osdc: replicate -opt node-packing optimizations into the regular runner/nodepool defs#902
osdc: replicate -opt node-packing optimizations into the regular runner/nodepool defs#902jeanschmidt wants to merge 2 commits into
Conversation
Capacity reportcommit ✅ simulate-cluster✅ analyze-utilization |
tofu plan — meta-prod-aws-uw1✅ Plan succeeded · commit Plan output |
tofu plan — meta-prod-aws-ue1✅ Plan succeeded · commit Plan output |
tofu plan — lf-prod-aws-ue1✅ Plan succeeded · commit Plan output |
tofu plan — lf-prod-aws-ue2✅ Plan succeeded · commit Plan output |
|
@claude Double check to see if all the changes are moved correctly from arc-runners-opt to arc-runners. Any recent changes on arc-runners to call out? |
|
Claude finished @huydhn's task in 4m 20s —— View job Verifying arc-runners-opt → arc-runners migration
✅ Everything moved correctlyI diffed the regular modules against the Runner defs (21 changed) — all identical to
Nodepool tier fleets (11 added/renamed) — all identical to
Monolithic fleets removed (6):
|
Stack from ghstack (oldest at bottom):
Impact: every cluster on the regular arc-runners/nodepools modules (all staging + meta-prod-aws-ue2, lf-prod-aws-ue1/ue2, and meta-prod-aws-uw1 nodepools); no cluster is redeployed by this change
Risk: medium
What
Copy the packing optimizations that until now lived only in the
-optmodules into the regular
arc-runners/nodepoolsdefs: 21 runner defs gainan explicit
node_fleeton a right-sized, smaller instance type; nodepoolsgain 11 per-size tier fleets, the
c7i-runnerpool switches to Graviton, andthe 6 superseded monolithic family fleets (c7a, c7i, m7g, m7i, m8g, r7i) are
removed. The
-optmodules are left untouched here.Why
Only meta-prod-aws-ue1 runs the
-optmodules today; the pilot showed acluster-wide vCPUhours / CIhousr gain (≈14%) by packing each runner onto a
size-matched node tier instead of stranding a small job on a huge instance.
Landing the same defs in the regular modules is the first step of rolling
that win out fleet-wide.