Fixes AI.MODELGET to return INPUTS,OUTPUTS, BATCHSIZE, and MINBATCHSIZE

docs/commands.md

@@ -216,6 +216,10 @@ An array of alternating key-value pairs as follows:
 1. **BACKEND**: the backend used by the model as a String
 1. **DEVICE**: the device used to execute the model as a String
 1. **TAG**: the model's tag as a String
+1. **BATCHSIZE**: The maximum size of any batch of incoming requests. If `BATCHSIZE` is equal to 0 each incoming request is served immediately. When `BATCHSIZE` is greater than 0, the engine will batch incoming requests from multiple clients that use the model with input tensors of the same shape.
+1. **MINBATCHSIZE**: The minimum size of any batch of incoming requests.
+1. **INPUTS**: array reply with one or more names of the model's input nodes (applicable only for TensorFlow models)
+1. **OUTPUTS**: array reply with one or more names of the model's output nodes (applicable only for TensorFlow models)
 1. **BLOB**: a blob containing the serialized model (when called with the `BLOB` argument) as a String
 
 **Examples**
@@ -224,12 +228,21 @@ Assuming that your model is stored under the 'mymodel' key, you can obtain its m
 
 ```
 redis> AI.MODELGET mymodel META
-1) "backend"
-2) TF
-3) "device"
-4) CPU
-5) "tag"
-6) imagenet:5.0
+ 1) "backend"
+ 2) "TF"
+ 3) "device"
+ 4) "CPU"
+ 5) "tag"
+ 6) "imagenet:5.0"
+ 7) "batchsize"
+ 8) (integer) 0
+ 9) "minbatchsize"
+10) (integer) 0
+11) "inputs"
+12) 1) "a"
+    2) "b"
+13) "outputs"
+14) 1) "c"
 ```
 
 You can also save it to the local file 'model.ext' with [`redis-cli`](https://redis.io/topics/cli) like so:

src/redisai.c

@@ -368,7 +368,7 @@ int RedisAI_ModelGet_RedisCommand(RedisModuleCtx *ctx, RedisModuleString **argv,
 
   RAI_Model *mto;
   RedisModuleKey *key;
-  const int status = RAI_GetModelFromKeyspace( ctx, argv[1], &key, &mto, REDISMODULE_READ | REDISMODULE_WRITE);
+  const int status = RAI_GetModelFromKeyspace( ctx, argv[1], &key, &mto, REDISMODULE_READ );
   if (status == REDISMODULE_ERR) {
     return REDISMODULE_ERR;
   }
@@ -418,12 +418,12 @@ int RedisAI_ModelGet_RedisCommand(RedisModuleCtx *ctx, RedisModuleString **argv,
     return REDISMODULE_OK;
   }
 
-  int outentries = blob ? 8 : 6;
+  const int outentries = blob ? 16 : 14;
 
   RedisModule_ReplyWithArray(ctx, outentries);
 
   RedisModule_ReplyWithCString(ctx, "backend");
-  const char* backendstr = RAI_BackendName(mto->backend);
+  const char *backendstr = RAI_BackendName(mto->backend);
   RedisModule_ReplyWithCString(ctx, backendstr);
 
   RedisModule_ReplyWithCString(ctx, "device");
@@ -432,6 +432,28 @@ int RedisAI_ModelGet_RedisCommand(RedisModuleCtx *ctx, RedisModuleString **argv,
   RedisModule_ReplyWithCString(ctx, "tag");
   RedisModule_ReplyWithCString(ctx, mto->tag ? mto->tag : "");
 
+  RedisModule_ReplyWithCString(ctx, "batchsize");
+  RedisModule_ReplyWithLongLong(ctx, (long)mto->opts.batchsize);
+
+  RedisModule_ReplyWithCString(ctx, "minbatchsize");
+  RedisModule_ReplyWithLongLong(ctx, (long)mto->opts.minbatchsize);
+
+  RedisModule_ReplyWithCString(ctx, "inputs");
+  const size_t ninputs = array_len(mto->inputs);
+  RedisModule_ReplyWithArray(ctx, (long)ninputs);
+
+  for (size_t i = 0; i < ninputs; i++) {
+    RedisModule_ReplyWithCString(ctx, mto->inputs[i]);
+  }
+
+  RedisModule_ReplyWithCString(ctx, "outputs");
+  const size_t noutputs = array_len(mto->outputs);
+  RedisModule_ReplyWithArray(ctx, (long)noutputs);
+
+  for (size_t i = 0; i < noutputs; i++) {
+    RedisModule_ReplyWithCString(ctx, mto->outputs[i]);
+  }
+
   if (meta && blob) {
     RedisModule_ReplyWithCString(ctx, "blob");
     RedisModule_ReplyWithStringBuffer(ctx, buffer, len);

test/tests_onnx.py

@@ -36,21 +36,27 @@ def test_onnx_modelrun_mnist(env):
     ensureSlaveSynced(con, env)
 
     ret = con.execute_command('AI.MODELGET', 'm', 'META')
-    env.assertEqual(len(ret), 6)
-    env.assertEqual(ret[-1], b'')
+    env.assertEqual(len(ret), 14)
+    env.assertEqual(ret[5], b'')
+    # assert there are no inputs or outputs
+    env.assertEqual(len(ret[11]), 0)
+    env.assertEqual(len(ret[13]), 0)
 
-    ret = con.execute_command('AI.MODELSET', 'm', 'ONNX', DEVICE, 'TAG', 'asdf', 'BLOB', model_pb)
+    ret = con.execute_command('AI.MODELSET', 'm', 'ONNX', DEVICE, 'TAG', 'version:2', 'BLOB', model_pb)
     env.assertEqual(ret, b'OK')
 
     ensureSlaveSynced(con, env)
 
     ret = con.execute_command('AI.MODELGET', 'm', 'META')
-    env.assertEqual(len(ret), 6)
-    env.assertEqual(ret[-1], b'asdf')
-
-    # TODO: enable me
-    # env.assertEqual(ret[0], b'ONNX')
-    # env.assertEqual(ret[1], b'CPU')
+    env.assertEqual(len(ret), 14)
+    # TODO: enable me. CI is having issues on GPU asserts of ONNX and CPU
+    if DEVICE == "CPU":
+        env.assertEqual(ret[1], b'ONNX')
+        env.assertEqual(ret[3], b'CPU')
+    env.assertEqual(ret[5], b'version:2')
+    # assert there are no inputs or outputs
+    env.assertEqual(len(ret[11]), 0)
+    env.assertEqual(len(ret[13]), 0)
 
     try:
         con.execute_command('AI.MODELSET', 'm', 'ONNX', DEVICE, 'BLOB', wrong_model_pb)
@@ -166,6 +172,19 @@ def test_onnx_modelrun_mnist_autobatch(env):
                               'BATCHSIZE', 2, 'MINBATCHSIZE', 2, 'BLOB', model_pb)
     env.assertEqual(ret, b'OK')
 
+    ret = con.execute_command('AI.MODELGET', 'm', 'META')
+    env.assertEqual(len(ret), 14)
+    # TODO: enable me. CI is having issues on GPU asserts of ONNX and CPU
+    if DEVICE == "CPU":
+        env.assertEqual(ret[1], b'ONNX')
+        env.assertEqual(ret[3], b'CPU')
+    env.assertEqual(ret[5], b'')
+    env.assertEqual(ret[7], 2)
+    env.assertEqual(ret[9], 2)
+    # assert there are no inputs or outputs
+    env.assertEqual(len(ret[11]), 0)
+    env.assertEqual(len(ret[13]), 0)
+
     con.execute_command('AI.TENSORSET', 'a', 'FLOAT', 1, 1, 28, 28, 'BLOB', sample_raw)
     con.execute_command('AI.TENSORSET', 'c', 'FLOAT', 1, 1, 28, 28, 'BLOB', sample_raw)
 

test/tests_pytorch.py

@@ -62,22 +62,31 @@ def test_pytorch_modelrun(env):
     ensureSlaveSynced(con, env)
 
     ret = con.execute_command('AI.MODELGET', 'm', 'META')
-    env.assertEqual(len(ret), 6)
-    env.assertEqual(ret[-1], b'')
-
-    ret = con.execute_command('AI.MODELSET', 'm', 'TORCH', DEVICE, 'TAG', 'asdf', 'BLOB', model_pb)
+    ret = con.execute_command('AI.MODELGET', 'm', 'META')
+    env.assertEqual(len(ret), 14)
+    # TODO: enable me. CI is having issues on GPU asserts of TORCH and CPU
+    if DEVICE == "CPU":
+        env.assertEqual(ret[1], b'TORCH')
+        env.assertEqual(ret[3], b'CPU')
+    env.assertEqual(ret[5], b'')
+    env.assertEqual(ret[7], 0)
+    env.assertEqual(ret[9], 0)
+    # assert there are no inputs or outputs
+    env.assertEqual(len(ret[11]), 0)
+    env.assertEqual(len(ret[13]), 0)
+
+    ret = con.execute_command('AI.MODELSET', 'm', 'TORCH', DEVICE, 'TAG', 'my:tag:v3', 'BLOB', model_pb)
     env.assertEqual(ret, b'OK')
 
     ensureSlaveSynced(con, env)
 
     ret = con.execute_command('AI.MODELGET', 'm', 'META')
-    env.assertEqual(len(ret), 6)
-    env.assertEqual(ret[-1], b'asdf')
-
-
-    # TODO: enable me
-    # env.assertEqual(ret[0], b'TORCH')
-    # env.assertEqual(ret[1], b'CPU')
+    env.assertEqual(len(ret), 14)
+    env.assertEqual(ret[5], b'my:tag:v3')
+    # TODO: enable me. CI is having issues on GPU asserts of TORCH and CPU
+    if DEVICE == "CPU":
+        env.assertEqual(ret[1], b'TORCH')
+        env.assertEqual(ret[3], b'CPU')
 
     try:
         con.execute_command('AI.MODELSET', 'm', 'TORCH', DEVICE, 'BLOB', wrong_model_pb)

test/tests_tensorflow.py

@@ -187,23 +187,28 @@ def test_run_tf_model(env):
     ensureSlaveSynced(con, env)
 
     ret = con.execute_command('AI.MODELGET', 'm', 'META')
-    env.assertEqual(len(ret), 6)
-    env.assertEqual(ret[-1], b'')
+    env.assertEqual(len(ret), 14)
+    env.assertEqual(ret[5], b'')
+    env.assertEqual(ret[11][0], b'a')
+    env.assertEqual(ret[11][1], b'b')
+    env.assertEqual(ret[13][0], b'mul')
 
-    ret = con.execute_command('AI.MODELSET', 'm', 'TF', DEVICE, 'TAG', 'asdf',
+    ret = con.execute_command('AI.MODELSET', 'm', 'TF', DEVICE, 'TAG', 'version:1',
                               'INPUTS', 'a', 'b', 'OUTPUTS', 'mul', 'BLOB', model_pb)
     env.assertEqual(ret, b'OK')
 
     ensureSlaveSynced(con, env)
 
     ret = con.execute_command('AI.MODELGET', 'm', 'META')
-    env.assertEqual(len(ret), 6)
-    env.assertEqual(ret[-1], b'asdf')
-
-
-    # TODO: enable me
-    # env.assertEqual(ret[0], b'TF')
-    # env.assertEqual(ret[1], b'CPU')
+    env.assertEqual(len(ret), 14)
+    # TODO: enable me. CI is having issues on GPU asserts of TF and CPU
+    if DEVICE == "CPU":
+        env.assertEqual(ret[1], b'TF')
+        env.assertEqual(ret[3], b'CPU')
+    env.assertEqual(ret[5], b'version:1')
+    env.assertEqual(ret[11][0], b'a')
+    env.assertEqual(ret[11][1], b'b')
+    env.assertEqual(ret[13][0], b'mul')
 
     con.execute_command('AI.TENSORSET', 'a', 'FLOAT',
                         2, 2, 'VALUES', 2, 3, 2, 3)
@@ -258,8 +263,10 @@ def test_run_tf2_model(env):
     ensureSlaveSynced(con, env)
 
     ret = con.execute_command('AI.MODELGET', 'm', 'META')
-    env.assertEqual(len(ret), 6)
-    env.assertEqual(ret[-1], b'')
+    env.assertEqual(len(ret), 14)
+    env.assertEqual(ret[5], b'')
+    env.assertEqual(ret[11][0], b'x')
+    env.assertEqual(ret[13][0], b'Identity')
 
     ret = con.execute_command('AI.MODELSET', 'm', 'TF', DEVICE, 'TAG', 'asdf',
                               'INPUTS', 'x', 'OUTPUTS', 'Identity', 'BLOB', model_pb)
@@ -268,8 +275,10 @@ def test_run_tf2_model(env):
     ensureSlaveSynced(con, env)
 
     ret = con.execute_command('AI.MODELGET', 'm', 'META')
-    env.assertEqual(len(ret), 6)
-    env.assertEqual(ret[-1], b'asdf')
+    env.assertEqual(len(ret), 14)
+    env.assertEqual(ret[5], b'asdf')
+    env.assertEqual(ret[11][0], b'x')
+    env.assertEqual(ret[13][0], b'Identity')
 
     zero_values = [0] * (28 * 28)
 

test/tests_tflite.py

@@ -35,26 +35,27 @@ def test_run_tflite_model(env):
     env.assertEqual(ret, b'OK')
 
     ret = con.execute_command('AI.MODELGET', 'm', 'META')
-    env.assertEqual(len(ret), 6)
-    env.assertEqual(ret[-1], b'')
+    env.assertEqual(len(ret), 14)
+    env.assertEqual(ret[5], b'')
 
     ret = con.execute_command('AI.MODELSET', 'm', 'TFLITE', 'CPU', 'TAG', 'asdf', 'BLOB', model_pb)
     env.assertEqual(ret, b'OK')
 
     ret = con.execute_command('AI.MODELGET', 'm', 'META')
-    env.assertEqual(len(ret), 6)
-    env.assertEqual(ret[-1], b'asdf')
+    env.assertEqual(len(ret), 14)
+    env.assertEqual(ret[5], b'asdf')
 
     ret = con.execute_command('AI.TENSORSET', 'a', 'FLOAT', 1, 1, 28, 28, 'BLOB', sample_raw)
     env.assertEqual(ret, b'OK')
 
     ensureSlaveSynced(con, env)
 
     ret = con.execute_command('AI.MODELGET', 'm', 'META')
-    env.assertEqual(len(ret), 6)
-    # TODO: enable me
-    # env.assertEqual(ret[0], b'TFLITE')
-    # env.assertEqual(ret[1], b'CPU')
+    env.assertEqual(len(ret), 14)
+    # TODO: enable me. CI is having issues on GPU asserts of TFLITE and CPU
+    if DEVICE == "CPU":
+        env.assertEqual(ret[1], b'TFLITE')
+        env.assertEqual(ret[3], b'CPU')
 
     con.execute_command('AI.MODELRUN', 'm', 'INPUTS', 'a', 'OUTPUTS', 'b', 'c')