xiaoxia-saas/scripts/ci/docker_build_push.sh

#!/bin/bash
# 通用Docker镜像构建+推送脚本（local cache为主 + registry cache兜底）
# M-2优化：解决registry缓存导入慢(247s)和推送不稳定问题
# 用法: docker_build_push.sh <Dockerfile> <image_tag> <cache_ref> [build_arg...]
set -eu

DOCKERFILE="$1"
IMAGE_TAG="$2"
CACHE_REF="$3"
shift 3
BUILD_ARGS=""
for arg in "$@"; do
  BUILD_ARGS="$BUILD_ARGS --build-arg $arg"
done

if ! docker buildx inspect ci-builder > /dev/null 2>&1; then
  docker buildx create --use --name ci-builder --driver docker-container
  echo "Created ci-builder"
else
  docker buildx use ci-builder
  echo "Using existing ci-builder"
fi
docker buildx inspect --bootstrap

# 从cache_ref中提取缓存名称（如 api-cache:develop -> api-cache-develop）
CACHE_NAME=$(echo "$CACHE_REF" | tr '/' '_' | tr ':' '-')
LOCAL_CACHE_DIR="/tmp/buildx-cache/${CACHE_NAME}"

mkdir -p "$LOCAL_CACHE_DIR"

# 缓存源：local优先（带自动修复），registry兜底
# 本地缓存损坏时自动清理后重试，避免snapshot not found导致构建全挂
build_with_cache_retry() {
  local attempt=1
  local max_attempts=2
  while [ $attempt -le $max_attempts ]; do
    local build_output
    local exit_code
    set +e
    build_output=$(docker buildx build \
      $BUILD_ARGS \
      --cache-from "type=local,src=${LOCAL_CACHE_DIR}" \
      --cache-from "type=registry,ref=${CACHE_REF},ignore-error=true" \
      --cache-to "type=local,dest=${LOCAL_CACHE_DIR},mode=max" \
      -f "${DOCKERFILE}" \
      -t "${IMAGE_TAG}" \
      --push \
      . 2>&1)
    exit_code=$?
    set -e
    if [ $exit_code -eq 0 ]; then
      echo "$build_output"
      return 0
    fi
    # 检测到缓存损坏类错误，清掉本地缓存重试
    if echo "$build_output" | grep -qE "parent snapshot.*not found|snapshot.*does not exist|cache.*corrupt|failed to compute cache key"; then
      echo "$build_output"
      echo ""
      echo "⚠️  Local cache appears corrupted, cleaning up and retrying (attempt $attempt/$max_attempts)..."
      rm -rf "${LOCAL_CACHE_DIR}"
      mkdir -p "${LOCAL_CACHE_DIR}"
      # 清理buildx builder的内部snapshot状态
      docker buildx prune -f -a >/dev/null 2>&1 || true
      attempt=$((attempt + 1))
    else
      # 非缓存类错误，直接输出并返回
      echo "$build_output"
      return $exit_code
    fi
  done
  # 重试完还是失败，不用本地缓存最后试一次（只从registry读）
  echo "⚠️  All cached attempts failed, building without local cache..."
  docker buildx build \
    $BUILD_ARGS \
    --cache-from "type=registry,ref=${CACHE_REF},ignore-error=true" \
    --cache-to "type=local,dest=${LOCAL_CACHE_DIR},mode=max" \
    -f "${DOCKERFILE}" \
    -t "${IMAGE_TAG}" \
    --push \
    .
}

echo "=== Step 1: Build & push image (local cache + registry read, with auto-repair) ==="
echo "Local cache: ${LOCAL_CACHE_DIR}"
echo "Registry cache: ${CACHE_REF}"
echo ""

build_with_cache_retry

echo ""
echo "Image pushed: ${IMAGE_TAG}"
echo "Local cache updated"

# DISABLED: registry cache too slow echo ""
# DISABLED: registry cache too slow echo "=== Step 2: Sync registry cache (best effort, retries 3x) ==="
# DISABLED: registry cache too slow CACHE_TO_REGISTRY="type=registry,ref=${CACHE_REF},mode=max,compression=zstd"
# DISABLED: registry cache too slow 
# DISABLED: registry cache too slow MAX_RETRIES=3
# DISABLED: registry cache too slow SUCCESS=0
# DISABLED: registry cache too slow for attempt in $(seq 1 $MAX_RETRIES); do
# DISABLED: registry cache too slow   echo "Registry cache sync attempt $attempt/$MAX_RETRIES"
# DISABLED: registry cache too slow   if docker buildx build \
# DISABLED: registry cache too slow     $BUILD_ARGS \
# DISABLED: registry cache too slow     --cache-from "${CACHE_FROM_LOCAL}" \
# DISABLED: registry cache too slow     --cache-to "${CACHE_TO_REGISTRY}" \
# DISABLED: registry cache too slow     -f "${DOCKERFILE}" \
# DISABLED: registry cache too slow     -t "${IMAGE_TAG}" \
# DISABLED: registry cache too slow     --push \
# DISABLED: registry cache too slow     .; then
# DISABLED: registry cache too slow     echo "Registry cache synced (attempt $attempt)"
# DISABLED: registry cache too slow     SUCCESS=1
# DISABLED: registry cache too slow     break
# DISABLED: registry cache too slow   else
# DISABLED: registry cache too slow     echo "Registry cache sync failed (attempt $attempt)"
# DISABLED: registry cache too slow     if [ $attempt -lt $MAX_RETRIES ]; then
# DISABLED: registry cache too slow       WAIT=$((attempt * 5))
# DISABLED: registry cache too slow       echo "Retrying in ${WAIT}s..."
# DISABLED: registry cache too slow       sleep $WAIT
# DISABLED: registry cache too slow     fi
# DISABLED: registry cache too slow   fi
# DISABLED: registry cache too slow done
# DISABLED: registry cache too slow 
# DISABLED: registry cache too slow if [ $SUCCESS -eq 0 ]; then
# DISABLED: registry cache too slow   echo "WARNING: Registry cache sync failed after $MAX_RETRIES attempts (non-fatal, local cache still works)"
# DISABLED: registry cache too slow fi

echo ""
echo "Build completed: ${IMAGE_TAG}"