simple_sequence_client must use stream with async (triton-inference-s…

…erver#580)
Marshall2M · Aug 23, 2019 · fd5eca9 · fd5eca9
1 parent 072566f
commit fd5eca9
Show file tree

Hide file tree

Showing 2 changed files with 21 additions and 9 deletions.
diff --git a/src/clients/c++/simple_sequence_client.cc b/src/clients/c++/simple_sequence_client.cc
@@ -236,14 +236,22 @@ main(int argc, char** argv)
   const ni::CorrelationID correlation_id0 = 1;
   const ni::CorrelationID correlation_id1 = 2;
 
-  // Create two different contexts, one is using streaming while the other
-  // isn't. Then we can compare their difference in sync/async runs
+  // Create two different contexts, in the sync case we can use one
+  // streaming and one not streaming. In the async case must use
+  // streaming for both since async+non-streaming means that order of
+  // requests reaching inference server is not guaranteed.
   err = nic::InferGrpcStreamContext::Create(
       &ctx0, correlation_id0, url, model_name, -1 /* model_version */, verbose);
   if (err.IsOk()) {
-    err = nic::InferGrpcContext::Create(
-        &ctx1, correlation_id1, url, model_name, -1 /* model_version */,
-        verbose);
+    if (async) {
+      err = nic::InferGrpcStreamContext::Create(
+          &ctx1, correlation_id1, url, model_name, -1 /* model_version */,
+          verbose);
+    } else {
+      err = nic::InferGrpcContext::Create(
+          &ctx1, correlation_id1, url, model_name, -1 /* model_version */,
+          verbose);
+    }
   }
 
   if (!err.IsOk()) {

diff --git a/src/clients/python/simple_sequence_client.py b/src/clients/python/simple_sequence_client.py
@@ -95,15 +95,19 @@ def async_receive(ctx, request_id):
     # correlation ID.
     values = [11, 7, 5, 3, 2, 0, 1]
 
-    # For the two different contexts, one is using streaming while the other
-    # isn't. Then we can compare their difference in sync/async runs
+    # Create two different contexts, in the sync case we can use one
+    # streaming and one not streaming. In the async case must use
+    # streaming for both since async+non-streaming means that order of
+    # requests reaching inference server is not guaranteed.
     correlation_id0 = 1000
     ctx0 = InferContext(FLAGS.url, protocol, model_name, model_version,
-                        correlation_id=correlation_id0, verbose=FLAGS.verbose, streaming=True)
+                        correlation_id=correlation_id0, verbose=FLAGS.verbose,
+                        streaming=True)
 
     correlation_id1 = 1001
     ctx1 = InferContext(FLAGS.url, protocol, model_name, model_version,
-                        correlation_id=correlation_id1, verbose=FLAGS.verbose, streaming=False)
+                        correlation_id=correlation_id1, verbose=FLAGS.verbose,
+                        streaming=FLAGS.async_set)
 
     # Now send the inference sequences..
     ctxs = []