feat(selenium-grid): handle selenium-grid autoscaler node type

ansible/configure-selenium-grid-local-oracle.yml

@@ -108,6 +108,8 @@
           cloud: "oracle"
     - { role: "jitsi-torture-assets", tags: "torture-assets", when: selenium_grid_enable_nomad == "true" }
     - { role: "consul-agent", tags: "consul", consul_install_flag: false, when: selenium_grid_enable_nomad == "true"}
+    - { role: "autoscaler-sidecar", tags: "autoscaler-sidecar",
+        autoscaler_instance_type: "selenium-grid", cloud_provider: "oracle", autoscaler_install_flag: false, autoscaler_configure_flag: true, when: nomad_enable_jitsi_autoscaler == "true"}
     - { role: "docker-daemon-config", tags: "docker-daemon-config", when: selenium_grid_enable_nomad == "true" }
     - { role: "docker-auth-config", tags: "docker-auth-config", docker_ecr_credentials: "{{ nomad_docker_ecr_credentials }}", when: selenium_grid_enable_nomad == "true" }
     - { role: "nomad-jitsi", tags: "nomad-jitsi", when: selenium_grid_enable_nomad == "true" }

ansible/nomad-client.yml

@@ -146,7 +146,7 @@
     - { role: "journald", tags: "journald"}
     - { role: "rsyslog", rsyslog_install_flag: false, tags: "rsyslog", rsyslog_tcp_output_enabled: true}
     - { role: "autoscaler-sidecar", tags: "autoscaler-sidecar",
-        autoscaler_instance_type: "nomad", cloud_provider: "oracle", autoscaler_install_flag: false, autoscaler_configure_flag: true, when: nomad_enable_jitsi_autoscaler}
+        autoscaler_instance_type: "nomad", cloud_provider: "oracle", autoscaler_install_flag: false, autoscaler_configure_flag: true, when: nomad_enable_jitsi_autoscaler == "true"}
     - { role: "consul-agent", tags: "consul", consul_install_flag: false}
     - { role: "nvidia-docker", tags: "nvidia-docker", when: nvidia_docker_flag }
     - { role: "docker-daemon-config", tags: "docker-daemon-config" }

ansible/roles/autoscaler-sidecar/defaults/main.yml

@@ -16,6 +16,7 @@ autoscaler_force_shutdown_command_by_type:
   'jigasi': 'service jigasi stop'
   'sip-jibri': 'service jibri stop'
   'nomad': 'nomad node drain -self -enable -force -detach -yes'
+  'selenium-grid': 'nomad node drain -self -enable -force -detach -yes'
 autoscaler_git_repo: https://github.com/jitsi/jitsi-autoscaler-sidecar.git
 autoscaler_graceful_shutdown_script: "{{
   autoscaler_graceful_shutdown_scripts[autoscaler_instance_type]
@@ -26,6 +27,7 @@ autoscaler_graceful_shutdown_scripts:
   'jigasi': /usr/local/bin/graceful-shutdown-wrapper-jigasi.sh
   'sip-jibri': /opt/jitsi/jibri/wait_graceful_shutdown.sh
   'nomad': /usr/local/bin/nomad_graceful_shutdown.sh
+  'selenium-grid': /usr/local/bin/graceful_shutdown_selenium_grid.sh
 autoscaler_graceful_shutdown_wrapper: /usr/local/bin/graceful_shutdown_wrapper.sh
 autoscaler_group: "default"
 autoscaler_groupname: jsidecar
@@ -67,6 +69,15 @@ autoscaler_sidecar_instance_version_command_by_type:
   'jigasi': "dpkg -s jigasi | grep Version | awk '{print $2}' | sed 's/..$//'"
   'sip-jibri': "dpkg -s jibri | grep Version | awk '{print $2}' | sed 's/..$//'"
   'nomad': "nomad --version | head -1 | cut -d' ' -f2 | tr -d 'v'"
+  'selenium-grid': >-
+    NODE_ID=$(curl -s http://localhost:4646/v1/agent/self |
+    jq -r '.stats.client.node_id') &&
+    JOB_ID=$(curl -s http://localhost:4646/v1/node/$NODE_ID/allocations |
+    jq -r '[.[] | select(.JobID | contains("grid-node")) |
+    select(.ClientStatus=="running")][0].JobID') &&
+    curl -s http://localhost:4646/v1/job/$JOB_ID |
+    jq -r '.TaskGroups[].Tasks[].Config.image' |
+    head -1 | rev | cut -d: -f1 | rev
 autoscaler_sidecar_private_ip: "{{ ansible_default_ipv4.address }}"
 autoscaler_sidecar_public_ip: "{{ ansible_ec2_public_ipv4 if ansible_ec2_public_ipv4 is defined else false }}"
 autoscaler_sidecar_service_name: jitsi-autoscaler-sidecar
@@ -80,6 +91,7 @@ autoscaler_stats_retrieve_urls:
   'jibri': http://localhost:2222/jibri/api/v1.0/health
   'jigasi': http://localhost:8788/about/stats
   'sip-jibri': http://localhost:2222/jibri/api/v1.0/health
+  'selenium-grid': http://localhost:4444/status
 autoscaler_status_url: "https://{{ autoscaler_server_host }}/sidecar/status"
 autoscaler_shutdown_url: "https://{{ autoscaler_server_host }}/sidecar/shutdown"
 autoscaler_terminate_script: '/usr/local/bin/terminate_instance.sh'

ansible/roles/autoscaler-sidecar/files/graceful_shutdown_selenium_grid.sh

@@ -0,0 +1,59 @@
+#!/bin/bash
+
+# Graceful shutdown for selenium-grid nodes running under nomad
+# 1. Drain the Selenium Grid 4 node (stop accepting new sessions, wait for existing to finish)
+# 2. Then perform the normal nomad graceful shutdown
+
+SELENIUM_NODE_PORT=${SELENIUM_NODE_PORT:-5555}
+SELENIUM_DRAIN_URL="http://localhost:${SELENIUM_NODE_PORT}/se/grid/node/drain"
+SELENIUM_STATUS_URL="http://localhost:${SELENIUM_NODE_PORT}/status"
+SELENIUM_DRAIN_TIMEOUT=${SELENIUM_DRAIN_TIMEOUT:-300}
+SLEEP_TIME=10
+
+echo "Starting selenium-grid graceful shutdown"
+
+# Step 1: Issue drain command to the selenium grid node
+echo "Sending drain request to Selenium Grid node at ${SELENIUM_DRAIN_URL}"
+DRAIN_RESPONSE=$(curl -s -o /dev/null -w "%{http_code}" -X POST "${SELENIUM_DRAIN_URL}")
+
+if [[ "$DRAIN_RESPONSE" == "200" ]]; then
+    echo "Drain request accepted by Selenium Grid node"
+else
+    echo "Warning: Drain request returned HTTP ${DRAIN_RESPONSE}, continuing with shutdown"
+fi
+
+# Step 2: Wait for active sessions to complete
+echo "Waiting for active sessions to drain (timeout: ${SELENIUM_DRAIN_TIMEOUT}s)"
+ELAPSED=0
+while [[ $ELAPSED -lt $SELENIUM_DRAIN_TIMEOUT ]]; do
+    STATUS=$(curl -s "${SELENIUM_STATUS_URL}" 2>/dev/null)
+    if [[ $? -ne 0 ]]; then
+        echo "Node status endpoint unreachable, assuming drained"
+        break
+    fi
+
+    # Check if node has any active sessions
+    ACTIVE_SESSIONS=$(echo "$STATUS" | jq -r '.value.node.slots // [] | map(select(.session != null)) | length' 2>/dev/null)
+    if [[ $? -ne 0 ]] || [[ -z "$ACTIVE_SESSIONS" ]]; then
+        echo "Could not parse session count, assuming drained"
+        break
+    fi
+
+    if [[ "$ACTIVE_SESSIONS" -eq 0 ]]; then
+        echo "All sessions drained from Selenium Grid node"
+        break
+    fi
+
+    echo "Still waiting for ${ACTIVE_SESSIONS} active session(s) to complete... (${ELAPSED}s/${SELENIUM_DRAIN_TIMEOUT}s)"
+    sleep $SLEEP_TIME
+    ELAPSED=$((ELAPSED + SLEEP_TIME))
+done
+
+if [[ $ELAPSED -ge $SELENIUM_DRAIN_TIMEOUT ]]; then
+    echo "Selenium drain timeout reached after ${SELENIUM_DRAIN_TIMEOUT}s, proceeding with nomad shutdown"
+fi
+
+# Step 3: Proceed with normal nomad graceful shutdown
+echo "Starting nomad graceful shutdown"
+/usr/local/bin/nomad_graceful_shutdown.sh
+exit $?

ansible/roles/autoscaler-sidecar/tasks/configure.yml

@@ -36,6 +36,13 @@
     regexp: "^%{{ autoscaler_groupname }}"
     line: "%{{ autoscaler_groupname }} ALL=(ALL) NOPASSWD: ALL"
 
+- name: Copy selenium-grid graceful shutdown script
+  ansible.builtin.copy:
+    dest: /usr/local/bin/graceful_shutdown_selenium_grid.sh
+    src: "graceful_shutdown_selenium_grid.sh"
+    mode: 0755
+  when: autoscaler_instance_type == 'selenium-grid'
+
 - name: Copy graceful shutdown wrapper script
   ansible.builtin.template:
     dest: "{{ autoscaler_graceful_shutdown_wrapper }}"

ansible/roles/jitsi-torture-assets/tasks/main.yml

@@ -4,6 +4,7 @@
     repo: "{{ jitsi_torture_assets_git_repo }}"
     version: "{{ jitsi_torture_assets_git_branch }}"
     dest: "{{ jitsi_torture_assets_path }}"
+    force: true
 
 - name: Ensure directory for assets
   ansible.builtin.file: