AI.INFO command

docs/commands.md

@@ -2,8 +2,6 @@
 
 ## AI.CONFIG LOADBACKEND
 
-### AI.CONFIG LOADBACKEND Example
-
 Load a DL/ML backend.
 
 By default, RedisAI starts with the ability to set and get tensor data, but setting and running models and scritps requires a computing backend to be loaded. This command allows to dynamically load a backend by specifying the backend identifier and the path to the backend library. Currently, once loaded, a backend cannot be unloaded, and there can be at most one backend per identifier loaded.
@@ -16,6 +14,8 @@ AI.CONFIG LOADBACKEND <backend_identifier> <location_of_backend_library>
 
 It is possible to specify backends at the command-line when starting `redis-server`, see example below.
 
+### AI.CONFIG LOADBACKEND Example
+
 > Load the TORCH backend
 
 ```sql
@@ -234,3 +234,80 @@ If needed, input tensors are copied to the device specified in `AI.SCRIPTSET` be
 ```sql
 AI.SCRIPTRUN addscript addtwo INPUTS a b OUTPUTS c
 ```
+
+## AI.INFO
+
+Return information about runs of a `MODEL` or a `SCRIPT`.
+
+At each `MODELRUN` or `SCRIPTRUN`, RedisAI will collect statistcs specific for each `MODEL` or `SCRIPT`,
+specific for the node (hence nodes in a cluster will have to be queried individually for their info).
+The following information is collected:
+
+- `KEY`: the key being run
+- `TYPE`: either `MODEL` or `SCRIPT`
+- `BACKEND`: the type of backend (always `TORCH` for `SCRIPT`)
+- `DEVICE`: the device where the run has been executed
+- `DURATION`: cumulative duration in microseconds
+- `SAMPLES`: cumulative number of samples obtained from the 0-th (batch) dimension (for `MODEL` only)
+- `CALLS`: number of calls
+- `ERRORS`: number of errors generated after the run has been submitted (i.e. excluding errors generated during parsing of the command)
+
+```sql
+AI.INFO <model_or_script_key>
+```
+
+Statistcs are accumulated until the same command with an extra `RESETSTAT` argument is called. This resets the statistics relative to the model or script.
+
+```sql
+AI.INFO <model_or_script_key> RESETSTAT
+```
+
+The command can be called on a key until that key is removed using `MODELDEL` or `SCRIPTDEL`.
+
+### AI.INFO Example
+
+```sql
+AI.INFO amodel
+
+>  1) KEY
+>  2) "amodel"
+>  3) TYPE
+>  4) MODEL
+>  5) BACKEND
+>  6) TORCH
+>  7) DEVICE
+>  8) CPU
+>  9) DURATION
+> 10) (integer) 6511
+> 11) SAMPLES
+> 12) (integer) 2
+> 13) CALLS
+> 14) (integer) 1
+> 15) ERRORS
+> 16) (integer) 0
+```
+
+```sql
+AI.INFO amodel RESETSTAT
+
+> OK
+
+AI.INFO amodel
+
+>  1) KEY
+>  2) "amodel"
+>  3) TYPE
+>  4) MODEL
+>  5) BACKEND
+>  6) TORCH
+>  7) DEVICE
+>  8) CPU
+>  9) DURATION
+> 10) (integer) 0
+> 11) SAMPLES
+> 12) (integer) 0
+> 13) CALLS
+> 14) (integer) 0
+> 15) ERRORS
+> 16) (integer) 0
+```

src/backends.c

@@ -21,7 +21,6 @@ RedisModuleString* RAI_GetModulePath(RedisModuleCtx *ctx) {
   return module_path;
 }
 
-
 RedisModuleString* RAI_GetBackendsPath(RedisModuleCtx *ctx) {
   Dl_info info;
   RedisModuleString* backends_path = NULL;
@@ -36,6 +35,20 @@ RedisModuleString* RAI_GetBackendsPath(RedisModuleCtx *ctx) {
   return backends_path;
 }
 
+const char* RAI_BackendName(int backend) {
+  switch (backend) {
+    case RAI_BACKEND_TENSORFLOW:
+      return "TF";
+    case RAI_BACKEND_TFLITE:
+      return "TFLITE";
+    case RAI_BACKEND_TORCH:
+      return "TORCH";
+    case RAI_BACKEND_ONNXRUNTIME:
+      return "ONNX";
+  }
+  return NULL;
+}
+
 int RAI_LoadBackend_TensorFlow(RedisModuleCtx *ctx, const char *path) {
   if (RAI_backends.tf.model_run != NULL) {
     RedisModule_Log(ctx, "warning", "Could not load TF backend: backend already loaded");

src/backends.h

@@ -35,4 +35,6 @@ char* RAI_BackendsPath;
 int RAI_LoadBackend(RedisModuleCtx *ctx, int backend, const char *path);
 int RAI_LoadDefaultBackend(RedisModuleCtx *ctx, int backend);
 
-#endif
+const char* RAI_BackendName(int backend);
+
+#endif

src/model.c

@@ -135,25 +135,11 @@ static void RAI_Model_AofRewrite(RedisModuleIO *aof, RedisModuleString *key, voi
     array_append(outputs_, RedisModule_CreateString(ctx, model->outputs[i], strlen(model->outputs[i])));
   }
 
-  char backend[256] = "";
-  switch (model->backend) {
-    case RAI_BACKEND_TENSORFLOW:
-      strcpy(backend, "TF");
-      break;
-    case RAI_BACKEND_TFLITE:
-      strcpy(backend, "TFLITE");
-      break;
-    case RAI_BACKEND_TORCH:
-      strcpy(backend, "TORCH");
-      break;
-    case RAI_BACKEND_ONNXRUNTIME:
-      strcpy(backend, "ONNX");
-      break;
-  }
+  const char* backendstr = RAI_BackendName(model->backend);
 
   RedisModule_EmitAOF(aof, "AI.MODELSET", "slccvcvb",
                       key,
-                      backend, model->devicestr,
+                      backendstr, model->devicestr,
                       "INPUTS", inputs_, model->ninputs,
                       "OUTPUTS", outputs_, model->noutputs,
                       buffer, len);

src/model_struct.h

@@ -17,8 +17,6 @@ typedef struct RAI_Model {
   char **outputs;
   size_t noutputs;
   long long refCount;
-  long long backend_calls;
-  long long backend_us;
   void* data;
 } RAI_Model;