From 74fc9445612a81de088146ea61b6ea07f44ca91c Mon Sep 17 00:00:00 2001
From: fewensa <37804932+fewensa@users.noreply.github.com>
Date: Mon, 29 Jul 2024 10:17:26 +0800
Subject: [PATCH] Monitor feature (#8)

* Add monitor

* monitor script

* monitor

* check ansible

* Test monitor
---
 .editorconfig                                 |   2 +-
 .github/workflows/check-ansible.yml           |  55 ++++++
 .github/workflows/monitor.yml                 |  25 +++
 ansible/inventories/hosts.ini                 |  23 +++
 .../playbooks/monitor/group_vars/monitor.yml  |   3 +
 ansible/playbooks/monitor/playbook.yml        |   3 +
 ansible/roles/monitor/defaults/main.yml       |  16 ++
 ansible/roles/monitor/tasks/main.yml          |  15 ++
 ansible/roles/monitor/templates/crawl.sh      | 168 ++++++++++++++++++
 9 files changed, 309 insertions(+), 1 deletion(-)
 create mode 100644 .github/workflows/check-ansible.yml
 create mode 100644 .github/workflows/monitor.yml
 create mode 100644 ansible/playbooks/monitor/group_vars/monitor.yml
 create mode 100644 ansible/playbooks/monitor/playbook.yml
 create mode 100644 ansible/roles/monitor/defaults/main.yml
 create mode 100644 ansible/roles/monitor/tasks/main.yml
 create mode 100755 ansible/roles/monitor/templates/crawl.sh

diff --git a/.editorconfig b/.editorconfig
index 24df5de..4de45b4 100644
--- a/.editorconfig
+++ b/.editorconfig
@@ -5,7 +5,7 @@ charset = utf-8
 end_of_line = lf
 insert_final_newline = true
 indent_style = space
-indent_size = 4
+indent_size = 2
 trim_trailing_whitespace = true
 
 [*.{yml,yaml}]
diff --git a/.github/workflows/check-ansible.yml b/.github/workflows/check-ansible.yml
new file mode 100644
index 0000000..02d8ad0
--- /dev/null
+++ b/.github/workflows/check-ansible.yml
@@ -0,0 +1,55 @@
+name: Check
+
+on:
+  pull_request:
+    branches: [main]
+
+env:
+  SLACK_INCOMING_WEBHOOK_URL: ${{ secrets.SLACK_INCOMING_WEBHOOK_URL }}
+
+jobs:
+  deploy-essentials:
+    name: Deploy essentials
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v2
+        with:
+          fetch-depth: 0
+
+      - name: Verify essentials file changed
+        uses: tj-actions/changed-files@v37.4.0
+        id: changed_files
+        with:
+          files: |
+            ansible/inventories/hosts.ini
+            ansible/playbooks/_essentials/*
+            ansible/playbooks/essentials.yml
+
+      - name: Deploy essentials
+        id: deploy-essentials
+        if: steps.changed_files.outputs.any_changed == 'true'
+        uses: dawidd6/action-ansible-playbook@v2
+        with:
+          directory: ansible
+          playbook: playbooks/essentials.yml
+          key: "${{ secrets.SSH_PRIVATE_KEY }}"
+          options: --user ansible
+
+  check-playbooks:
+    name: Check playbook
+    runs-on: ubuntu-latest
+    needs: [deploy-essentials]
+    strategy:
+      matrix:
+        playbook:
+          - snapshots_crab
+    steps:
+      - uses: actions/checkout@v2
+
+      - name: Run playbook
+        uses: dawidd6/action-ansible-playbook@v2
+        with:
+          directory: ansible
+          playbook: playbooks/${{ matrix.playbook }}/playbook.yml
+          key: "${{ secrets.SSH_PRIVATE_KEY }}"
+          options: --user ansible --verbose --diff --check
diff --git a/.github/workflows/monitor.yml b/.github/workflows/monitor.yml
new file mode 100644
index 0000000..7666d50
--- /dev/null
+++ b/.github/workflows/monitor.yml
@@ -0,0 +1,25 @@
+name: Monitor
+
+on:
+  schedule:
+    - cron: "0 */2 * * *"
+  workflow_dispatch:
+
+jobs:
+  monitor:
+    name: monitor
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        playbook:
+          - monitor
+    steps:
+      - uses: actions/checkout@v2
+
+      - name: Run playbook
+        uses: dawidd6/action-ansible-playbook@v2
+        with:
+          directory: ansible
+          playbook: playbooks/${{ matrix.playbook }}/playbook.yml
+          key: "${{ secrets.SSH_PRIVATE_KEY }}"
+          options: --user ansible --verbose --diff
diff --git a/ansible/inventories/hosts.ini b/ansible/inventories/hosts.ini
index f69dd59..d651e74 100644
--- a/ansible/inventories/hosts.ini
+++ b/ansible/inventories/hosts.ini
@@ -4,3 +4,26 @@ g1.crab2.darwinia.network
 [darwinia_nodes]
 g1.darwinia2.darwinia.network
 
+[monitor]
+g1.crab2.darwinia.network
+c1.crab2.darwinia.network
+g1.darwinia2.darwinia.network
+c1.darwinia2.darwinia.network
+g1.testnets.darwinia.network
+g2.testnets.darwinia.network
+g3.testnets.darwinia.network
+g1.generic.darwinia.network
+g2.generic.darwinia.network
+
+c1.darwinia-rpc.itering.io
+c2.darwinia-rpc.itering.io
+c1.crab-rpc.itering.io
+c2.crab-rpc.itering.io
+
+c1.collator.itering.io
+c2.collator.itering.io
+c3.collator.itering.io
+c4.collator.itering.io
+
+c5.collator.itering.io
+c6.collator.itering.io
diff --git a/ansible/playbooks/monitor/group_vars/monitor.yml b/ansible/playbooks/monitor/group_vars/monitor.yml
new file mode 100644
index 0000000..4ba9821
--- /dev/null
+++ b/ansible/playbooks/monitor/group_vars/monitor.yml
@@ -0,0 +1,3 @@
+
+monitor:
+  notify_slack_webhook: "{{ lookup('env', 'SLACK_INCOMING_WEBHOOK_URL') }}"
diff --git a/ansible/playbooks/monitor/playbook.yml b/ansible/playbooks/monitor/playbook.yml
new file mode 100644
index 0000000..fdefde7
--- /dev/null
+++ b/ansible/playbooks/monitor/playbook.yml
@@ -0,0 +1,3 @@
+- hosts: monitor
+  roles:
+    - monitor
diff --git a/ansible/roles/monitor/defaults/main.yml b/ansible/roles/monitor/defaults/main.yml
new file mode 100644
index 0000000..59ed131
--- /dev/null
+++ b/ansible/roles/monitor/defaults/main.yml
@@ -0,0 +1,16 @@
+
+monitor:
+  workdir: /tmp/monitor
+  notify_slack_webhook: ''
+  notify_slack_channel: 'darwinia-alert-notification'
+  server_name: ''
+  check_disks:
+    - /dev/sda
+    - /dev/sdb
+  alert_thread_cpu_p2: 90
+  alert_thread_cpu_p1: 98
+  alert_thread_ram_p2: 90
+  alert_thread_ram_p1: 98
+  alert_thread_disk_p2: 90
+  alert_thread_disk_p1: 98
+
diff --git a/ansible/roles/monitor/tasks/main.yml b/ansible/roles/monitor/tasks/main.yml
new file mode 100644
index 0000000..489dfc4
--- /dev/null
+++ b/ansible/roles/monitor/tasks/main.yml
@@ -0,0 +1,15 @@
+
+- name: Creates workdir
+  file:
+    path: "{{ monitor.workdir }}"
+    state: directory
+
+- name: Generate scripts file
+  template:
+    src: crawl.sh
+    dest: "{{ monitor.workdir }}/crawl.sh"
+    mode: "0644"
+
+- name: Run snapshot
+  command: bash {{ monitor.workdir }}/crawl.sh
+
diff --git a/ansible/roles/monitor/templates/crawl.sh b/ansible/roles/monitor/templates/crawl.sh
new file mode 100755
index 0000000..b149b52
--- /dev/null
+++ b/ansible/roles/monitor/templates/crawl.sh
@@ -0,0 +1,168 @@
+#!/bin/bash
+
+DISKS_TO_MONITOR=({{ monitor.check_disks | join(' ') }})
+
+SERVER_NAME='{{ monitor.server_name }}'
+NOTIFY_SLACK_WEBHOOK='{{ monitor.notify_slack_webhook }}'
+NOTIFY_SLACK_CHANNEL='{{ monitor.notify_slack_channel }}'
+
+ALERT_THREAD_CPU_P2={{ monitor.alert_thread_cpu_p2 }}
+ALERT_THREAD_CPU_P1={{ monitor.alert_thread_cpu_p1 }}
+ALERT_THREAD_RAM_P2={{ monitor.alert_thread_ram_p2 }}
+ALERT_THREAD_RAM_P1={{ monitor.alert_thread_ram_p1 }}
+ALERT_THREAD_DISK_P2={{ monitor.alert_thread_disk_p2 }}
+ALERT_THREAD_DISK_P1={{ monitor.alert_thread_disk_p1 }}
+
+timestamp() {
+  date +"%Y-%m-%d %H:%M:%S"
+}
+
+cpu_usage() {
+  top -bn1 | grep "Cpu(s)" | awk '{print $2 + $4}' | sed 's/%//'
+}
+
+memory_usage() {
+  free | grep Mem | awk '{print $3/$2 * 100.0}'
+}
+
+disk_usage() {
+  for disk in "${DISKS_TO_MONITOR[@]}"; do
+    usage=$(df -h | grep "^$disk" | awk '{print $5}' | sed 's/%//')
+    echo "$disk $usage"
+  done
+}
+
+request_count() {
+  ss -s | grep 'estab' | awk '{print $2}'
+}
+
+generate_alert_message() {
+  local cpu=$(cpu_usage)
+  local ram=$(memory_usage)
+  local tcp=$(request_count)
+  local alert_message="[]"
+  local priority='P2'
+
+  if (( $(echo "$cpu > $ALERT_THREAD_CPU_P1" | bc -l) )); then
+    priority='P1'
+  fi
+  if (( $(echo "$ram > $ALERT_THREAD_RAM_P1" | bc -l) )); then
+    priority='P1'
+  fi
+  if [[ "P1" == "$priority" ]]; then
+    priority_alert=$(jq -n --arg priority "${priority}" '[{"type":"mrkdwn","text":"*Priority*"},{"type":"plain_text","text":$priority}]')
+    alert_message=$(echo "$alert_message" | jq --argjson priority_alert "$priority_alert" '. += $priority_alert')
+  fi
+
+  if (( $(echo "$cpu > $ALERT_THREAD_CPU_P2" | bc -l) )); then
+    cpu_alert=$(jq -n --arg cpu "${cpu}%" '[{"type":"mrkdwn","text":"*CPU*"},{"type":"plain_text","text":$cpu}]')
+    alert_message=$(echo "$alert_message" | jq --argjson cpu_alert "$cpu_alert" '. += $cpu_alert')
+  fi
+
+  if (( $(echo "$ram > $ALERT_THREAD_RAM_P2" | bc -l) )); then
+    ram_alert=$(jq -n --arg ram "${ram}%" '[{"type":"mrkdwn","text":"*RAM*"},{"type":"plain_text","text":$ram}]')
+    alert_message=$(echo "$alert_message" | jq --argjson ram_alert "$ram_alert" '. += $ram_alert')
+  fi
+
+  if [[ "$alert_message" != "[]" ]]; then
+    tcp_alert=$(jq -n --arg tcp "${tcp}" '[{"type":"mrkdwn","text":"*TCP*"},{"type":"plain_text","text":$tcp}]')
+    alert_message=$(echo "$alert_message" | jq --argjson tcp_alert "$tcp_alert" '. += $tcp_alert')
+  fi
+
+  echo "$alert_message"
+}
+
+
+generate_disk_alert_message() {
+  local alert_message="[]"
+  local priority='P2'
+
+  while IFS= read -r line; do
+    local disk=$(echo $line | awk '{print $1}')
+    local usage=$(echo $line | awk '{print $2}')
+    if [[ -z "$usage" ]]; then
+      continue
+    fi
+
+    if (( $(echo "$usage > $ALERT_THREAD_DISK_P1" | bc -l) )); then
+      priority='P1'
+    fi
+    if (( $(echo "$usage > $ALERT_THREAD_DISK_P2" | bc -l) )); then
+      disk_alert=$(jq -n --arg disk "*DISK* ($disk)" --arg usage "${usage}%" '[{"type":"mrkdwn","text":$disk},{"type":"plain_text","text":$usage}]')
+      alert_message=$(echo "$alert_message" | jq --argjson disk_alert "$disk_alert" '. += $disk_alert')
+    fi
+  done < <(disk_usage)
+
+  if [[ "P1" == "$priority" ]]; then
+    priority_alert=$(jq -n --arg priority "${priority}" '[{"type":"mrkdwn","text":"*Priority*"},{"type":"plain_text","text":$priority}]')
+    alert_message=$(echo "$alert_message" | jq --argjson priority_alert "$priority_alert" '. += $priority_alert')
+  fi
+
+  echo "$alert_message"
+}
+
+
+check_and_send_alert() {
+  local alert_message=$(generate_alert_message)
+  local disk_alert_message=$(generate_disk_alert_message)
+  local HOSTNAME=${SERVER_NAME:-$(hostname)}
+
+  local blocks="[]"
+
+  if [[ "$alert_message" != "[]" ]]; then
+    alert_block=$(
+      jq -n \
+        --arg warning "[*WARNING*]: New server alert > $HOSTNAME" \
+        --argjson msg "$alert_message" \
+        '{ "type": "section", "text": {"type": "mrkdwn", "text": $warning}, "fields": $msg }'
+    )
+    blocks=$(echo "$blocks" | jq --argjson block "$alert_block" '. += [$block]')
+  fi
+
+  if [[ "$disk_alert_message" != "[]" ]]; then
+    disk_block=$(
+      jq -n \
+        --arg warning "[*WARNING*]: New disk alert > $HOSTNAME" \
+        --argjson msg "$disk_alert_message" \
+        '{ "type": "section", "text": {"type": "mrkdwn", "text": $warning}, "fields": $msg }'
+    )
+    blocks=$(echo "$blocks" | jq --argjson block "$disk_block" '. += [$block]')
+  fi
+
+  if [[ "$blocks" != "[]" ]]; then
+    local data=$(
+      jq -n \
+        --arg channel "$NOTIFY_SLACK_CHANNEL" \
+        --argjson blocks "$blocks" \
+        '{
+          "username": "ServerBot",
+          "icon_emoji": ":loudspeaker:",
+          "channel": $channel,
+          "blocks": $blocks
+        }'
+    )
+
+    send_alert "$data"
+  fi
+}
+
+send_alert() {
+  local message=$1
+
+  curl -X POST \
+    -H "Content-type: application/json" \
+    $NOTIFY_SLACK_WEBHOOK \
+    --data "$message"
+}
+
+main() {
+  local cpu=$(cpu_usage)
+  local ram=$(memory_usage)
+  local disk=$(disk_usage)
+  local requests=$(request_count)
+  echo "$(timestamp) CPU: ${cpu}% RAM: ${ram}% Disk: ${disk}% Requests: ${requests}"
+
+  check_and_send_alert
+}
+
+main