Choose timezone
Your profile timezone:
Ibex and AI HOWTO
→
Europe/Zurich
513/1-024 (CERN)
Attendance:
IT-PES: Alex Lossent, Vitor Gomes, Nacho Barrentos, Steve traylen, Maite Barroso, Manuel Guijarro, Nils Hoimyr, Peter Jones, John Evans, Evangelos Atlidakis, David Asbury, Akos Hencz
IT-ES: Pabo Saiz
IT-DI: Denise Heagerty
IT-DSS: Alex Iribarren
IT-CF: Veromique Lefebure
IT-GT: Andres Abad Rodriguez
IT-OIS: Jan Van Eldik, Tim Bell
ATLAS: Sergey Barnov, Alexey Buzykaev
CMS: Jorge Molina
LHCb: Joel Closier
LCD: Stephane Poss, Andre Sailer
Questions:
(PS) Pet VM type used in demo, but we said we support only cattle at the moment?
=> (NB) indeed. At the moment it makes no difference.
(PS) We can define parameters such as root users in both foreman and hiera. What is the recommended way?
=> (ST) do as you like.
(VL) How to define settings for a single machine?
=> either create a hostgroup with one machine, or use something like: if ($::fqdn=="myvm.cern.ch") { some puppet code }
Why specify "require=>Package['emacs']" in the Puppet declaration of the emacs configuration file?
=> (NB) often you need to create a cofnig file in a folder created by the package. So this statement ensures that the package (and hence relevant folders) is installed before the confg file is created. The require parameter is used to order Puppet actions.
(JC) do we have a selinux module?
=> (ST) a module exists, will be made available in AI as necessary
What if there is a typo in my puppet code ?
(ST) Syntax errors will cause the commit in git to be rejected. Other mistakes will cause the configuration to fail and there will be an error message shown when puppet runs.
Does it work only on KVM?
=> yes AI will eventually run all Linux virtual machines on KVM, contrary to Quattor-manage VMs that were running on HyperV
(SB) Can we identify RPMs installed manually on a node? (this was easy to detect in Quattor)
=> (AL) not with Puppet. But other approaches can be used to make sure that a machine possibly subject to manual operations (e.g. dev nodes) is in its reference configuration (e.g. discard and recreate reference machines to ensure that manual configurations do not persist – this is the cattle use case)
(SB) how many cores/RAM for a VM in AI?
=> (JvE) you have the flexibility to create your own flavors if you need large VMs, technically the limit is the physical host (typically 12 cores/48GB RAM)
Any plan to decrease the time of a VM creation ?
=> actually this depends on landb/dns – at least for pets; cattle use case is faster because we don't get to set the hostname, we just get one.
(JM) when can we create production "pets"?
=> JvE: after summer. You are encouraged to change the model to something compatible with 'cattle' whenever possible.
Next meeting in April: yum, managing and deploying updates, how to deploy custom software (custom repositories), osdate-like feature
NB: Questions and answers can be found in the Indico meeting minutes.
IT-PES: Alex Lossent, Vitor Gomes, Nacho Barrentos, Steve traylen, Maite Barroso, Manuel Guijarro, Nils Hoimyr, Peter Jones, John Evans, Evangelos Atlidakis, David Asbury, Akos Hencz
IT-ES: Pabo Saiz
IT-DI: Denise Heagerty
IT-DSS: Alex Iribarren
IT-CF: Veromique Lefebure
IT-GT: Andres Abad Rodriguez
IT-OIS: Jan Van Eldik, Tim Bell
ATLAS: Sergey Barnov, Alexey Buzykaev
CMS: Jorge Molina
LHCb: Joel Closier
LCD: Stephane Poss, Andre Sailer
Questions:
(PS) Pet VM type used in demo, but we said we support only cattle at the moment?
=> (NB) indeed. At the moment it makes no difference.
(PS) We can define parameters such as root users in both foreman and hiera. What is the recommended way?
=> (ST) do as you like.
(VL) How to define settings for a single machine?
=> either create a hostgroup with one machine, or use something like: if ($::fqdn=="myvm.cern.ch") { some puppet code }
Why specify "require=>Package['emacs']" in the Puppet declaration of the emacs configuration file?
=> (NB) often you need to create a cofnig file in a folder created by the package. So this statement ensures that the package (and hence relevant folders) is installed before the confg file is created. The require parameter is used to order Puppet actions.
(JC) do we have a selinux module?
=> (ST) a module exists, will be made available in AI as necessary
What if there is a typo in my puppet code ?
(ST) Syntax errors will cause the commit in git to be rejected. Other mistakes will cause the configuration to fail and there will be an error message shown when puppet runs.
Does it work only on KVM?
=> yes AI will eventually run all Linux virtual machines on KVM, contrary to Quattor-manage VMs that were running on HyperV
(SB) Can we identify RPMs installed manually on a node? (this was easy to detect in Quattor)
=> (AL) not with Puppet. But other approaches can be used to make sure that a machine possibly subject to manual operations (e.g. dev nodes) is in its reference configuration (e.g. discard and recreate reference machines to ensure that manual configurations do not persist – this is the cattle use case)
(SB) how many cores/RAM for a VM in AI?
=> (JvE) you have the flexibility to create your own flavors if you need large VMs, technically the limit is the physical host (typically 12 cores/48GB RAM)
Any plan to decrease the time of a VM creation ?
=> actually this depends on landb/dns – at least for pets; cattle use case is faster because we don't get to set the hostname, we just get one.
(JM) when can we create production "pets"?
=> JvE: after summer. You are encouraged to change the model to something compatible with 'cattle' whenever possible.
Next meeting in April: yum, managing and deploying updates, how to deploy custom software (custom repositories), osdate-like feature
NB: Questions and answers can be found in the Indico meeting minutes.
There are minutes attached to this event.
Show them.
