Skip to content

Open Source alternative to Algolia + Pinecone and an Easier-to-Use alternative to ElasticSearch ⚡ 🔍 ✨ Fast, typo tolerant, in-memory fuzzy Search Engine for building delightful search experiences

Fix updating auto embedding field indexes

ozanarmagan46b5847869d9
diff --git a/src/index.cpp b/src/index.cpp
index f027c09e..f9d755d8 100644
--- a/src/index.cpp
+++ b/src/index.cpp
@@ -6722,13 +6722,10 @@ void Index::batch_embed_fields(std::vector<index_record*>& records,
texts_to_embed[i].first->index_failure(embedding_res.status_code, "");
continue;
}
- nlohmann::json* document;
if(texts_to_embed[i].first->is_update) {
- document = &texts_to_embed[i].first->new_doc;
- } else {
- document = &texts_to_embed[i].first->doc;
- }
- (*document)[field.name] = embedding_res.embedding;
+ texts_to_embed[i].first->new_doc[field.name] = embedding_res.embedding;
+ }
+ texts_to_embed[i].first->doc[field.name] = embedding_res.embedding;
}
}
}
diff --git a/test/collection_vector_search_test.cpp b/test/collection_vector_search_test.cpp
index 04cd0096..72244efe 100644
--- a/test/collection_vector_search_test.cpp
+++ b/test/collection_vector_search_test.cpp
@@ -2729,4 +2729,97 @@ TEST_F(CollectionVectorTest, TestSearchNonIndexedVectorField) {
ASSERT_FALSE(search_result.ok());
ASSERT_EQ("Field `vec` is marked as a non-indexed field in the schema.", search_result.error());
+}
+
+
+TEST_F(CollectionVectorTest, TestSemanticSearchAfterUpdate) {
+ nlohmann::json schema = R"({
+ "name": "test",
+ "fields": [
+ {
+ "name": "name",
+ "type": "string"
+ },
+ {
+ "name": "embedding",
+ "type": "float[]",
+ "embed": {
+ "from": [
+ "name"
+ ],
+ "model_config": {
+ "model_name": "ts/e5-small"
+ }
+ }
+ }
+ ]
+ })"_json;
+
+ TextEmbedderManager::set_model_dir("/tmp/typesense_test/models");
+
+ auto collection_create_op = collectionManager.create_collection(schema);
+ ASSERT_TRUE(collection_create_op.ok());
+
+ auto coll = collection_create_op.get();
+
+ auto add_op = coll->add(R"({
+ "name": "soccer",
+ "id": "0"
+ })"_json.dump());
+
+ ASSERT_TRUE(add_op.ok());
+
+ add_op = coll->add(R"({
+ "name": "basketball",
+ "id": "1"
+ })"_json.dump());
+
+ ASSERT_TRUE(add_op.ok());
+
+ add_op = coll->add(R"({
+ "name": "typesense",
+ "id": "2"
+ })"_json.dump());
+
+ ASSERT_TRUE(add_op.ok());
+
+ add_op = coll->add(R"({
+ "name": "potato",
+ "id": "3"
+ })"_json.dump());
+
+ ASSERT_TRUE(add_op.ok());
+
+ auto result = coll->search("*", {}, "", {}, {}, {0}, 20, 1, FREQUENCY, {true}, Index::DROP_TOKENS_THRESHOLD,
+ spp::sparse_hash_set<std::string>(),
+ spp::sparse_hash_set<std::string>(), 10, "", 30, 5,
+ "", 10, {}, {}, {}, 0,
+ "<mark>", "</mark>", {}, 1000, true, false, true, "", false, 6000 * 1000, 4, 7,
+ fallback,
+ 4, {off}, 32767, 32767, 2,
+ false, true, "embedding:([], id:0, k:1)");
+
+ ASSERT_TRUE(result.ok());
+ ASSERT_EQ(1, result.get()["hits"].size());
+ ASSERT_EQ("basketball", result.get()["hits"][0]["document"]["name"]);
+
+ auto update_op = coll->add(R"({
+ "name": "onion",
+ "id": "0"
+ })"_json.dump(), index_operation_t::UPDATE, "0");
+
+ ASSERT_TRUE(update_op.ok());
+
+ result = coll->search("*", {}, "", {}, {}, {0}, 20, 1, FREQUENCY, {true}, Index::DROP_TOKENS_THRESHOLD,
+ spp::sparse_hash_set<std::string>(),
+ spp::sparse_hash_set<std::string>(), 10, "", 30, 5,
+ "", 10, {}, {}, {}, 0,
+ "<mark>", "</mark>", {}, 1000, true, false, true, "", false, 6000 * 1000, 4, 7,
+ fallback,
+ 4, {off}, 32767, 32767, 2,
+ false, true, "embedding:([], id:0, k:1)");
+
+ ASSERT_TRUE(result.ok());
+ ASSERT_EQ(1, result.get()["hits"].size());
+ ASSERT_EQ("potato", result.get()["hits"][0]["document"]["name"]);
}
\ No newline at end of file