Differentiate batch and online feature requests

zhilingc · zhilingc · commit e1973aace291 · 2019-10-14T11:32:48.000+08:00
diff --git a/protos/feast/serving/ServingService.proto b/protos/feast/serving/ServingService.proto
@@ -34,17 +34,17 @@ service ServingService {
     rpc GetFeastServingType (GetFeastServingTypeRequest) returns (GetFeastServingTypeResponse);
 
     // Get online features synchronously.
-    rpc GetOnlineFeatures (GetFeaturesRequest) returns (GetOnlineFeaturesResponse);
+    rpc GetOnlineFeatures (GetOnlineFeaturesRequest) returns (GetOnlineFeaturesResponse);
 
-    // Get batch features asynchronously. 
-    // 
+    // Get batch features asynchronously.
+    //
     // The client should check the status of the returned job periodically by
-    // calling ReloadJob to determine if the job has completed successfully 
+    // calling ReloadJob to determine if the job has completed successfully
     // or with an error. If the job completes successfully i.e.
     // status = JOB_STATUS_DONE with no error, then the client can check
     // the file_uris for the location to download feature values data.
     // The client is assumed to have access to these file URIs.
-    rpc GetBatchFeatures (GetFeaturesRequest) returns (GetBatchFeaturesResponse);
+    rpc GetBatchFeatures (GetBatchFeaturesRequest) returns (GetBatchFeaturesResponse);
 
     // Get the latest job status for batch feature retrieval.
     rpc GetJob (GetJobRequest) returns (GetJobResponse);
@@ -62,7 +62,7 @@ message GetFeastServingTypeResponse {
     FeastServingType type = 1;
 }
 
-message GetFeaturesRequest {
+message GetOnlineFeaturesRequest {
 
     // List of feature sets and their features that are being retrieved
     repeated FeatureSet feature_sets = 1;
@@ -72,6 +72,8 @@ message GetFeaturesRequest {
     // rows into a final dataset
     repeated EntityRow entity_rows = 2;
 
+    // Option to omit entities from the response. If true, only feature
+    // values will be returned.
     bool omit_entities_in_response = 3;
 
     message FeatureSet {
@@ -87,7 +89,7 @@ message GetFeaturesRequest {
         // The features will be retrieved if:
         // entity_timestamp - max_age  <= event_timestamp <= entity_timestamp
         //
-        // If unspecified the default max_age specified in FeatureSetSpec will 
+        // If unspecified the default max_age specified in FeatureSetSpec will
         // be used.
         google.protobuf.Duration max_age = 4;
     }
@@ -103,12 +105,41 @@ message GetFeaturesRequest {
     }
 }
 
+message GetBatchFeaturesRequest {
+
+    // List of feature sets and their features that are being retrieved.
+    repeated FeatureSet feature_sets = 1;
+
+    // Source of the entity dataset containing the timestamps and entity keys to retrieve
+    // features for.
+    DatasetSource dataset_source = 2;
+
+    message FeatureSet {
+        // Feature set name
+        string name = 1;
+
+        // Feature set version
+        int32 version = 2;
+
+        // Features that should be retrieved from this feature set
+        repeated string feature_names = 3;
+
+        // The features will be retrieved if:
+        // entity_timestamp - max_age  <= event_timestamp <= entity_timestamp
+        //
+        // If unspecified the default max_age specified in FeatureSetSpec will
+        // be used.
+        google.protobuf.Duration max_age = 4;
+    }
+}
+
 message GetOnlineFeaturesResponse {
+    // Feature values retrieved from feast.
     repeated FieldValues field_values = 1;
 
-    // TODO: update this comment
-    // does not include timestamp, includes features and entities
     message FieldValues {
+        // Map of feature or entity name to feature/entity values.
+        // Timestamps are not returned in this response.
         map<string,feast.types.Value> fields = 1;
     }
 }
@@ -127,10 +158,10 @@ message GetJobResponse {
 
 enum FeastServingType {
     FEAST_SERVING_TYPE_INVALID = 0;
-    // Online serving receives entity data directly and synchronously and will 
+    // Online serving receives entity data directly and synchronously and will
     // respond immediately.
     FEAST_SERVING_TYPE_ONLINE = 1;
-    // Batch serving receives entity data asynchronously and orchestrates the 
+    // Batch serving receives entity data asynchronously and orchestrates the
     // retrieval through a staging location.
     FEAST_SERVING_TYPE_BATCH = 2;
 }
@@ -163,10 +194,31 @@ message Job {
     JobStatus status = 3;
     // Output only. If not empty, the job has failed with this error message.
     string error = 4;
-    // Output only. The list of URIs for the files to be downloaded or 
+    // Output only. The list of URIs for the files to be downloaded or
     // uploaded (depends on the job type) for this particular job.
     repeated string file_uris = 5;
     // Output only. The data format for all the files.
     // For CSV format, the files contain both feature values and a column header.
     DataFormat data_format = 6;
 }
+
+message DatasetSource {
+    oneof dataset_source {
+        // SQL query to generate the dataset from. It is recommended that the SQL follows the
+        // ANSI standard.
+        string sql_query = 1;
+
+        // File source to load the dataset from.
+        FileSource file_source = 2;
+    }
+
+    message FileSource {
+        // URIs to retrieve the dataset from, e.g. gs://bucket/directory/object.csv. Wildcards are
+        // supported. This data must be compatible to be uploaded to the serving store, and also be
+        // accessible by this serving instance.
+        repeated string file_uris = 1;
+
+        // Format of the data.
+        DataFormat data_format = 2;
+    }
+}
