Page MenuHomePhabricator

Create public vlans in eqiad and codfw
Open, MediumPublic

Description

See https://wikitech.wikimedia.org/wiki/Public_baremetal_hosts_in_core_sites for context.

Next step:

  • Decide which switches to use to host the public VLANs eqiad and codfw:
  • eqiad A/B : A3
  • eqiad C/D : C1
  • eqiad E/F : E8
  • codfw A/B : B3
  • codfw C/D : D3
  • codfw E/F : E5
  • Decide on a name public1-{rackname}-{site}
  • Assign a v4 /28 (and v6 /64) for each one
  • eqiad A/B :
  • eqiad C/D :
  • eqiad E/F :
  • codfw A/B :
  • codfw C/D :
  • codfw E/F :
  • Update ACL and routing policies if any
  • move a sretest hosts in one of those racks to test provisioning

Event Timeline

Restricted Application added a subscriber: Aklapper. · View Herald Transcript

If we are going to have one public-enabled rack per "pod" then should we not have just one vlan assigned for codfw row E/F (and then one also for a/b and c/d)?

Maybe we should have a parent task to plan out the overall approach and delivery of all of them? public1-{rackname}-{site} makes sense as a name, as does a /28 for the v4. Hopefully the deploy will also be quite easy, so perhaps instead of an sretest host we could just try to move one of the less-critical existing hosts?

My initial thought was to start with E/F only but you're right better plan it fully here, especially the IP allocations. (task description updated)

ayounsi renamed this task from Create public vlan on eqiad and codfw pods E/F to Create public vlans in eqiad and codfw.Apr 2 2026, 12:41 PM
ayounsi updated the task description. (Show Details)

Is it maybe an idea to re-use some of the existing vlans? Like if we assign rack A1 as the public rack for the A/B POD we could add all the hosts to public1-a-eqiad as we move them? And then when complete rename the vlan to public1-a1-eqiad?

No preference might be a bit of a shortcut.

ayounsi triaged this task as Medium priority.Apr 7 2026, 2:32 PM

That could help slightly for the few hosts that are in the matching row (allow to move them to a rack without re-numbering).
If we look at eqiad and its 18 hosts, that means ~4.5 per row (which we can round up to 4 as some hosts are to be decom), that means it could make the transition smoother for 6 hosts 3 in A/B, 3 in C/D (assuming the rack we pick already have one public hosts).
To be considered, but not strictly needed.

DC-Ops (@Papaul, @Jhancock.wm, @VRiley-WMF , @Jclark-ctr, @RobH).

Which rack, from each "pods" (see task description) could we use to have an additional "public" vlan ? That means those racks will need to have some space for at least 6 hosts day 1, and probably a few more as time goes. But deployment can be staggered (eg. A/B month X, C/D month Y, etc)

Ideally those would be the same in all sites (Eg. A3, C3, E3 in both codfw and eqiad) but that's not a strict requirement.

imho, i'd prefer a rack not in A row cause of the two CR racks already taking up real estate.
D row has no specialty rack at all so we can easily work around that for future private vlan installs.
codfw's E row is 5 racks long but the F row is 4 racks + 1 Frack, so E would be the better choice. and not E-3 cause it already has less room cause of all the patch panels.

codfw A/B : B2-B8 (i'd pick B3)
codfw C/D : any D row (i'd pick D3)
codfw E/F : E-1, E-2, E-4, or E-5 (i'd pick E-5)

also papaul is on vacation and i'd like to have his weight in as well

D row has no specialty rack at all so we can easily work around that for future private vlan installs.

To be clear these racks would not be "as special" as say our Fundraising our Cloud racks. We can host regular production servers in them just fine. That said they will be the only place we will put servers that need to be on the public vlan, so while we can put private vlan hosts in there it might be best to not do that and save the space. Anyway just to be clear on what the constraints are.

That could help slightly for the few hosts that are in the matching row (allow to move them to a rack without re-numbering).

Yeah that and just being lazy to not have to assign new ranges and vlan ids right away etc.

For Eqiad, I would choose A3, C1, and either E8 or F8. A3 is currently 1G, and C1 is pending the arrival of new switches. It was previously out for fundraising.

ayounsi updated the task description. (Show Details)

Great, thanks ! Task description updated.