Before you start

Make sure that you can compile and run core tests and read a basic example Writing your first HTTP server.

Step by step guide

Consider the case when you need to write an HTTP handler that greets a user. However, the localized strings are stored at some remote HTTP server, and those localizations rarely change. To reduce the network traffic and improve response times it is profitable to bulk retrieve the translations from remote and cache them in a local cache.

In this sample we show how to write and test a cache along with creating HTTP requests. If you wish to cache data from database prefer using cache specializations for DB.

Remote HTTP server description

For the purpose of this there is some HTTP server that has HTTP handler http://localhost:8090/v1/translations.

Handler returns all the available translations as JSON on GET:

curl http://localhost:8090/v1/translations -s | jq
{
  "content": {
    "hello": {
      "ru": "Привет",
      "en": "Hello"
    },
    "wellcome": {
      "ru": "Добро пожаловать",
      "en": "Wellcome"
    }
  },
  "update_time": "2021-11-01T12:00:00Z"
}

In case of query parameter "last_update" the handler returns only the changed translations since requested time:

$ curl http://localhost:8090/v1/translations?last_update=2021-11-01T12:00:00Z -s | jq
{
  "content": {
    "hello": {
      "ru": "Приветище"
    }
  },
  "update_time": "2021-12-01T12:00:00Z"
}

We are planning to cache those translations in a std::unordered_map:

namespace samples::http_cache {
 
struct KeyLang {
  std::string key;
  std::string language;
};
 
struct KeyLangEq {
  bool operator()(const KeyLang& x, const KeyLang& y) const noexcept;
};
 
struct KeyLangHash {
  bool operator()(const KeyLang& x) const noexcept;
};
 
using KeyLangToTranslation =
    std::unordered_map<KeyLang, std::string, KeyLangHash, KeyLangEq>;
 
}  // namespace samples::http_cache

Cache component

Our cache component should have the following fields:

Reference to HTTP client, that is required to make HTTP requests
URL to the incremental and full update handle
String with last update time from remote. This string used only from the components::CachingComponentBase::Update overload and it is guaranteed that Update is not called concurrently, so there is no need to protect it from concurrent access.

namespace samples::http_cache {
 
class HttpCachedTranslations final
    : public components::CachingComponentBase<KeyLangToTranslation> {
 public:
  static constexpr std::string_view kName = "cache-http-translations";
 
  HttpCachedTranslations(const components::ComponentConfig& config,
                         const components::ComponentContext& context);
  ~HttpCachedTranslations() override;
 
  void Update(
      cache::UpdateType type,
      [[maybe_unused]] const std::chrono::system_clock::time_point& last_update,
      [[maybe_unused]] const std::chrono::system_clock::time_point& now,
      cache::UpdateStatisticsScope& stats_scope) override;
 
  static yaml_config::Schema GetStaticConfigSchema();
 
 private:
  clients::http::Client& http_client_;
  const std::string translations_url_;
  std::string last_update_remote_;
 
  formats::json::Value GetAllData() const;
  formats::json::Value GetUpdatedData() const;
 
  void MergeAndSetData(KeyLangToTranslation&& content,
                       formats::json::Value json,
                       cache::UpdateStatisticsScope& stats_scope);
};
 
}  // namespace samples::http_cache

To create a non LRU cache cache you have to derive from components::CachingComponentBase, call CacheUpdateTrait::StartPeriodicUpdates() at the component constructor and CacheUpdateTrait::StopPeriodicUpdates() at the destructor:

HttpCachedTranslations::HttpCachedTranslations(
    const components::ComponentConfig& config,
    const components::ComponentContext& context)
    : CachingComponentBase(config, context),
      http_client_(
          context.FindComponent<components::HttpClient>().GetHttpClient()),
      translations_url_(config["translations-url"].As<std::string>()) {
  CacheUpdateTrait::StartPeriodicUpdates();
}
 
HttpCachedTranslations::~HttpCachedTranslations() {
  CacheUpdateTrait::StopPeriodicUpdates();
}

The constructor initializes component fields with data from static configuration and with references to clients.

Update

Depending on cache configuration settings the overloaded Update function is periodically called with different options:

void HttpCachedTranslations::Update(
    cache::UpdateType type,
    [[maybe_unused]] const std::chrono::system_clock::time_point& last_update,
    [[maybe_unused]] const std::chrono::system_clock::time_point& now,
    cache::UpdateStatisticsScope& stats_scope) {
  formats::json::Value json;
  switch (type) {
    case cache::UpdateType::kFull:
      json = GetAllData();
      break;
    case cache::UpdateType::kIncremental:
      json = GetUpdatedData();
      break;
    default:
      UASSERT(false);
  }
 
  if (json.IsEmpty()) {
    stats_scope.FinishNoChanges();
    return;
  }
 
  KeyLangToTranslation content;
  if (type == cache::UpdateType::kIncremental) {
    const auto snapshot = Get();  // smart pointer to a shared cache data
    content = *snapshot;          // copying the shared data
  }
 
  MergeAndSetData(std::move(content), json, stats_scope);
}

In this sample full and incremental data retrieval is implemented in GetAllData() and GetUpdatedData() respectively.

Both functions return data in the same format and we parse both responses using MergeAndSetData:

void HttpCachedTranslations::MergeAndSetData(
    KeyLangToTranslation&& content, formats::json::Value json,
    cache::UpdateStatisticsScope& stats_scope) {
  for (const auto& [key, value] : Items(json["content"])) {
    for (const auto& [lang, text] : Items(value)) {
      content.insert_or_assign(KeyLang{key, lang}, text.As<std::string>());
    }
    stats_scope.IncreaseDocumentsReadCount(value.GetSize());
  }
 
  auto update_time = json["update_time"].As<std::string>();
 
  const auto size = content.size();
  Set(std::move(content));
  last_update_remote_ = std::move(update_time);
  stats_scope.Finish(size);
}

At the end of the MergeAndSetData function the components::CachingComponentBase::Set() invocation stores data as a new cache.

Update time from remote stored into last_update_remote_. Clocks on different hosts are usually out of sync, so it is better to store the remote time if possible, rather than using a local times from last_update and now input parameters.

Data retrieval

To make an HTTP request call clients::http::Client::CreateRequest() to get an instance of clients::http::Request builder. Work with builder is quite straightforward:

formats::json::Value HttpCachedTranslations::GetAllData() const {
  auto response = http_client_.CreateRequest()
                      .get(translations_url_)  // HTTP GET translations_url_ URL
                      .retry(2)                // retry once in case of error
                      .timeout(std::chrono::milliseconds{500})
                      .perform();  // start performing the request
  response->raise_for_status();
  return formats::json::FromString(response->body_view());
}

HTTP requests for incremental update differ only in URL and query parameter last_update:

formats::json::Value HttpCachedTranslations::GetUpdatedData() const {
  const auto url =
      http::MakeUrl(translations_url_, {{"last_update", last_update_remote_}});
  auto response = http_client_.CreateRequest()
                      .get(url)
                      .retry(2)
                      .timeout(std::chrono::milliseconds{500})
                      .perform();
  response->raise_for_status();
  return formats::json::FromString(response->body_view());
}

Static configuration

To configure the new cache component provide its own options, options from components::CachingComponentBase:

components_manager:
    components:                       # Configuring components that were registered via component_list
        cache-http-translations:
            translations-url: 'mockserver/v1/translations'  # Some other microservice listens on this URL
 
            update-types: full-and-incremental
            full-update-interval: 1h
            update-interval: 15m

Options for dependent components components::HttpClient, components::TestsuiteSupport and support handler server::handlers::TestsControl should be provided:

        http-client:
            fs-task-processor: fs-task-processor
 
        dns-client:                     # Asynchronous DNS component
            fs-task-processor: fs-task-processor
 
        testsuite-support:
        tests-control:
            # Some options from server::handlers::HttpHandlerBase
            path: /tests/{action}
            method: POST
            task_processor: main-task-processor
 
        server:
            # ...

Dynamic configuration

Dynamic configuration is close to the basic configuration from Writing your first HTTP server but should have additional options for HTTP client: samples/http_caching/dynamic_config_fallback.json

All the values are described in at Dynamic configs.

A production ready service would dynamically retrieve the above options at runtime from a configuration service. See Writing your own configs server for insights on how to change the above options on the fly, without restarting the service.

Cache component usage

Now the cache could be used just as any other component. For example, a handler could get a reference to the cache and use it in HandleRequestThrow:

class GreetUser final : public server::handlers::HttpHandlerBase {
 public:
  static constexpr std::string_view kName = "handler-greet-user";
 
  GreetUser(const components::ComponentConfig& config,
            const components::ComponentContext& context)
      : HttpHandlerBase(config, context),
        cache_(context.FindComponent<HttpCachedTranslations>()) {}
 
  std::string HandleRequestThrow(
      const server::http::HttpRequest& request,
      server::request::RequestContext&) const override {
    const auto cache_snapshot = cache_.Get();
 
    using samples::http_cache::KeyLang;
    const auto& hello = cache_snapshot->at(KeyLang{"hello", "ru"});
    const auto& welcome = cache_snapshot->at(KeyLang{"welcome", "ru"});
    return fmt::format("{}, {}! {}", hello, request.GetArg("username"),
                       welcome);
  }
 
 private:
  samples::http_cache::HttpCachedTranslations& cache_;
};

Note that the cache is concurrency safe as all the components.

int main()

Finally, after writing down the dynamic configuration values into file at dynamic-config-fallbacks.fallback-path, we add our component to the components::MinimalServerComponentList(), and start the server with static configuration kStaticConfig.

int main(int argc, char* argv[]) {
  const auto component_list =
      components::MinimalServerComponentList()
          .Append<samples::http_cache::HttpCachedTranslations>()
          .Append<samples::http_cache::GreetUser>()
          .Append<components::TestsuiteSupport>()
          .Append<server::handlers::TestsControl>()
          .Append<clients::dns::Component>()
          .Append<components::HttpClient>();
  return utils::DaemonMain(argc, argv, component_list);
}

Build and Run

To build the sample, execute the following build steps at the userver root directory:

mkdir build_release
cd build_release
cmake -DCMAKE_BUILD_TYPE=Release ..
make userver-samples-http_caching

Note that you need a running translations service with bulk handlers to start the service. You could use the mongo service for that purpose.

The sample could be started by running make start-userver-samples-http_caching. The command would invoke testsuite start target that sets proper paths in the configuration files and starts the service.

To start the service manually run ./samples/http_caching/userver-samples-http_caching -c </path/to/static_config.yaml> (do not forget to prepare the configuration files!).

Now you can send a request to your server from another terminal:

$ curl -X POST http://localhost:8089/samples/greet?username=Dear+Developer  -i
HTTP/1.1 200 OK
Date: Thu, 09 Dec 2021 17:01:44 UTC
Content-Type: text/html; charset=utf-8
X-YaRequestId: 94193f99ebf94eb58252139f2e9dace4
X-YaSpanId: 4e17e02dfa7b8322
X-YaTraceId: 306d2d54fd0543c09376a5c4bb120a88
Server: userver/1.0.0 (20211209085954; rv:d05d059a3)
Connection: keep-alive
Content-Length: 61
 
Привет, Dear Developer! Добро пожаловать

Auto testing

The server::handlers::TestsControl and components::TestsuiteSupport components among other things provide necessary control over the caches. You can stop/start periodic updates or even force updates of caches by HTTP requests to the server::handlers::TestsControl handler.

For example the following JSON forces incremental update of the cache-http-translations cache:

{"invalidate_caches": {
    "update_type": "incremental",
    "names": ["cache-http-translations"]
}}

Fortunately, the testsuite API provides all the required functionality via simpler to use Python functions.

Functional testing

Functional tests for the service could be implemented using the testsuite. To do that you have to:

Mock the translations service data:
@pytest.fixture

def translations():

return {

'hello': {'en': 'hello', 'ru': 'Привет'},

'welcome': {'ru': 'Добро пожаловать', 'en': 'Welcome'},

}
Mock the translations service API:
@pytest.fixture(autouse=True)

def mock_translations(mockserver, translations, mocked_time):

@mockserver.json_handler('/v1/translations')

def mock(request):

return {

'content': translations,

'update_time': utils.timestring(mocked_time.now()),

}

return mock
Import the pytest_userver.plugins.core plugin and teach testsuite how to patch the service config to use the mocked URL:
pytest_plugins = ['pytest_userver.plugins.core']

USERVER_CONFIG_HOOKS = ['userver_config_translations']

@pytest.fixture(scope='session')

def userver_config_translations(mockserver_info):

def do_patch(config_yaml, config_vars):

components = config_yaml['components_manager']['components']

components['cache-http-translations'][

'translations-url'

] = mockserver_info.url('v1/translations')

return do_patch
Write the test:
async def test_http_caching(service_client, translations, mocked_time):

response = await service_client.post(

'/samples/greet', params={'username': 'дорогой разработчик'},

)

assert response.status == 200

assert response.text == 'Привет, дорогой разработчик! Добро пожаловать'

translations['hello']['ru'] = 'Приветище'

mocked_time.sleep(10)

await service_client.invalidate_caches()

response = await service_client.post(

'/samples/greet', params={'username': 'дорогой разработчик'},

)

assert response.status == 200

assert response.text == 'Приветище, дорогой разработчик! Добро пожаловать'

Full sources

See the full example:

Go backTCP full-duplex server with metrics and Spans Up nextHTTP Flatbuf handler and requests

Table of Contents