Merge pull request #38001 from dchen1107/master

cluster/common.sh

@@ -627,6 +627,7 @@ NETWORK_PROVIDER: $(yaml-quote ${NETWORK_PROVIDER:-})
 NETWORK_POLICY_PROVIDER: $(yaml-quote ${NETWORK_POLICY_PROVIDER:-})
 PREPULL_E2E_IMAGES: $(yaml-quote ${PREPULL_E2E_IMAGES:-})
 HAIRPIN_MODE: $(yaml-quote ${HAIRPIN_MODE:-})
+SOFTLOCKUP_PANIC: $(yaml-quote ${SOFTLOCKUP_PANIC:-})
 OPENCONTRAIL_TAG: $(yaml-quote ${OPENCONTRAIL_TAG:-})
 OPENCONTRAIL_KUBERNETES_TAG: $(yaml-quote ${OPENCONTRAIL_KUBERNETES_TAG:-})
 OPENCONTRAIL_PUBLIC_SUBNET: $(yaml-quote ${OPENCONTRAIL_PUBLIC_SUBNET:-})

cluster/gce/config-default.sh

@@ -173,3 +173,7 @@ SCHEDULING_ALGORITHM_PROVIDER="${SCHEDULING_ALGORITHM_PROVIDER:-}"
 
 # Optional: install a default StorageClass
 ENABLE_DEFAULT_STORAGE_CLASS="${ENABLE_DEFAULT_STORAGE_CLASS:-true}"
+
+# TODO(dawn1107): Remove this once the flag is built into CVM image.
+# Kernel panic upon soft lockup issue
+SOFTLOCKUP_PANIC="${SOFTLOCKUP_PANIC:-false}" # true, false

cluster/gce/config-test.sh

@@ -210,3 +210,7 @@ SCHEDULING_ALGORITHM_PROVIDER="${SCHEDULING_ALGORITHM_PROVIDER:-}"
 
 # Optional: install a default StorageClass
 ENABLE_DEFAULT_STORAGE_CLASS="${ENABLE_DEFAULT_STORAGE_CLASS:-true}"
+
+# TODO(dawn1107): Remove this once the flag is built into CVM image.
+# Kernel panic upon soft lockup issue
+SOFTLOCKUP_PANIC="${SOFTLOCKUP_PANIC:-true}" # true, false

cluster/gce/configure-vm.sh

@@ -446,6 +446,7 @@ admission_control: '$(echo "$ADMISSION_CONTROL" | sed -e "s/'/''/g")'
 network_provider: '$(echo "$NETWORK_PROVIDER" | sed -e "s/'/''/g")'
 prepull_e2e_images: '$(echo "$PREPULL_E2E_IMAGES" | sed -e "s/'/''/g")'
 hairpin_mode: '$(echo "$HAIRPIN_MODE" | sed -e "s/'/''/g")'
+softlockup_panic: '$(echo "$SOFTLOCKUP_PANIC" | sed -e "s/'/''/g")'
 opencontrail_tag: '$(echo "$OPENCONTRAIL_TAG" | sed -e "s/'/''/g")'
 opencontrail_kubernetes_tag: '$(echo "$OPENCONTRAIL_KUBERNETES_TAG")'
 opencontrail_public_subnet: '$(echo "$OPENCONTRAIL_PUBLIC_SUBNET")'

cluster/saltbase/salt/docker/init.sls

@@ -261,6 +261,20 @@ net.ipv4.ip_forward:
   sysctl.present:
     - value: 1
 
+{% if pillar.get('softlockup_panic', '').lower() == 'true' %}
+# TODO(dchen1107) Remove this once kernel.softlockup_panic is built into the CVM image.
+/etc/sysctl.conf:
+  file.append:
+    - text:
+      - "kernel.softlockup_panic = 1"
+      - "kernel.softlockup_all_cpu_backtrace = 1"
+
+'sysctl-reload':
+  cmd.run:
+    - name: 'sysctl --system'
+    - unless: 'sysctl -a | grep "kernel.softlockup_panic = 1"'
+{% endif %}
+ 
 {{ environment_file }}:
   file.managed:
     - source: salt://docker/docker-defaults