Skip to content

Commit 3b49be0

Browse files
committed
test(e2e): give tenant workers two vCPUs
The tenant worker boots Talos and unpacks the kubelet image on a single vCPU whose CFS quota also covers the QEMU emulator and IO threads, and measurements on both quiet and loaded runs show the compute container pinned at 93-98% of that one-core ceiling for the whole node-Ready budget. The join deadline is then decided entirely by how much CPU work the boot happens to need, and under parallel e2e load it does not fit. Doubling the worker to u1.large raises the ceiling instead of the weight, which measurements showed changes nothing since the workers never lose a scheduling fight. Assisted-By: Claude <[email protected]> Signed-off-by: Aleksei Sviridkin <[email protected]>
1 parent ef96292 commit 3b49be0

1 file changed

Lines changed: 1 addition & 1 deletion

File tree

‎hack/e2e-chainsaw/_lib/run-kubernetes.sh‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -2651,7 +2651,7 @@ ${ouroboros_addon}
26512651
md0:
26522652
diskSize: 20Gi
26532653
gpus: []
2654-
instanceType: u1.medium
2654+
instanceType: u1.large
26552655
# The failure this suite keeps hitting is a worker that stalls in the
26562656
# guest before Talos apid answers, which leaves nothing to read on the
26572657
# management side and nothing for talosctl to connect to. Turning this on

0 commit comments

Comments
 (0)