Set short timeouts for api test to prevent startup delays

Since endpointAvailable is called on construct, this can happen in premain when running with the agent. I included a simple retry just in case…

Not sure in what case the trace agent would respond so slowly…

dd-trace-ot/src/main/java/datadog/trace/common/writer/DDApi.java

@@ -32,7 +32,7 @@ public class DDApi {
 
   private final String tracesEndpoint;
   private final String servicesEndpoint;
-  private final List<ResponseListener> responseListeners = new ArrayList<ResponseListener>();
+  private final List<ResponseListener> responseListeners = new ArrayList<>();
 
   private final RateLimiter loggingRateLimiter =
       RateLimiter.create(1.0 / SECONDS_BETWEEN_ERROR_LOG);
@@ -51,7 +51,7 @@ public class DDApi {
     }
   }
 
-  public void addResponseListener(ResponseListener listener) {
+  public void addResponseListener(final ResponseListener listener) {
     if (!responseListeners.contains(listener)) {
       responseListeners.add(listener);
     }
@@ -137,12 +137,12 @@ public class DDApi {
         if (null != responseString
             && !"".equals(responseString.trim())
             && !"OK".equalsIgnoreCase(responseString.trim())) {
-          JsonNode response = objectMapper.readTree(responseString);
+          final JsonNode response = objectMapper.readTree(responseString);
-          for (ResponseListener listener : responseListeners) {
+          for (final ResponseListener listener : responseListeners) {
             listener.onResponse(endpoint, response);
           }
         }
-      } catch (IOException e) {
+      } catch (final IOException e) {
         log.debug("failed to parse DD agent response: " + responseString, e);
       }
       return true;
@@ -164,13 +164,25 @@ public class DDApi {
   }
 
   private boolean endpointAvailable(final String endpoint) {
+    return endpointAvailable(endpoint, true);
+  }
+
+  private boolean endpointAvailable(final String endpoint, final boolean retry) {
     try {
       final HttpURLConnection httpCon = getHttpURLConnection(endpoint);
-      OutputStreamWriter out = new OutputStreamWriter(httpCon.getOutputStream());
+
+      // This is potentially called in premain, so we want to fail fast.
+      httpCon.setConnectTimeout((int) TimeUnit.SECONDS.toMillis(1));
+      httpCon.setReadTimeout((int) TimeUnit.SECONDS.toMillis(1));
+
+      final OutputStreamWriter out = new OutputStreamWriter(httpCon.getOutputStream());
       out.flush();
       out.close();
       return httpCon.getResponseCode() == 200;
-    } catch (IOException e) {
+    } catch (final IOException e) {
+      if (retry) {
+        return endpointAvailable(endpoint, false);
+      }
     }
     return false;
   }

dd-trace-ot/src/test/groovy/datadog/trace/api/writer/DDApiTest.groovy

@@ -8,9 +8,11 @@ import datadog.trace.common.Service
 import datadog.trace.common.writer.DDApi
 import datadog.trace.common.writer.DDApi.ResponseListener
 import org.msgpack.jackson.dataformat.MessagePackFactory
+import ratpack.exec.Blocking
 import ratpack.http.Headers
 import ratpack.http.MediaType
 import spock.lang.Specification
+import spock.lang.Unroll
 
 import java.util.concurrent.atomic.AtomicReference
 
@@ -253,12 +255,11 @@ class DDApiTest extends Specification {
     // response not seen because of non-200 status
     agentResponse.get() == 'not-set'
 
-
     cleanup:
     agent.close()
   }
 
-    def "Api Downgrades to v3"() {
+  def "Api Downgrades to v3 if v0.4 not available"() {
     setup:
     def v3Agent = ratpack {
       handlers {
@@ -280,6 +281,52 @@ class DDApiTest extends Specification {
     v3Agent.close()
   }
 
+  @Unroll
+  def "Api Downgrades to v3 if timeout exceeded (#delayTrace, #delayServices, #badPort)"() {
+    // This test is unfortunately only exercising the read timeout, not the connect timeout.
+    setup:
+    def agent = ratpack {
+      handlers {
+        put("v0.3/traces") {
+          response.status(200).send()
+        }
+        put("v0.3/services") {
+          response.status(200).send()
+        }
+        put("v0.4/traces") {
+          Blocking.exec {
+            Thread.sleep(delayTrace)
+            response.status(200).send()
+          }
+        }
+        put("v0.4/services") {
+          Blocking.exec {
+            Thread.sleep(delayServices)
+            response.status(200).send()
+          }
+        }
+      }
+    }
+    def port = badPort ? 999 : agent.address.port
+    def client = new DDApi("localhost", port)
+
+    expect:
+    client.tracesEndpoint == "http://localhost:${port}/$endpointVersion/traces"
+    client.servicesEndpoint == "http://localhost:${port}/$endpointVersion/services"
+
+    cleanup:
+    agent.close()
+
+    where:
+    endpointVersion | delayTrace | delayServices | badPort
+    "v0.4"          | 0          | 0             | false
+    "v0.3"          | 0          | 0             | true
+    "v0.4"          | 500        | 0             | false
+    "v0.4"          | 0          | 500           | false
+    "v0.3"          | 3000       | 0             | false
+    "v0.3"          | 0          | 3000          | false
+  }
+
   static List<TreeMap<String, Object>> convertList(byte[] bytes) {
     return mapper.readValue(bytes, new TypeReference<List<TreeMap<String, Object>>>() {})
   }