[#37] 增加sys/swarm_tests(cpu) ；单独构建的node bundle镜像

2025-11-06 16:43:14 +08:00
parent 94b3e910b3
commit d1fad4a05a
34 changed files with 2494 additions and 32 deletions
--- a/deployment/build/build_images.sh
+++ b/deployment/build/build_images.sh
@@ -0,0 +1,98 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
+. "$ROOT_DIR/deployment/build/common.sh"
+
+usage() {
+cat <<EOF
+Build Argus images (optional node-bundle)
+
+Usage: build_images.sh [--with-node-bundle] [--client-version YYYYMMDD] [--base-image NAME[:TAG]]
+
+Examples:
+  ./deployment/build/build_images.sh --with-node-bundle --client-version 20251106
+EOF
+}
+
+WITH_BUNDLE=false
+CLIENT_VERSION=""
+BASE_IMAGE="argus-sys-metric-test-node:latest"
+while [[ $# -gt 0 ]]; do
+  case "$1" in
+    --with-node-bundle) WITH_BUNDLE=true; shift;;
+    --client-version) CLIENT_VERSION="$2"; shift 2;;
+    --base-image) BASE_IMAGE="$2"; shift 2;;
+    -h|--help) usage; exit 0;;
+    *) err "unknown arg: $1"; usage; exit 1;;
+  esac
+done
+
+if [[ "$WITH_BUNDLE" == true ]]; then
+  require_cmd docker tar gzip
+  BUNDLE_DIR="$ROOT_DIR/src/sys/build/node-bundle"
+  CTX_DIR="$BUNDLE_DIR"
+  TMP_BUNDLE="$BUNDLE_DIR/bundle"
+  rm -rf "$TMP_BUNDLE"; mkdir -p "$TMP_BUNDLE"
+
+  # Build or locate client artifact
+  PLUGIN_DIR="$ROOT_DIR/src/metric/client-plugins/all-in-one-full"
+  # CLIENT_VERSION 支持两种形式：
+  #  - 形如 1.42.0 的 artifact 版本（默认）
+  #  - 形如 YYYYMMDD 的打包日期，将从 deployment/artifact/client/ 下解析出内部 artifact 版本
+  if [[ -z "$CLIENT_VERSION" ]]; then
+    pushd "$PLUGIN_DIR" >/dev/null
+    bash scripts/package_artifact.sh --force
+    CLIENT_VERSION=$(cat artifact/*/version.json 2>/dev/null | sed -n 's/.*"version"\s*:\s*"\([^"]\+\)".*/\1/p' | tail -n1)
+    popd >/dev/null
+    [[ -n "$CLIENT_VERSION" ]] || { err "failed to detect client version"; exit 1; }
+  else
+    if [[ "$CLIENT_VERSION" =~ ^[0-9]{8}$ ]]; then
+      PKG_DIR="$ROOT_DIR/deployment/artifact/client/$CLIENT_VERSION"
+      TAR_PKG="$PKG_DIR/argus-metric_${CLIENT_VERSION}.tar.gz"
+      [[ -f "$TAR_PKG" ]] || { err "client date package not found: $TAR_PKG"; exit 1; }
+      # 解包读取内部 version.json
+      tmpd=$(mktemp -d); trap 'rm -rf "$tmpd"' EXIT
+      tar -xzf "$TAR_PKG" -C "$tmpd"
+      if [[ -f "$tmpd/version.json" ]]; then
+        ART_VER=$(sed -n 's/.*"version"\s*:\s*"\([^"]\+\)".*/\1/p' "$tmpd/version.json" | head -n1)
+        [[ -n "$ART_VER" ]] || { err "failed to parse artifact version from date package"; exit 1; }
+        CLIENT_VERSION="$ART_VER"
+        # 直接使用该 tar 作为 bundle 源
+        cp "$TAR_PKG" "$TMP_BUNDLE/argus-metric_$(echo "$ART_VER" | tr '.' '_').tar.gz"
+        # 同时尝试复制 setup.sh（若存在）
+        [[ -f "$PKG_DIR/setup.sh" ]] && cp "$PKG_DIR/setup.sh" "$TMP_BUNDLE/" || true
+      else
+        err "version.json missing in client date package"
+        exit 1
+      fi
+    else
+      # 假定为 artifact 版本目录
+      pushd "$PLUGIN_DIR" >/dev/null
+      [[ -d "artifact/$CLIENT_VERSION" ]] || bash scripts/package_artifact.sh --force
+      popd >/dev/null
+    fi
+  fi
+
+  # 若未通过日期包预置 tar，则从插件 artifact 目录取
+  TAR_NAME="argus-metric_$(echo "$CLIENT_VERSION" | tr '.' '_').tar.gz"
+  if [[ ! -f "$TMP_BUNDLE/$TAR_NAME" ]]; then
+    SRC_TAR="$PLUGIN_DIR/artifact/$CLIENT_VERSION/$TAR_NAME"
+    [[ -f "$SRC_TAR" ]] || { err "missing client tar: $SRC_TAR"; exit 1; }
+    cp "$SRC_TAR" "$TMP_BUNDLE/"
+    # also include setup.sh for fallback
+    if [[ -f "$ROOT_DIR/deployment/artifact/client/$(today_version)/setup.sh" ]]; then
+      cp "$ROOT_DIR/deployment/artifact/client/$(today_version)/setup.sh" "$TMP_BUNDLE/" || true
+    fi
+  fi
+
+  log "Building node-bundle image with client version: $CLIENT_VERSION"
+  DOCKER_BUILDKIT=0 docker build \
+    --build-arg CLIENT_VER="$CLIENT_VERSION" \
+    --build-arg BASE_IMAGE="$BASE_IMAGE" \
+    -t argus-sys-metric-test-node-bundle:latest \
+    -f "$BUNDLE_DIR/Dockerfile" "$BUNDLE_DIR"
+  log "Built image: argus-sys-metric-test-node-bundle:latest"
+fi
+
+log "Done."
--- a/deployment/build/templates/scripts/fix-prom-targets-overlay.sh
+++ b/deployment/build/templates/scripts/fix-prom-targets-overlay.sh
@@ -0,0 +1,103 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+# Quick fix tool: replace 172.22/16 targets in nodes.json with overlay IPs resolved from hostname.
+# Usage: run on server package host: scripts/fix-prom-targets-overlay.sh
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+ROOT="$(cd "$SCRIPT_DIR/.." && pwd)"
+NODES_JSON="$ROOT/private/argus/metric/prometheus/nodes.json"
+
+require_cmd() { command -v "$1" >/dev/null 2>&1 || { echo "[ERROR] missing command: $1" >&2; exit 1; }; }
+
+backup() {
+  local src="$1"; local ts; ts=$(date -u +%Y%m%d-%H%M%SZ)
+  cp "$src" "${src%.json}_bak_${ts}.json"
+}
+
+prefer_overlay_ip() {
+  local host="$1"
+  # prefer 10.0/8 then 172.31/16
+  getent hosts "$host" | awk '{print $1}' | while read -r ip; do
+    if [[ "$ip" =~ ^10\. ]]; then echo "$ip"; return; fi
+  done
+  getent hosts "$host" | awk '{print $1}' | while read -r ip; do
+    if [[ "$ip" =~ ^172\.31\. ]]; then echo "$ip"; return; fi
+  done
+  # fallback: first A record
+  getent hosts "$host" | awk '{print $1; exit}'
+}
+
+require_cmd awk
+require_cmd sed
+
+if [[ ! -f "$NODES_JSON" ]]; then
+  echo "[WARN] nodes.json not found: $NODES_JSON" >&2
+  exit 0
+fi
+
+backup "$NODES_JSON"
+
+tmp=$(mktemp)
+trap 'rm -f "$tmp"' EXIT
+
+changed=0
+python3 - "$NODES_JSON" <<'PY' > "$tmp" || {
+import ipaddress, json, sys, socket
+path=sys.argv[1]
+data=json.load(open(path)) if path else []
+def resolve(host):
+    try:
+        infos=socket.getaddrinfo(host,None,family=socket.AF_INET)
+        ips=[i[4][0] for i in infos]
+        # prefer 10. over 172.31.
+        for ip in ips:
+            if ip.startswith('10.'): return ip
+        for ip in ips:
+            if ip.startswith('172.31.'): return ip
+        return ips[0] if ips else None
+    except OSError:
+        return None
+gw=ipaddress.ip_network('172.22.0.0/16')
+out=[]
+changed=False
+for item in data:
+    ip=item.get('ip')
+    host=item.get('hostname') or ''
+    try:
+        bad = ip and ipaddress.ip_address(ip) in gw
+    except Exception:
+        bad = False
+    if bad and host:
+        new=resolve(host)
+        if new:
+            item=dict(item)
+            item['ip']=new
+            changed=True
+    out.append(item)
+json.dump(out, sys.stdout, ensure_ascii=False)
+sys.stderr.write('CHANGED' if changed else 'UNCHANGED')
+PY
+
+status=$?
+marker=$(tail -n1 /dev/stderr 2>/dev/null || true)
+if [[ "$status" -ne 0 ]]; then
+  echo "[ERROR] failed to rewrite nodes.json" >&2
+  exit 1
+fi
+
+if grep -q '"ip"\s*:\s*"172\.22\.' "$tmp"; then
+  echo "[WARN] some gwbridge targets remain; manual fix may be required" >&2
+fi
+
+mv "$tmp" "$NODES_JSON"
+echo "[OK] nodes.json updated"
+
+# try to reload Prometheus
+if docker ps --format '{{.Names}}' | grep -q '^argus-prometheus$'; then
+  docker exec argus-prometheus sh -lc 'pidof prometheus >/dev/null 2>&1 && kill -HUP $(pidof prometheus) || supervisorctl restart prometheus' >/dev/null 2>&1 || true
+  echo "[INFO] Prometheus reloaded"
+fi
+
+exit 0
+
--- a/deployment/build/templates/scripts/server-diagnose.sh
+++ b/deployment/build/templates/scripts/server-diagnose.sh
@@ -155,6 +155,34 @@ gfcode=$(http_code "http://localhost:${GRAFANA_PORT:-3000}/api/health"); [[ "$gf
 # Deduplicate errors
 sort -u -o "$ERRORS" "$ERRORS"

+# --- Prometheus targets & nodes.json checks ---
+section PROMETHEUS-TARGETS
+nodes_json_path="$ROOT/private/argus/metric/prometheus/nodes.json"
+if [[ -f "$nodes_json_path" ]]; then
+  logd "nodes.json present: $nodes_json_path"
+  # detect gwbridge addresses (172.22/16)
+  if grep -E '"ip"\s*:\s*"172\.22\.' "$nodes_json_path" >/dev/null 2>&1; then
+    append_err "[prometheus][targets] contains gwbridge (172.22/16) addresses; prefer overlay (10.0/8)."
+    echo "HINT: nodes.json has 172.22.x.x targets. Ensure agent publishes overlay_ip and master prefers it." >&2
+  fi
+else
+  logd "nodes.json missing at $nodes_json_path"
+fi
+
+# Query Prometheus activeTargets and list down items when possible
+pt_json=$(curl -s --max-time 3 "http://localhost:${PROMETHEUS_PORT:-9090}/api/v1/targets" || true)
+if command -v jq >/dev/null 2>&1; then
+  downs=$(printf '%s' "$pt_json" | jq -r '.data.activeTargets[] | select(.health=="down") | "[prometheus][activeTargets] down url=\(.scrapeUrl) lastError=\(.lastError)"' 2>/dev/null || true)
+  if [[ -n "$downs" ]]; then
+    printf '%s\n' "$downs" >> "$ERRORS"
+  fi
+else
+  # best-effort grep when jq is unavailable
+  if printf '%s' "$pt_json" | grep -q '"health":"down"'; then
+    append_err "[prometheus][activeTargets] some targets are down (enable jq for detailed reasons)"
+  fi
+fi
+
 echo "Diagnostic details -> $DETAILS"
 echo "Detected errors   -> $ERRORS"