Skip to content

Commit ef12565

Browse files
authored
fix(recipes): raise nvidia-setup-full resources for EFA DKMS build (#1738)
Signed-off-by: Alex Yuskauskas <ayuskauskas@nvidia.com>
1 parent bc3ea17 commit ef12565

1 file changed

Lines changed: 11 additions & 4 deletions

File tree

  • recipes/components/nodewright-customizations/manifests

recipes/components/nodewright-customizations/manifests/tuning.yaml

Lines changed: 11 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -126,10 +126,17 @@ spec:
126126
interrupt:
127127
type: reboot
128128
resources:
129-
cpuLimit: 4000m
130-
cpuRequest: 2000m
131-
memoryLimit: 8192Mi
132-
memoryRequest: 4096Mi
129+
# Elevated from 4000m/8192Mi for the EFA DKMS build. efa_installer.sh's
130+
# "Inspecting kernel" autoconf is a highly parallel kernel-module build;
131+
# inside the skyhook chroot nproc reports the host core count, so make -j
132+
# oversubscribes. Under the prior 4 CPU / 8Gi cgroup it was CFS-throttled
133+
# and its feature probes were starved/OOM-killed, so efa misdetected the
134+
# kernel and emitted compat shims that fail to compile on 6.17. More CPU
135+
# and memory headroom lets the probes complete and detect correctly.
136+
cpuLimit: 8000m
137+
cpuRequest: 4000m
138+
memoryLimit: 16384Mi
139+
memoryRequest: 8192Mi
133140
configMap:
134141
{{- if $cust.service }}
135142
service: {{ $cust.service }}

0 commit comments

Comments
 (0)