@arturo thanks for this. The hosts can go in any rack, but we should make sure hosts of the same type go into different ones for redundancy.

I.e. the two cloudservices should be in separate racks, the 3 cloudcontrols should be in separate racks.

C8/D5 are good choices if there is room in those, but E4/F4 will work as well (1 more hop in some cases but that won’t make much difference).

Thanks. So if ops-eqiad don't have any other preference, we could do something like:

• cloudcontrol1005 --> C8
• cloudcontrol1006 --> D5
• cloudcontrol1007 --> E4
• cloudservices1005 --> F4
• cloudservices1006 --> C8 (new)

Put cloudservices1005 in D5 if there is room there instead of F4.

In T341494#9004690, @cmooney wrote:

Put cloudservices1005 in D5 if there is room there instead of F4.

Done. What should go to F4 instead?

I would like to have something in every rack if possible, for failure distribution.

In T341494#9004717, @aborrero wrote:

In T341494#9004690, @cmooney wrote:

Put cloudservices1005 in D5 if there is room there instead of F4.

Done. What should go to F4 instead?

I would like to have something in every rack if possible, for failure distribution.

Correct me if I’m wrong but should the distribution be considered separately for cloudcontrol vs cloudservice?

So as long as the two cloudservice are in different racks that’s as distributed as it can get? Ditto for the 3 cloudcontrols?

Based on that I’d put one of each in c8/d5 (these are effectively the “core”, used for transit between e4 and f4 as well as being closest to external internet). Cloud control there are 3 so place the 3rd in either e4/f4 for diversity of course.

Unless you’re considering it like a cluster of 5, and how to spread over the 4 racks?

In T341494#9009786, @cmooney wrote:

Unless you’re considering it like a cluster of 5, and how to spread over the 4 racks?

I guess it is more a psychological factor, related to the amount of hardware in the racks, even if they are different services :-)

I'm OK for now with the plan as it is.