fix: [Geneva Uploader] Eliminate hash collisions and redundant MD5 computation in schema deduplication - Perf improvement 5-10% (#486)

Author: lalitb

opentelemetry-exporter-geneva/geneva-uploader/src/payload_encoder/bond_encoder.rs

@@ -107,14 +107,14 @@ pub(crate) struct FieldDef {
 
 /// Schema definition that can be built dynamically
 #[derive(Clone)]
-pub(crate) struct DynamicSchema {
+pub(crate) struct DynamicSchema<'a> {
     pub struct_name: String,
     pub qualified_name: String,
-    pub fields: Vec<FieldDef>,
+    pub fields: &'a [FieldDef],
 }
 
-impl DynamicSchema {
-    pub(crate) fn new(name: &str, namespace: &str, fields: Vec<FieldDef>) -> Self {
+impl<'a> DynamicSchema<'a> {
+    pub(crate) fn new(name: &str, namespace: &str, fields: &'a [FieldDef]) -> Self {
         Self {
             struct_name: name.to_string(),
             qualified_name: format!("{namespace}.{name}"),
@@ -380,7 +380,7 @@ pub(crate) struct BondEncodedSchema {
 }
 
 impl BondEncodedSchema {
-    pub(crate) fn from_fields(name: &str, namespace: &str, fields: Vec<FieldDef>) -> Self {
+    pub(crate) fn from_fields(name: &str, namespace: &str, fields: &[FieldDef]) -> Self {
         let schema = DynamicSchema::new(name, namespace, fields);
         let encoded_bytes = schema.encode().expect("Schema encoding failed");
 
@@ -428,7 +428,7 @@ mod tests {
             },
         ];
 
-        let schema = DynamicSchema::new("TestStruct", "test.namespace", fields);
+        let schema = DynamicSchema::new("TestStruct", "test.namespace", &fields);
         let encoded = schema.encode().unwrap();
         assert!(!encoded.is_empty());
     }
@@ -453,7 +453,7 @@ mod tests {
             },
         ];
 
-        let schema = BondEncodedSchema::from_fields("OtlpLogRecord", "telemetry", fields);
+        let schema = BondEncodedSchema::from_fields("OtlpLogRecord", "telemetry", &fields);
         let bytes = schema.as_bytes();
         assert!(!bytes.is_empty());
     }
@@ -475,7 +475,7 @@ mod tests {
             },
         ];
 
-        let schema = BondEncodedSchema::from_fields("TestStruct", "test.namespace", fields);
+        let schema = BondEncodedSchema::from_fields("TestStruct", "test.namespace", &fields);
         let bytes = schema.as_bytes();
         assert!(!bytes.is_empty());
     }
@@ -504,7 +504,7 @@ mod tests {
                 })
                 .collect();
 
-            let schema = DynamicSchema::new("TestStruct", "test.namespace", fields);
+            let schema = DynamicSchema::new("TestStruct", "test.namespace", &fields);
             let exact_size = schema.calculate_exact_encoded_size();
             let encoded = schema.encode().unwrap();
 

opentelemetry-exporter-geneva/geneva-uploader/src/payload_encoder/central_blob.rs

@@ -1,6 +1,6 @@
 //use md5;
 
-use crate::payload_encoder::bond_encoder::BondEncodedSchema;
+use crate::payload_encoder::bond_encoder::{BondEncodedSchema, FieldDef};
 use chrono::{DateTime, Datelike, Timelike, Utc};
 use std::sync::Arc;
 
@@ -72,6 +72,10 @@ pub(crate) struct CentralSchemaEntry {
     pub id: u64,
     pub md5: [u8; 16],
     pub schema: BondEncodedSchema,
+    // Store fields for deduplication
+    // TODO: It might be faster to encode schema per event and compare encoded bytes with stored schemas
+    // instead of checking each field. Needs benchmarking to determine which approach is faster.
+    pub fields: Vec<FieldDef>,
 }
 
 /// Event/row entry for central blob
@@ -250,7 +254,7 @@ mod tests {
                 field_id: 2u16,
             },
         ];
-        let schema_obj = BondEncodedSchema::from_fields("TestStruct", "test.namespace", fields);
+        let schema_obj = BondEncodedSchema::from_fields("TestStruct", "test.namespace", &fields);
         let schema_bytes = schema_obj.as_bytes().to_vec();
         let schema_md5 = md5_bytes(&schema_bytes);
         let schema_id = 1234u64;
@@ -259,6 +263,7 @@ mod tests {
             id: schema_id,
             md5: schema_md5,
             schema: schema_obj,
+            fields,
         };
 
         // Prepare a row

opentelemetry-exporter-geneva/geneva-uploader/src/payload_encoder/mod.rs

@@ -13,7 +13,7 @@ mod tests {
     use crate::payload_encoder::bond_encoder::FieldDef;
 
     fn create_payload(fields: Vec<FieldDef>, row_data: Vec<u8>) -> Vec<u8> {
-        let schema_obj = BondEncodedSchema::from_fields("MdsContainer", "testNamespace", fields);
+        let schema_obj = BondEncodedSchema::from_fields("MdsContainer", "testNamespace", &fields);
         let schema_bytes = schema_obj.as_bytes();
         let schema_md5 = md5::compute(schema_bytes).0;
         let schema_id = 1u64;
@@ -22,6 +22,7 @@ mod tests {
             id: schema_id,
             md5: schema_md5,
             schema: schema_obj,
+            fields,
         };
 
         let event = CentralEventEntry {