eqiad: the servers can be divided into 4 groups by procurement ticket:

group A - T159963 - mw1307 through mw1348 - R430 Dual Intel Xeon E5-2650 CPU (2.2GHz/12 cores per cpu)

group B - T228214 - mw1349 through mw1384 - R440 Intel Xeon Silver 4114 2.2GHz

group C - T233639 - mw1385 through mw1413 - R440 Intel Xeon Silver 4214 2.2GHz

group D - T271155 - mw1414 through mw1456 - R440 Intel Xeon Silver 4214 2.2GHz

codfw:

group A - T151779 - mw2251 through mw2258 - R430 Xeon E5 2650 - 2.2GHz/12core - already currently set to 25

group B - T183404 - mw2559 through mw2290) - R430 Xeon E5 2650 - 2.3GHz/10 cores - already currently set to 25

group C - T231255 - mw2291 through mw2376 - R440 Intel Xeon Silver 4214 2.2GHz - already currently set to 30

Based on this I would say we should set mw1307 through mw1348 to 25 (or 20?) and then we'd be done here.

There is a new type of servers now:

group D - mw2416, mw2417 and mw2418 - R440 - Xeon Silver 4210R 2.4G - (40 processors, 128GB RAM), that's only 40 processors vs 48 but MORE RAM, 128 vs 96)

are assigned as API and they are of the new type: Dell PowerEdge R440 - ConfigC 202107

role assignments: https://gerrit.wikimedia.org/r/c/operations/puppet/+/785918

procurement ticket: https://phabricator.wikimedia.org/T286516

@RLazarus @Joe Just saw this again in the history after a while. re: https://config-master.wikimedia.org/pybal/eqiad/api-https

My suggestion was to set mw1307 through mw1348 to 25 (or 20?) because they are of type A as above.

Presumably after that this ticket would then be resolved. Should I do it? Would you want to check as well?

That sounds right to me; it would give us the same distribution as codfw, which is probably as much work as we need to do on this. I don't think it's worth investing time into any deliberate benchmarking, but if the cluster happens to saturate unevenly again in a future incident, we can tweak opportunistically.

By the way, we should do the same thing for the appservers-https cluster: https://config-master.wikimedia.org/pybal/eqiad/appservers-https. If you don't mind lowering those Xeon E5s to 25 at the same time, we can just repurpose this ticket to cover both. (But if it'll be a longer-lived item, we can open a new task for it, either way works for me.)

Thanks for grabbing this! If it turns out you don't have time after all, just let me know and I can pick it up.

Mentioned in SAL (#wikimedia-operations) [2022-08-24T21:20:05Z] <mutante> setting weight to 25 (from 30) for appservers and API servers in the range mw1307 through mw1348 because they are of an older hardware type (not changing weights of jobrunners/videoscalers even if in this range) (T304800)

@RLazarus Thank you! Done! I changed the value from 30 to 25 for any server within the range mw1307 through mw1348 that was either appserver or api_appserver cluster. I ignored the weights for jobscaler/videoscaler even if within this range.

https://config-master.wikimedia.org/pybal/eqiad/api-https

https://config-master.wikimedia.org/pybal/eqiad/appservers-https

https://config-master.wikimedia.org/pybal/eqiad/jobrunner

And here is the netbox query to show that is the correct range that comes from procurement ticket T159963

https://netbox.wikimedia.org/dcim/devices/?q=mw13&serial=&asset_tag=&mac_address=&console_ports=&console_server_ports=&power_ports=&power_outlets=&interfaces=&pass_through_ports=&has_primary_ip=&virtual_chassis_member=&local_context_data=&cf_purchase_date=&cf_ticket=+T159963