如何在CDKTF运行时获取可用区并创建可直接引用的子网？

我原本用以下代码为us-west-2区域的每个AWS可用区创建子网：

const dataAwsAvailabilityZonesAll = 
  new aws.datasources.DataAwsAvailabilityZones(this, "all", {});


const zoneNames = Fn.lookup(
  dataAwsAvailabilityZonesAll.fqn,
  "names",
  undefined
);
const availabilityZone = Fn.element(
  zoneNames,
  Token.asNumber("count.index")
);

const publicSubnet = new aws.vpc.Subnet(this, "pub_subnet", {
  availabilityZone,
  cidrBlock: "${cidrsubnet(aws_vpc.vpc.cidr_block, 8, count.index)}",
  mapPublicIpOnLaunch: true,
  tags: {
    name: environment + "-public-subnet-${(count.index + 1)}",
    ["kubernetes.io/cluster/" + eksClusterName]: "shared",
    "kubernetes.io/role/elb": "1",
    environment: environment,
    public: "true",
  },
  vpcId: "${aws_vpc.vpc.id}",
});
publicSubnet.addOverride("count", Fn.lengthOf(zoneNames));

我想重构代码，去掉Fn.count、Fn.lookup、count.index这类用法，直接在TypeScript中引用每个子网。尝试了下面的写法，但没有创建任何子网：

const testSubnets = []
for (let i = 0; i++; i < Fn.lengthOf(zoneNames)) {
  const s = Fn.element(zoneNames, i)
  testSubnets.push(new aws.vpc.Subnet(this, `subnet-${s}`, {
    availabilityZone: s,
    cidrBlock:
      `\${cidrsubnet(aws_vpc.vpc.cidr_block, 4, ${i + 3*dataAwsAvailabilityZonesAll.names.length})}`,
    mapPublicIpOnLaunch: true,
    tags: {
      name: environment + `${environment}-large-public-subnet-${i}`,
      ["kubernetes.io/cluster/" + eksClusterName]: "shared",
      "kubernetes.io/role/elb": "1",
      environment: environment,
      public: "true",
    },
    vpcId: clusterVpc.id
  }))
}

看起来dataAwsAvailabilityZonesAll.names在运行时为空，要等到CDKTF生命周期后期才会解析。请问：

1. 在CDKTF中如何正确获取指定区域的可用区名称？
2. 是否应该放弃自动获取可用区，转而将AZ声明为常量来创建子网？

1. 理解CDKTF的Token机制 #

dataAwsAvailabilityZonesAll.names是一个Token，它在CDKTF的synth阶段（生成Terraform配置的阶段）只是占位符，实际的可用区列表要等到Terraform执行terraform plan/apply时才会被解析。普通TypeScript循环（如for）在synth阶段运行，此时Token没有实际值，循环不会执行，自然不会创建子网。

2. 正确重构代码：用Token感知的迭代 #

使用CDKTF提供的Token.asList()将可用区列表转为Token数组，再用map方法创建每个子网。这种方式会被CDKTF正确转换为Terraform的资源迭代逻辑，同时保留TS层面的直观引用：

const dataAwsAvailabilityZonesAll = new aws.datasources.DataAwsAvailabilityZones(this, "all", {});

// 将可用区列表转为Token数组，支持迭代
const zoneNames = Token.asList(dataAwsAvailabilityZonesAll.names);

// 为每个可用区创建子网，同时在TS中保留子网引用
const publicSubnets = zoneNames.map((zone, index) => {
  // 将索引转为Token，确保在Terraform中正确解析
  const subnetIndex = Token.asNumber(index);

  return new aws.vpc.Subnet(this, `pub-subnet-${zone}`, {
    availabilityZone: zone,
    // 使用Fn.cidrSubnet替代字符串插值，更符合CDKTF规范
    cidrBlock: Fn.cidrSubnet(clusterVpc.cidrBlock, 8, subnetIndex),
    mapPublicIpOnLaunch: true,
    tags: {
      name: `${environment}-public-subnet-${Token.asString(Token.add(subnetIndex, 1))}`,
      [`kubernetes.io/cluster/${eksClusterName}`]: "shared",
      "kubernetes.io/role/elb": "1",
      environment: environment,
      public: "true",
    },
    vpcId: clusterVpc.id,
  });
});

3. 是否需要用常量声明AZ？ #

不需要放弃自动获取可用区。自动获取的优势在于：

• 跨区域部署时无需手动修改AZ列表
• 适配AWS区域的AZ数量变化
• 减少硬编码带来的维护成本

只有当业务需要固定使用特定AZ（比如合规要求、关联已有资源）时，才考虑将AZ声明为常量，例如：

const fixedAZs = ["us-west-2a", "us-west-2b", "us-west-2c"];
const publicSubnets = fixedAZs.map((zone, index) => {
  return new aws.vpc.Subnet(this, `pub-subnet-${zone}`, {
    // ... 其他配置
  });
});

内容的提问来源于stack exchange，提问作者Paymahn Moghadasian