"Managed Nodes" in chef for network switched, PDUs


#1

Ohai chefs,

I stumbled accross this old session summary (
http://wiki.opscode.com/display/chef/Managed+Nodes+Part+2 ) while looking
for a way to automate our network switch, and other custom hardware
configuration where we can’t run chef-client. Do you think this could ever
happen?

We are currently managing our bare metal infrastructure with Chef and we
looked into Crowbar but decided not to use it for many reasons (focus on
Openstack, works for one DC only and we have multiple, very heavy process
to use barclamps and wrap Chef). We went lean and basically built a set of
cookbooks to provide PXE boot, OS installation, DHCP/DNS/IPMI and other
things (BIOS upgrade, RAID firmware…). Thus we don’t rely on anything but
chef and we can use community cookbooks as much as possible.

However now we look into automating configuration for other hardware
devices and Managed Nodes look like an interesting solution, especially
since there isn’t any system available that’s not vendor specific. We
already have a database that handle the hardware location part (like, rack,
room, DC, serial number, etc…) and we use ohai plugins to populate the
data into Chef.

Any idea? What are other people using?

Cheers,
Maxime


#2

Maxime, I have seen that some people are using
pxe_dusthttp://community.opscode.com/cookbooks/pxe_dust;
we currently have our own internal developed solution that does exactly
what you describe.

I’d love to hear what other people are doing, because I too think Crowbar
is a little heavyweight for our needs, and I am currently in a position to
replace our existing provisioning infrastructure.

Thanks for starting this topic,
Brian

On Tue, Feb 5, 2013 at 11:58 AM, Maxime Brugidou
maxime.brugidou@gmail.comwrote:

Ohai chefs,

I stumbled accross this old session summary (
http://wiki.opscode.com/display/chef/Managed+Nodes+Part+2 ) while looking
for a way to automate our network switch, and other custom hardware
configuration where we can’t run chef-client. Do you think this could ever
happen?

We are currently managing our bare metal infrastructure with Chef and we
looked into Crowbar but decided not to use it for many reasons (focus on
Openstack, works for one DC only and we have multiple, very heavy process
to use barclamps and wrap Chef). We went lean and basically built a set of
cookbooks to provide PXE boot, OS installation, DHCP/DNS/IPMI and other
things (BIOS upgrade, RAID firmware…). Thus we don’t rely on anything but
chef and we can use community cookbooks as much as possible.

However now we look into automating configuration for other hardware
devices and Managed Nodes look like an interesting solution, especially
since there isn’t any system available that’s not vendor specific. We
already have a database that handle the hardware location part (like, rack,
room, DC, serial number, etc…) and we use ohai plugins to populate the
data into Chef.

Any idea? What are other people using?

Cheers,
Maxime


#3

Hi Brian, we will probably move ou tftp/pxe cookbooks to pxe_dust since it
looks well maintained and we didn’t notice it at the time we built our
first cookbooks.

Our internal solution has an initramfs image (similar to sledgehammer in
Crowbar) that is configured to boot and run chef-client, set its initial
run list to “role[firstboot]” if the node does not have a a run list yet.
New nodes pop up in chef just like they would on Crowbar and we can
provision them using “standard” chef operations (basically assigning a run
list).

We have an OS cookbook that install the operating system and reboots (only
if needed, some nodes are also running diskless).

I’d love to discuss more about the topic, since we are also building our
next provisioning infrastructure. My current concern is about hardware
devices that can’t run chef-client (but have a management API).

Maxime

On Tue, Feb 5, 2013 at 6:07 PM, Brian Hatfield bhatfield@brightcove.comwrote:

Maxime, I have seen that some people are using pxe_dusthttp://community.opscode.com/cookbooks/pxe_dust;
we currently have our own internal developed solution that does exactly
what you describe.

I’d love to hear what other people are doing, because I too think Crowbar
is a little heavyweight for our needs, and I am currently in a position to
replace our existing provisioning infrastructure.

Thanks for starting this topic,
Brian

On Tue, Feb 5, 2013 at 11:58 AM, Maxime Brugidou <
maxime.brugidou@gmail.com> wrote:

Ohai chefs,

I stumbled accross this old session summary (
http://wiki.opscode.com/display/chef/Managed+Nodes+Part+2 ) while
looking for a way to automate our network switch, and other custom hardware
configuration where we can’t run chef-client. Do you think this could ever
happen?

We are currently managing our bare metal infrastructure with Chef and we
looked into Crowbar but decided not to use it for many reasons (focus on
Openstack, works for one DC only and we have multiple, very heavy process
to use barclamps and wrap Chef). We went lean and basically built a set of
cookbooks to provide PXE boot, OS installation, DHCP/DNS/IPMI and other
things (BIOS upgrade, RAID firmware…). Thus we don’t rely on anything but
chef and we can use community cookbooks as much as possible.

However now we look into automating configuration for other hardware
devices and Managed Nodes look like an interesting solution, especially
since there isn’t any system available that’s not vendor specific. We
already have a database that handle the hardware location part (like, rack,
room, DC, serial number, etc…) and we use ohai plugins to populate the
data into Chef.

Any idea? What are other people using?

Cheers,
Maxime


#4

pxe_dust is intended for simple pxe booting and Chef client installation, not nearly as ambitious as some of the other tools out there (no BIOS management for example). There are additional features planned, but I plan on keeping the feature set fairly small and focused. Razor has recently gained Chef support thanks to Fletcher Nichol’s work, that’s another alternative.

The original discussion of Managed Nodes was around using Chef to automate devices that could not install a Chef client on them. It’s still a goal for Chef, but progress has been made with installation of Chef on Arista and similar switches in the meantime.

Thanks,
Matt Ray
Senior Technical Evangelist | Opscode Inc.
matt@opscode.com | (512) 731-2218
Twitter, IRC, GitHub: mattray


From: Maxime Brugidou [maxime.brugidou@gmail.com]
Sent: Tuesday, February 05, 2013 11:30 AM
To: chef@lists.opscode.com
Subject: [chef] Re: Re: “Managed Nodes” in chef for network switched, PDUs…

Hi Brian, we will probably move ou tftp/pxe cookbooks to pxe_dust since it looks well maintained and we didn’t notice it at the time we built our first cookbooks.

Our internal solution has an initramfs image (similar to sledgehammer in Crowbar) that is configured to boot and run chef-client, set its initial run list to “role[firstboot]” if the node does not have a a run list yet. New nodes pop up in chef just like they would on Crowbar and we can provision them using “standard” chef operations (basically assigning a run list).

We have an OS cookbook that install the operating system and reboots (only if needed, some nodes are also running diskless).

I’d love to discuss more about the topic, since we are also building our next provisioning infrastructure. My current concern is about hardware devices that can’t run chef-client (but have a management API).

Maxime

On Tue, Feb 5, 2013 at 6:07 PM, Brian Hatfield <bhatfield@brightcove.commailto:bhatfield@brightcove.com> wrote:
Maxime, I have seen that some people are using pxe_dusthttp://community.opscode.com/cookbooks/pxe_dust; we currently have our own internal developed solution that does exactly what you describe.

I’d love to hear what other people are doing, because I too think Crowbar is a little heavyweight for our needs, and I am currently in a position to replace our existing provisioning infrastructure.

Thanks for starting this topic,
Brian

On Tue, Feb 5, 2013 at 11:58 AM, Maxime Brugidou <maxime.brugidou@gmail.commailto:maxime.brugidou@gmail.com> wrote:
Ohai chefs,

I stumbled accross this old session summary (http://wiki.opscode.com/display/chef/Managed+Nodes+Part+2 ) while looking for a way to automate our network switch, and other custom hardware configuration where we can’t run chef-client. Do you think this could ever happen?

We are currently managing our bare metal infrastructure with Chef and we looked into Crowbar but decided not to use it for many reasons (focus on Openstack, works for one DC only and we have multiple, very heavy process to use barclamps and wrap Chef). We went lean and basically built a set of cookbooks to provide PXE boot, OS installation, DHCP/DNS/IPMI and other things (BIOS upgrade, RAID firmware…). Thus we don’t rely on anything but chef and we can use community cookbooks as much as possible.

However now we look into automating configuration for other hardware devices and Managed Nodes look like an interesting solution, especially since there isn’t any system available that’s not vendor specific. We already have a database that handle the hardware location part (like, rack, room, DC, serial number, etc…) and we use ohai plugins to populate the data into Chef.

Any idea? What are other people using?

Cheers,
Maxime


Managing Embedded Devices with Chef?
#5

I agree with you, I do not intend to use pxe_dust for something other than
provide a pxe boot image.

In general, we tend to have small focused cookbooks that do one thing well.
Our BIOS cookbook focuses on bios firmware upgrade and configuration (it’s
very vendor specific for now). Our OS cookbook only install a bare
operating system and manage disk partitions…etc.

That’s good news about running chef on other devices! Hope we’ll hear more
about it.
On Feb 5, 2013 6:58 PM, “Matt Ray” matt@opscode.com wrote:

pxe_dust is intended for simple pxe booting and Chef client
installation, not nearly as ambitious as some of the other tools out there
(no BIOS management for example). There are additional features planned,
but I plan on keeping the feature set fairly small and focused. Razor has
recently gained Chef support thanks to Fletcher Nichol’s work, that’s
another alternative.

The original discussion of Managed Nodes was around using Chef to
automate devices that could not install a Chef client on them. It’s still a
goal for Chef, but progress has been made with installation of Chef on
Arista and similar switches in the meantime.

Thanks,
Matt Ray
Senior Technical Evangelist | Opscode Inc.
matt@opscode.com | (512) 731-2218
Twitter, IRC, GitHub: mattray

From: Maxime Brugidou [maxime.brugidou@gmail.com]
Sent: Tuesday, February 05, 2013 11:30 AM
To: chef@lists.opscode.com
Subject: [chef] Re: Re: “Managed Nodes” in chef for network switched,
PDUs…

Hi Brian, we will probably move ou tftp/pxe cookbooks to pxe_dust since
it looks well maintained and we didn’t notice it at the time we built our
first cookbooks.

Our internal solution has an initramfs image (similar to sledgehammer in
Crowbar) that is configured to boot and run chef-client, set its initial
run list to “role[firstboot]” if the node does not have a a run list yet.
New nodes pop up in chef just like they would on Crowbar and we can
provision them using “standard” chef operations (basically assigning a run
list).

We have an OS cookbook that install the operating system and reboots
(only if needed, some nodes are also running diskless).

I’d love to discuss more about the topic, since we are also building our
next provisioning infrastructure. My current concern is about hardware
devices that can’t run chef-client (but have a management API).

Maxime

On Tue, Feb 5, 2013 at 6:07 PM, Brian Hatfield bhatfield@brightcove.comwrote:

Maxime, I have seen that some people are using pxe_dusthttp://community.opscode.com/cookbooks/pxe_dust;
we currently have our own internal developed solution that does exactly
what you describe.

I’d love to hear what other people are doing, because I too think
Crowbar is a little heavyweight for our needs, and I am currently in a
position to replace our existing provisioning infrastructure.

Thanks for starting this topic,
Brian

On Tue, Feb 5, 2013 at 11:58 AM, Maxime Brugidou <
maxime.brugidou@gmail.com> wrote:

Ohai chefs,

I stumbled accross this old session summary (
http://wiki.opscode.com/display/chef/Managed+Nodes+Part+2 ) while
looking for a way to automate our network switch, and other custom hardware
configuration where we can’t run chef-client. Do you think this could ever
happen?

We are currently managing our bare metal infrastructure with Chef and
we looked into Crowbar but decided not to use it for many reasons (focus on
Openstack, works for one DC only and we have multiple, very heavy process
to use barclamps and wrap Chef). We went lean and basically built a set of
cookbooks to provide PXE boot, OS installation, DHCP/DNS/IPMI and other
things (BIOS upgrade, RAID firmware…). Thus we don’t rely on anything but
chef and we can use community cookbooks as much as possible.

However now we look into automating configuration for other hardware
devices and Managed Nodes look like an interesting solution, especially
since there isn’t any system available that’s not vendor specific. We
already have a database that handle the hardware location part (like, rack,
room, DC, serial number, etc…) and we use ohai plugins to populate the
data into Chef.

Any idea? What are other people using?

Cheers,
Maxime