = template was ofcourse
> >>systemvm-kvm-4.2
> >> -
> >> > but that was also Debian 7 x64 type... so this should not be the
> >>issues
> >> > (guest not supported by host...)
> >> >
> >> > The only thing that might be out of "standard" = all SVMs are on CEPH
> >>-
> >> > there are official docs on altering database to make some new System
> >> > Offering as default for SSVM and CPVM - what I did, I also have done
> >>same
> >> > config in DB, to make VR use another System Offering as default -
> >>which
> >> is
> >> > NOT explained in the docs - you could use "Change Offering..." button
> >>on
> >> > exiting, shutdown VR to change it per docs...
> >> > But still this worked all fine on 4.2.1...
> >> >
> >> > - regarding /var/cache/cloud/cmdline the content is folowing at the
> >> moment
> >> > root@r-801-VM:~# cat /var/cache/cloud/cmdline
> >> > vpccidr=10.0.0.0/8 domain=cscloud.internal dns1=8.8.8.8 dns2=
> >> template=domP
> >> > name=r-801-VM eth0ip=169.254.0.75 eth0mask=255.255.0.0 type=vpcrouter
> >> > disable_rp_filter=true
> >> >
> >> > Also please note that only eth1 does not have IP info, eth0 (control
> >> > 169.xxx) and all other eh2 and up that are used for Tiers get IP info
> >> fine.
> >> > I could also manually add IP for eth1 (public NIC) and start ifup
> >>eth1 -
> >> > and it works fine, but adding new IP Port Forwarding etc does not
> >>work...
> >> >
> >> > Daan or somebody said it could be realted to my "Public" network (in
> >>the
> >> > Zones, Physical Network, eth1 listing) is NOT tagged
> >>(vlan://untagged)...
> >> > Interestingly the only VR that does work fine is the VR used in Shared
> >> > network, but that VR is using IP from Guest IP range (also efectively
> >> > public IPs but on vlan 500)
> >> >
> >> > I was instructed to try to change Public IP range from untagged to
> >>vlan
> >> > 500, but I'm not sure how to do this, if there is any way at all
> >>(editing
> >> > "vlan" table and changing to vlan 500 does not work, after rebooting
> >>VR
> >> > from ACS gui).
> >> >
> >> > :)
> >> >
> >> > So, not sure what is roughly expected date for 4.4, but right now, I'm
> >> > pretty stuck with a big problem of all VPC not operational at all...
> >> >
> >> > Thanks,
> >> >
> >> >
> >> >
> >> >
> >> >
> >> >
> >> >
> >> >
> >> >
> >> >
> >> >
> >> >
> >> >
> >> > On 30 May 2014 08:27, Joris van Lieshout <
> >> JvanLieshout@schubergphilis.com>
> >> > wrote:
> >> >
> >> >> Hi Andrija,
> >> >>
> >> >> Daan asked me to have a look at this as well. Looking at you issue
I
> >> >> recall having seen something similar. Back then when upgrading 4.2.1
> >>to
> >> >> 4.3 I though it had to do with out own custom build svm template.
> >> >> Let me fire off some questions before explaining what the cause was
> >>in
> >> our
> >> >> case. :)
> >> >>
> >> >> - what hypervisor (and version) are you using?
> >> >> - if XS, is the new VR a para-virtualised instance (PV) or hardware
> >> >> assisted (HVM)? Do a "xe vm-param-list" on the VR uuid and check that
> >> >> param PV-args is set and HVM-boot-policy is unset.
> >> >> - what is the OS type of the VR in ACS (guest_os_id in vm_instance
> >>table
> >> >> and match with table guest_os)
> >> >> - what is the OS type of the SVM template?
> >> >>
> >> >> Now for the explaining. :)
> >> >> In our case the OS type of the new template was not supported on the
> >> >> XenServer version we are running. Therefore the VR was started by XS
> >>as
> >> a
> >> >> HVM guest. System vms on XS rely on the arguments passed to them in
> >>the
> >> >> PV-args param (ends up on the guest in /var/cache/cloud/cmdline
> >>which in
> >> >> turn is used by cloud-early-config) in order to work. cmdline
> >>contains
> >> the
> >> >> NIC configuration information.
> >> >> So, long story short, if a VR gets started as a HVM it will not get
> >>the
> >> >> information needed to configure it's NICs.
> >> >>
> >> >> Workaround
> >> >> We corrected the os_type_id in the DB (yes I know editing the DB is
> >> >> something you usually don't want but there is no other way in this
> >>case)
> >> >> of the existing VR's and of the systemvmtemplate to something
> >>supported
> >> by
> >> >> XenServer.
> >> >>
> >> >> Kind regards,
> >> >> Joris van Lieshout
> >> >>
> >> >> Schuberg Philis
> >> >>
> >> >>
> >> >>
> >> >>
> >> >> On 29/05/14 12:18, "Andrija Panic" <andrija.panic@gmail.com>
wrote:
> >> >>
> >> >> >They are 2 traffic types on 1 physical net (that is both tagged
vlan
> >> 500,
> >> >> >and untagged packets travel over same KVM bridge, and over eth1
to
> >> outside
> >> >> >world)...
> >> >> >
> >> >> >
> >> >> >On 29 May 2014 12:04, Daan Hoogland <daan.hoogland@gmail.com>
> wrote:
> >> >> >
> >> >> >> Are these two traffic types in one physical net? or two physical
> >>nets
> >> >> >> on the same interface (seems wrong).
> >> >> >>
> >> >> >> On Thu, May 29, 2014 at 11:35 AM, Jayapal Reddy Uradi
> >> >> >> <jayapalreddy.uradi@citrix.com> wrote:
> >> >> >> > I don't think editing DB table will work.
> >> >> >> >
> >> >> >> > -Jayapal
> >> >> >> > On 29-May-2014, at 2:52 PM, Andrija Panic
> >><andrija.panic@gmail.com
> >> >
> >> >> >> wrote:
> >> >> >> >
> >> >> >> >> It's like this:
> >> >> >> >>
> >> >> >> >> I have public subnet /24.
> >> >> >> >>
> >> >> >> >> half is dedicated for Guest traffic (vlan 500) and
the second
> >> half is
> >> >> >> >> dedicated to Public traffic/network (no vlan tags,
that is
> >> untagged
> >> >> >> packets)
> >> >> >> >>
> >> >> >> >> Both vlan500 and untagged packets travel over physical
eth1
> >> >> >>interface on
> >> >> >> >> hypervisors and can reach Internet.
> >> >> >> >>
> >> >> >> >> Thanks,
> >> >> >> >>
> >> >> >> >>
> >> >> >> >> On 29 May 2014 11:06, Daan Hoogland <daan.hoogland@gmail.com>
> >> wrote:
> >> >> >> >>
> >> >> >> >>> On Thu, May 29, 2014 at 10:57 AM, Andrija Panic
<
> >> >> >> andrija.panic@gmail.com>
> >> >> >> >>> wrote:
> >> >> >> >>>> 500
> >> >> >> >>>
> >> >> >> >>>
> >> >> >> >>> is 500 the vlan of your guestnetwork or your
physical network?
> >> You
> >> >> >> >>> wouldn't want to have two nets with vlan 500!
> >> >> >> >>>
> >> >> >> >>> --
> >> >> >> >>> Daan
> >> >> >> >>>
> >> >> >> >>
> >> >> >> >>
> >> >> >> >>
> >> >> >> >> --
> >> >> >> >>
> >> >> >> >> Andrija Panić
> >> >> >> >> --------------------------------------
> >> >> >> >>  http://admintweets.com
> >> >> >> >> --------------------------------------
> >> >> >> >
> >> >> >>
> >> >> >>
> >> >> >>
> >> >> >> --
> >> >> >> Daan
> >> >> >>
> >> >> >
> >> >> >
> >> >> >
> >> >> >--
> >> >> >
> >> >> >Andrija Panić
> >> >> >--------------------------------------
> >> >> >  http://admintweets.com
> >> >> >--------------------------------------
> >> >>
> >> >>
> >> >
> >> >
> >> > --
> >> >
> >> > Andrija Panić
> >> > --------------------------------------
> >> >   http://admintweets.com
> >> > --------------------------------------
> >>
> >>
> >>
> >> --
> >> Daan
> >>
> >
> >
> >
> >--
> >
> >Andrija Panić
> >--------------------------------------
> >  http://admintweets.com
> >--------------------------------------
>
>


-- 

Andrija Panić
--------------------------------------
  http://admintweets.com
--------------------------------------

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message