Skip to content

es-mapping.md

Latest commit

 

History

History
43 lines (29 loc) · 3.96 KB

es-mapping.md

File metadata and controls

43 lines (29 loc) · 3.96 KB
Raw

SOP: edge switch mapping

Every edge switch has a DNS name from its asset tag (../asset-tags.md), <asset-tag>.venue.rubykaigi.net (for example rk-srnp19a2), and a fixed DHCP assignment in the management VLAN (hosts.txt, rendered into the DHCP and DNS config). A switch therefore comes up at the same address wherever it is plugged in, and nobody configures an IP address on it at setup.

The location name es-<area>-NN (see ../topology.md) is a CNAME to the asset tag name. It is what monitoring (k8s/prom/snmp.jsonnet, glazed/glazed.jsonnet) and humans use. Moving a switch to another area or swapping a unit means editing one CNAME.

Which switch goes where is not planned in advance: whoever mounts a switch records its asset tag against the location, as for APs (ap-naming.md).

Inputs (prepared by humans)

  1. The deployment map: one row per edge switch with its es-<area>-NN name and asset tag. In 2026 this was the edge switch sheet of the NOC spreadsheet; ask for it as TSV in a file under tmp/.
  2. Which management VLAN each switch sits on, only if some switches are on the second venue network (venue2-vlan1000). Check hosts.txt for the tag; a tag listed only under mgmt2 must point at venue2-vlan1000.<asset-tag>.venue.rubykaigi.net. instead.

Steps

  1. Confirm every asset tag in the map has a fixed assignment: grep <tag> hosts.txt. A tag missing there has no address and needs adding to hosts.txt first (regenerate with gen-hosts.rb).

  2. Edit the private zone block (hosted_zone ... "Z05547502KT77L0O53UWK") in route53/rubykaigi.net.route. Replace the es-* CNAMEs with the new map, one rrset 'es-<area>-NN.venue.rubykaigi.net.', 'CNAME' each, ttl 300, pointing at <asset-tag>.venue.rubykaigi.net.. Delete the entries for areas the venue does not have, so stale names do not look like down switches.

  3. Dry-run and apply with roadworker from route53/ (bundle exec roadwork -a --dry-run, then -a without it; the exact flags follow the other .route changes).

  4. Update the same name list in k8s/prom/snmp.jsonnet and glazed/glazed.jsonnet (ios(name, 'venue')), then run ./gen-k8s.rb and ./gen-glazed.rb and commit the gen/ output with them. A name left out of these is not monitored or backed up.

  5. Verify from inside the venue network: dig +short es-<area>-NN.venue.rubykaigi.net resolves through the asset tag to the management address for every switch, and show lldp neighbors on cs-01 agrees with the uplink table in the event README.

  6. Set each switch's own hostname to its location name, so the prompt and show output identify the chassis and parallel sessions are not confused:

    configure terminal
hostname es-<area>-NN
end
write memory

    Do it over the DNS name from step 5 and confirm the new prompt matches the name you connected to. The hostname is stored on the switch, so a switch moved to another area needs it changed again.

  7. Record the area to asset tag mapping in docs/events/<event>/ so the next event can see what was where.

A switch moved after setup needs only steps 2, 3, 5 and 6; the monitoring name list does not change.

Teardown

Set each hostname back to its asset tag (hostname <asset-tag>, write memory) before the switch is pulled, as APs go back to their asset tags (ap-naming.md). A switch that leaves under a location name shows the wrong area next year if mounted elsewhere.

When it goes wrong

  • es-* does not resolve: the CNAME has not applied, or points at a tag with no hosts.txt entry.
  • It resolves but the switch does not answer: the asset tag in the map is wrong (someone mounted another unit). Compare the label on the switch with the map; show lldp neighbors on the uplink shows the hostname the switch reports.
  • The prompt shows a different name from the one you connected to: the hostname was not updated (step 6) or the CNAME points at another unit. Stop and compare before changing anything.