Skip to content

fix: remove EXIT trap from wait_for_available_zone.sh and modularise the daily tests of kueue pod cleanup #6154

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
sudheer-quad merged 4 commits into from
Aug 28, 2026
Merged

fix: remove EXIT trap from wait_for_available_zone.sh and modularise the daily tests of kueue pod cleanup #6154

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
aa5b42d
fix: remove EXIT trap from wait_for_available_zone.sh to fix resource…
sudheer-quad Aug 17, 2026
c237b0c
Refactor kueue daily tests to use shared Kueue pod cleanup script
sudheer-quad Aug 18, 2026
3ee99c9
address gemini review
sudheer-quad Aug 20, 2026
69f0599
Merge branch 'develop' into fix/prevent-cleanup-trap-override
sudheer-quad Aug 27, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
41 changes: 3 additions & 38 deletions tools/cloud-build/daily-tests/builds/ansible-vm.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -123,50 +123,15 @@ steps:
gsutil rm "gs://daily-tests-tf-state/*/\$$DEPLOYMENT_NAME/**.tflock" 2>/dev/null || true

RUN_CLEANUP=true
# Trap function that runs the rescue playbook (terraform destroy)
# if the pod is terminated by Kueue, fails, or is cancelled.
cleanup_pod() {
local exit_code=\$$?
trap - EXIT SIGTERM SIGINT ERR
set +e

if [ "\$$RUN_CLEANUP" = "false" ]; then
exit 0
fi
if [ \$$exit_code -eq 0 ]; then exit_code=1; fi

echo ""
echo "=========================================================================="
echo "CAUGHT SIGTERM OR SCRIPT ERROR!"
echo "Halting primary Ansible execution..."
echo "=========================================================================="
if [ -n "\$${ANSIBLE_PID:-}" ]; then
kill -TERM \$$ANSIBLE_PID 2>/dev/null || true
wait \$$ANSIBLE_PID 2>/dev/null || true
echo "Waiting 15s for Terraform to release GCS backend state locks..."
sleep 15
fi

echo ""
echo "INITIATING RESCUE PLAYBOOK: Destroying leaked infrastructure for \$$DEPLOYMENT_NAME..."
echo "- hosts: localhost" > /workspace/cleanup-playbook.yml
echo " tasks:" >> /workspace/cleanup-playbook.yml
echo " - ansible.builtin.include_tasks:" >> /workspace/cleanup-playbook.yml
echo " file: tools/cloud-build/daily-tests/ansible_playbooks/tasks/rescue_gcluster_failure.yml" >> /workspace/cleanup-playbook.yml

ansible-playbook /workspace/cleanup-playbook.yml -e deployment_name="\$$DEPLOYMENT_NAME" -e workspace="/workspace" || true

echo "Graceful cleanup finished."
exit \$$exit_code
}
source /workspace/tools/cloud-build/kueue_cleanup_pod.sh
trap cleanup_pod EXIT SIGTERM SIGINT

bash tools/get_binary.sh "\$$_TEST_PREFIX"
bash tools/get_binary.sh "${_TEST_PREFIX}"
BLUEPRINT="tools/cloud-build/daily-tests/blueprints/ansible-vm.yaml"
bash tools/add_ttl_label.sh \$$BLUEPRINT

ansible-playbook tools/cloud-build/daily-tests/ansible_playbooks/base-integration-test.yml \
--user="$$OSLOGIN_USER" --extra-vars="project=\$$PROJECT_ID build=$$BUILD_ID_SHORT full_build_id=$BUILD_ID" \
--user="$$OSLOGIN_USER" --extra-vars="project=$PROJECT_ID build=$$BUILD_ID_SHORT full_build_id=$BUILD_ID" \
--extra-vars="@tools/cloud-build/daily-tests/tests/ansible-vm.yml" \
--extra-vars="triage_gcs_bucket_override=$$TRIAGE_GCS_BUCKET" \
--extra-vars="triage_project_number_override=$$TRIAGE_PROJECT_NUMBER" \
Expand Down
37 changes: 1 addition & 36 deletions tools/cloud-build/daily-tests/builds/batch-mpi.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -128,42 +128,7 @@ steps:
gsutil rm "gs://daily-tests-tf-state/*/\$$DEPLOYMENT_NAME/**.tflock" 2>/dev/null || true

RUN_CLEANUP=true
# Trap function that runs the rescue playbook (terraform destroy)
# if the pod is terminated by Kueue, fails, or is cancelled.
cleanup_pod() {
local exit_code=\$$?
trap - EXIT SIGTERM SIGINT ERR
set +e

if [ "\$$RUN_CLEANUP" = "false" ]; then
exit 0
fi
if [ \$$exit_code -eq 0 ]; then exit_code=1; fi

echo ""
echo "=========================================================================="
echo "CAUGHT SIGTERM OR SCRIPT ERROR!"
echo "Halting primary Ansible execution..."
echo "=========================================================================="
if [ -n "\$${ANSIBLE_PID:-}" ]; then
kill -TERM \$$ANSIBLE_PID 2>/dev/null || true
wait \$$ANSIBLE_PID 2>/dev/null || true
echo "Waiting 15s for Terraform to release GCS backend state locks..."
sleep 15
fi

echo ""
echo "INITIATING RESCUE PLAYBOOK: Destroying leaked infrastructure for \$$DEPLOYMENT_NAME..."
echo "- hosts: localhost" > /workspace/cleanup-playbook.yml
echo " tasks:" >> /workspace/cleanup-playbook.yml
echo " - ansible.builtin.include_tasks:" >> /workspace/cleanup-playbook.yml
echo " file: tools/cloud-build/daily-tests/ansible_playbooks/tasks/rescue_gcluster_failure.yml" >> /workspace/cleanup-playbook.yml

ansible-playbook /workspace/cleanup-playbook.yml -e deployment_name="\$$DEPLOYMENT_NAME" -e workspace="/workspace" || true

echo "Graceful cleanup finished."
exit \$$exit_code
}
source /workspace/tools/cloud-build/kueue_cleanup_pod.sh
trap cleanup_pod EXIT SIGTERM SIGINT

bash tools/get_binary.sh "${_TEST_PREFIX}"
Expand Down
37 changes: 1 addition & 36 deletions tools/cloud-build/daily-tests/builds/chrome-remote-desktop-ubuntu.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -122,42 +122,7 @@ steps:
gsutil rm "gs://daily-tests-tf-state/*/\$$DEPLOYMENT_NAME/**.tflock" 2>/dev/null || true

RUN_CLEANUP=true
# Trap function that runs the rescue playbook (terraform destroy)
# if the pod is terminated by Kueue, fails, or is cancelled.
cleanup_pod() {
local exit_code=\$$?
trap - EXIT SIGTERM SIGINT ERR
set +e

if [ "\$$RUN_CLEANUP" = "false" ]; then
exit 0
fi
if [ \$$exit_code -eq 0 ]; then exit_code=1; fi

echo ""
echo "=========================================================================="
echo "CAUGHT SIGTERM OR SCRIPT ERROR!"
echo "Halting primary Ansible execution..."
echo "=========================================================================="
if [ -n "\$${ANSIBLE_PID:-}" ]; then
kill -TERM \$$ANSIBLE_PID 2>/dev/null || true
wait \$$ANSIBLE_PID 2>/dev/null || true
echo "Waiting 15s for Terraform to release GCS backend state locks..."
sleep 15
fi

echo ""
echo "INITIATING RESCUE PLAYBOOK: Destroying leaked infrastructure for \$$DEPLOYMENT_NAME..."
echo "- hosts: localhost" > /workspace/cleanup-playbook.yml
echo " tasks:" >> /workspace/cleanup-playbook.yml
echo " - ansible.builtin.include_tasks:" >> /workspace/cleanup-playbook.yml
echo " file: tools/cloud-build/daily-tests/ansible_playbooks/tasks/rescue_gcluster_failure.yml" >> /workspace/cleanup-playbook.yml

ansible-playbook /workspace/cleanup-playbook.yml -e deployment_name="\$$DEPLOYMENT_NAME" -e workspace="/workspace" || true

echo "Graceful cleanup finished."
exit \$$exit_code
}
source /workspace/tools/cloud-build/kueue_cleanup_pod.sh
trap cleanup_pod EXIT SIGTERM SIGINT

bash tools/get_binary.sh "${_TEST_PREFIX}"
Expand Down
37 changes: 1 addition & 36 deletions tools/cloud-build/daily-tests/builds/chrome-remote-desktop.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -120,42 +120,7 @@ steps:
gsutil rm "gs://daily-tests-tf-state/*/\$$DEPLOYMENT_NAME/**.tflock" 2>/dev/null || true

RUN_CLEANUP=true
# Trap function that runs the rescue playbook (terraform destroy)
# if the pod is terminated by Kueue, fails, or is cancelled.
cleanup_pod() {
local exit_code=\$$?
trap - EXIT SIGTERM SIGINT ERR
set +e

if [ "\$$RUN_CLEANUP" = "false" ]; then
exit 0
fi
if [ \$$exit_code -eq 0 ]; then exit_code=1; fi

echo ""
echo "=========================================================================="
echo "CAUGHT SIGTERM OR SCRIPT ERROR!"
echo "Halting primary Ansible execution..."
echo "=========================================================================="
if [ -n "\$${ANSIBLE_PID:-}" ]; then
kill -TERM \$$ANSIBLE_PID 2>/dev/null || true
wait \$$ANSIBLE_PID 2>/dev/null || true
echo "Waiting 15s for Terraform to release GCS backend state locks..."
sleep 15
fi

echo ""
echo "INITIATING RESCUE PLAYBOOK: Destroying leaked infrastructure for \$$DEPLOYMENT_NAME..."
echo "- hosts: localhost" > /workspace/cleanup-playbook.yml
echo " tasks:" >> /workspace/cleanup-playbook.yml
echo " - ansible.builtin.include_tasks:" >> /workspace/cleanup-playbook.yml
echo " file: tools/cloud-build/daily-tests/ansible_playbooks/tasks/rescue_gcluster_failure.yml" >> /workspace/cleanup-playbook.yml

ansible-playbook /workspace/cleanup-playbook.yml -e deployment_name="\$$DEPLOYMENT_NAME" -e workspace="/workspace" || true

echo "Graceful cleanup finished."
exit \$$exit_code
}
source /workspace/tools/cloud-build/kueue_cleanup_pod.sh
trap cleanup_pod EXIT SIGTERM SIGINT

bash tools/get_binary.sh "${_TEST_PREFIX}"
Expand Down
45 changes: 7 additions & 38 deletions tools/cloud-build/daily-tests/builds/e2e.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -102,38 +102,7 @@ steps:
gsutil rm "gs://daily-tests-tf-state/*/\$$DEPLOYMENT_NAME/**.tflock" 2>/dev/null || true

RUN_CLEANUP=true
cleanup_pod() {
local exit_code=\$$?
trap - EXIT SIGTERM SIGINT ERR
set +e

if [ "\$$RUN_CLEANUP" = "false" ]; then
exit 0
fi
if [ \$$exit_code -eq 0 ]; then exit_code=1; fi

echo ""
echo "=========================================================================="
echo "CAUGHT SIGTERM OR SCRIPT ERROR!"
echo "Halting execution..."
echo "=========================================================================="
if [ -n "\$${CMD_PID:-}" ]; then
kill -TERM \$$CMD_PID 2>/dev/null || true
wait \$$CMD_PID 2>/dev/null || true
fi

echo ""
echo "INITIATING RESCUE PLAYBOOK: Destroying leaked infrastructure for \$$DEPLOYMENT_NAME..."
echo "- hosts: localhost" > /workspace/cleanup-playbook.yml
echo " tasks:" >> /workspace/cleanup-playbook.yml
echo " - ansible.builtin.include_tasks:" >> /workspace/cleanup-playbook.yml
echo " file: tools/cloud-build/daily-tests/ansible_playbooks/tasks/rescue_gcluster_failure.yml" >> /workspace/cleanup-playbook.yml

ansible-playbook /workspace/cleanup-playbook.yml -e deployment_name="\$$DEPLOYMENT_NAME" -e workspace="/workspace" || true

echo "Graceful cleanup finished."
exit \$$exit_code
}
source /workspace/tools/cloud-build/kueue_cleanup_pod.sh
Comment thread
sudheer-quad marked this conversation as resolved.
trap cleanup_pod EXIT SIGTERM SIGINT

region="us-central1"
Expand All @@ -145,17 +114,17 @@ steps:
bash tools/add_ttl_label.sh "\$$BLUEPRINT"

./gcluster deploy "\$$BLUEPRINT" --vars="\$$vars" --auto-approve &
CMD_PID=\$$!
wait \$$CMD_PID
ANSIBLE_PID=\$$!
wait \$$ANSIBLE_PID

# check instance was created
gcloud compute instances describe "\$${DEPLOYMENT_NAME}-0" --project="\$$PROJECT_ID" --zone="\$$zone" >/dev/null &
CMD_PID=\$$!
wait \$$CMD_PID
ANSIBLE_PID=\$$!
wait \$$ANSIBLE_PID

./gcluster destroy "\$$DEPLOYMENT_NAME" --auto-approve --robust &
CMD_PID=\$$!
wait \$$CMD_PID
ANSIBLE_PID=\$$!
wait \$$ANSIBLE_PID

RUN_CLEANUP=false
resources:
Expand Down
19 changes: 9 additions & 10 deletions tools/cloud-build/daily-tests/builds/gke-g4-onspot.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -130,19 +130,10 @@ steps:
command: ["/bin/bash", "-c"]
args:
- |
set -e -u -o pipefail
echo "Sourcing find_available_zone.sh to determine zone."
source /workspace/tools/cloud-build/wait_for_available_zone.sh
if [ -z "\$${ZONE:-}" ]; then
echo "ERROR: ZONE not found" >&2
exit 1
fi
set -x -e
echo "\"test_file_path\": tools/cloud-build/daily-tests/builds/gke-g4-onspot.yaml"
cd /workspace

BUILD_ID_FULL=$$BUILD_ID
BUILD_ID_SHORT=\$${BUILD_ID_FULL:0:6}
DEPLOYMENT_NAME="gke-g4-spot-$$BUILD_ID_SHORT"

# Force clear any leaked Terraform state locks from previously preempted attempts of this test run.
Expand All @@ -153,6 +144,14 @@ steps:
source /workspace/tools/cloud-build/kueue_cleanup_pod.sh
trap cleanup_pod EXIT SIGTERM SIGINT

set -e -u -o pipefail
echo "Sourcing find_available_zone.sh to determine zone."
source /workspace/tools/cloud-build/wait_for_available_zone.sh
if [ -z "\$${ZONE:-}" ]; then
echo "ERROR: ZONE not found" >&2
exit 1
fi
set -x
bash tools/get_binary.sh "${_TEST_PREFIX}"
REGION="\$${ZONE%-*}"
EXAMPLE_BP=examples/gke-g4/gke-g4.yaml
Expand Down Expand Up @@ -183,7 +182,7 @@ steps:
cat \$$EXAMPLE_BP

ansible-playbook tools/cloud-build/daily-tests/ansible_playbooks/base-integration-test.yml \
--extra-vars="test_prefix=${_TEST_PREFIX}" \
--extra-vars="test_prefix=${_TEST_PREFIX}" \
--user="$$OSLOGIN_USER" \
--extra-vars="project=$PROJECT_ID build=$$BUILD_ID_SHORT full_build_id=$BUILD_ID" \
--extra-vars="region=\$$REGION zone=\$$ZONE" \
Expand Down
20 changes: 10 additions & 10 deletions tools/cloud-build/daily-tests/builds/gke-h4d-onspot.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -131,15 +131,7 @@ steps:
command: ["/bin/bash", "-c"]
args:
- |
set -e -u -o pipefail
echo "\"test_file_path\": tools/cloud-build/daily-tests/builds/gke-h4d-onspot.yaml"
echo "Sourcing find_available_zone.sh to determine zone."
source /workspace/tools/cloud-build/wait_for_available_zone.sh
if [ -z "\$${ZONE:-}" ]; then
echo "ERROR: ZONE not found" >&2
exit 1
fi
set -x
set -x -e
echo "\"test_file_path\": tools/cloud-build/daily-tests/builds/gke-h4d-onspot.yaml"
cd /workspace

Expand All @@ -153,8 +145,16 @@ steps:
source /workspace/tools/cloud-build/kueue_cleanup_pod.sh
trap cleanup_pod EXIT SIGTERM SIGINT

set -e -u -o pipefail
echo "\"test_file_path\": tools/cloud-build/daily-tests/builds/gke-h4d-onspot.yaml"
echo "Sourcing find_available_zone.sh to determine zone."
source /workspace/tools/cloud-build/wait_for_available_zone.sh
if [ -z "\$${ZONE:-}" ]; then
echo "ERROR: ZONE not found" >&2
exit 1
fi
set -x
bash tools/get_binary.sh "${_TEST_PREFIX}"

REGION="\$${ZONE%-*}"
EXAMPLE_BP="examples/gke-h4d/gke-h4d.yaml"

Expand Down
37 changes: 1 addition & 36 deletions tools/cloud-build/daily-tests/builds/gke-inactive-reservation.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -125,42 +125,7 @@ steps:
gsutil rm "gs://daily-tests-tf-state/*/\$$DEPLOYMENT_NAME/**.tflock" 2>/dev/null || true

RUN_CLEANUP=true
# Trap function that runs the rescue playbook (terraform destroy)
# if the pod is terminated by Kueue, fails, or is cancelled.
cleanup_pod() {
local exit_code=\$$?
trap - EXIT SIGTERM SIGINT ERR
set +e

if [ "\$$RUN_CLEANUP" = "false" ]; then
exit 0
fi
if [ \$$exit_code -eq 0 ]; then exit_code=1; fi

echo ""
echo "=========================================================================="
echo "CAUGHT SIGTERM OR SCRIPT ERROR!"
echo "Halting primary Ansible execution..."
echo "=========================================================================="
if [ -n "\$${ANSIBLE_PID:-}" ]; then
kill -TERM \$$ANSIBLE_PID 2>/dev/null || true
wait \$$ANSIBLE_PID 2>/dev/null || true
echo "Waiting 15s for Terraform to release GCS backend state locks..."
sleep 15
fi

echo ""
echo "INITIATING RESCUE PLAYBOOK: Destroying leaked infrastructure for \$$DEPLOYMENT_NAME..."
echo "- hosts: localhost" > /workspace/cleanup-playbook.yml
echo " tasks:" >> /workspace/cleanup-playbook.yml
echo " - ansible.builtin.include_tasks:" >> /workspace/cleanup-playbook.yml
echo " file: tools/cloud-build/daily-tests/ansible_playbooks/tasks/rescue_gcluster_failure.yml" >> /workspace/cleanup-playbook.yml

ansible-playbook /workspace/cleanup-playbook.yml -e deployment_name="\$$DEPLOYMENT_NAME" -e workspace="/workspace" || true

echo "Graceful cleanup finished."
exit \$$exit_code
}
source /workspace/tools/cloud-build/kueue_cleanup_pod.sh
trap cleanup_pod EXIT SIGTERM SIGINT

bash tools/get_binary.sh "\$$_TEST_PREFIX"
Expand Down
Loading
Loading