Restore old Evaluator switch, and put the new Evaluator behind a flag

[stephenamar-db]

Before:

Result "com.databricks.jsonnet.bundle.test.InternalJsonnetBuilderBenchmark.main":
  239.623 ±(99.9%) 2.487 ms/op [Average]
  (min, avg, max) = (238.165, 239.623, 242.976), stdev = 1.645
  CI (99.9%): [237.136, 242.110] (assumes normal distribution)

After:

Result "com.databricks.jsonnet.bundle.test.InternalJsonnetBuilderBenchmark.main":
  224.443 ±(99.9%) 1.335 ms/op [Average]
  (min, avg, max) = (223.504, 224.443, 226.269), stdev = 0.883
  CI (99.9%): [223.108, 225.779] (assumes normal distribution)

Looks like there's still a small regression somewhere - will continue investigating
Initial 3.3.5 commit

Result "com.databricks.jsonnet.bundle.test.InternalJsonnetBuilderBenchmark.main":
  216.718 ±(99.9%) 2.125 ms/op [Average]
  (min, avg, max) = (214.841, 216.718, 218.835), stdev = 1.406
  CI (99.9%): [214.593, 218.843] (assumes normal distribution)

[stephenamar-db]

@He-Pin

[He-Pin]

@stephenamar-db seems #329 affects performance too.

I think I need to reshape it to avoid allocation.

[He-Pin]

// 定义一个微服务配置生成器
local generateServiceConfig(serviceParams) =
  local defaults = {
    port: 8080,
    replicas: 1,
    memoryLimit: '512Mi',
    cpuLimit: '0.5',
    environmentVars: {},
    dependencies: [],
  };

  local params = defaults + serviceParams;
  local params2 = {
                      name: params.name,
                      version: params.version,
                      replicas: params.replicas,
                      memory: params.memoryLimit,
                      cpu: params.cpuLimit,
                      port: params.port,
                      env: std.manifestJsonEx(params.environmentVars, '  '),
                      deps: std.join(',    ', ['"%s"' % dep for dep in params.dependencies]),
                    };

  |||
    // Service configuration for %(name)s
    {
      name: "%(name)s",
      version: "%(version)s",

      deployment: {
        replicas: %(replicas)d,
        resources: {
          limits: {
            memory: "%(memory)s",
            cpu: "%(cpu)s",
          },
        },
      },

      networking: {
        port: %(port)d,
        service: "%(name)s-service",
      },

      environment: %(env)s,
      unset: input.unset,

      dependencies: [
        %(deps)s
      ],
    }
  ||| % params2;

// 生成多个服务的配置
local services = {
  'user-service': {
    name: 'user-service',
    version: '1.2.0',
    port: 8081,
    replicas: 3,
    environmentVars: {
      DB_HOST: 'postgres.example.com',
      LOG_LEVEL: 'info',
    },
    dependencies: ['auth-service', 'email-service'],
  },
  'auth-service': {
    name: 'auth-service',
    version: '1.1.0',
    port: 8082,
    memoryLimit: '1Gi',
    environmentVars: {
      JWT_SECRET: 'secret-placeholder',
      TOKEN_EXPIRY: '24h',
    },
  },
};

// 生成每个服务的配置文件内容
{
  [svcName + '.jsonnet']: generateServiceConfig(services[svcName])
  for svcName in std.objectFields(services)
}

we are testing something like this:

render it 10_000_000 times;
cost time: 61926.0 ms
cost time per time: 0.0061926 ms

[He-Pin]

We can collect statistics about Expr and sort them by frequency?