#105 AI command inconsistencies (#143)

Author: lordsarcastic

ahnlich/ai/src/server/task.rs

@@ -1,11 +1,8 @@
 use crate::engine::ai::models::Model;
 use ahnlich_client_rs::{builders::db as db_params, db::DbClient};
-use ahnlich_types::ai::{
-    AIQuery, AIServerQuery, AIServerResponse, AIServerResult, PreprocessAction,
-};
+use ahnlich_types::ai::{AIQuery, AIServerQuery, AIServerResponse, AIServerResult};
 use ahnlich_types::client::ConnectedClient;
 use ahnlich_types::db::{ServerInfo, ServerResponse};
-use ahnlich_types::keyval::StoreInput;
 use ahnlich_types::metadata::MetadataValue;
 use ahnlich_types::predicate::{Predicate, PredicateCondition};
 use ahnlich_types::version::VERSION;
@@ -345,20 +342,15 @@ impl AhnlichProtocol for AIProxyTask {
                     condition,
                     closest_n,
                     algorithm,
+                    preprocess_action,
                 } => {
-                    // TODO: Replace this with calls to self.model_manager.handle_request
-                    // TODO (HAKSOAT): Shouldn't preprocess action also be in the params?
-                    let preprocess = match search_input {
-                        StoreInput::RawString(_) => PreprocessAction::ModelPreprocessing,
-                        StoreInput::Image(_) => PreprocessAction::ModelPreprocessing,
-                    };
                     let repr = self
                         .store_handler
                         .get_ndarray_repr_for_store(
                             &store,
                             search_input,
                             &self.model_manager,
-                            preprocess,
+                            preprocess_action,
                         )
                         .await;
                     match repr {
@@ -405,6 +397,39 @@ impl AhnlichProtocol for AIProxyTask {
                     let destoryed = self.store_handler.purge_stores();
                     Ok(AIServerResponse::Del(destoryed))
                 }
+                AIQuery::ListClients => {
+                    Ok(AIServerResponse::ClientList(self.client_handler.list()))
+                }
+                AIQuery::GetKey { store, keys } => {
+                    let metadata_values: HashSet<MetadataValue> =
+                        keys.into_iter().map(|value| value.into()).collect();
+                    let get_key_condition = PredicateCondition::Value(Predicate::In {
+                        key: AHNLICH_AI_RESERVED_META_KEY.clone(),
+                        value: metadata_values,
+                    });
+
+                    let get_pred_params = db_params::GetPredParams::builder()
+                        .store(store.to_string())
+                        .condition(get_key_condition)
+                        .tracing_id(parent_id.clone())
+                        .build();
+
+                    match self.db_client.get_pred(get_pred_params).await {
+                        Ok(res) => {
+                            if let ServerResponse::Get(response) = res {
+                                // conversion to store input here
+                                let output = self
+                                    .store_handler
+                                    .store_key_val_to_store_input_val(response);
+                                Ok(AIServerResponse::Get(output))
+                            } else {
+                                Err(AIProxyError::UnexpectedDBResponse(format!("{:?}", res))
+                                    .to_string())
+                            }
+                        }
+                        Err(err) => Err(format!("{err}")),
+                    }
+                }
             })
         }
         result

ahnlich/ai/src/tests/aiproxy_test.rs

@@ -11,11 +11,16 @@ use ahnlich_types::{
     predicate::{Predicate, PredicateCondition},
     similarity::Algorithm,
 };
+// use flurry::HashMap;
 use utils::server::AhnlichServerUtils;
 
 use once_cell::sync::Lazy;
 use pretty_assertions::assert_eq;
-use std::{collections::HashSet, num::NonZeroUsize, sync::atomic::Ordering};
+use std::{
+    collections::{HashMap, HashSet},
+    num::NonZeroUsize,
+    sync::atomic::Ordering,
+};
 
 use crate::{
     cli::{server::SupportedModels, AIProxyConfig},
@@ -162,6 +167,73 @@ async fn test_ai_proxy_create_store_success() {
     query_server_assert_result(&mut reader, message, expected.clone()).await;
 }
 
+#[tokio::test]
+async fn test_ai_store_get_key_works() {
+    let address = provision_test_servers().await;
+    let first_stream = TcpStream::connect(address).await.unwrap();
+    let second_stream = TcpStream::connect(address).await.unwrap();
+    let store_name = StoreName(String::from("Deven Kicks"));
+    let store_input = StoreInput::RawString(String::from("Jordan 3"));
+    let store_data: (StoreInput, HashMap<MetadataKey, MetadataValue>) =
+        (store_input.clone(), HashMap::new());
+
+    let message = AIServerQuery::from_queries(&[
+        AIQuery::CreateStore {
+            store: store_name.clone(),
+            query_model: AIModel::AllMiniLML6V2,
+            index_model: AIModel::AllMiniLML6V2,
+            predicates: HashSet::new(),
+            non_linear_indices: HashSet::new(),
+            error_if_exists: true,
+            store_original: false,
+        },
+        AIQuery::Set {
+            store: store_name.clone(),
+            inputs: vec![store_data.clone()],
+            preprocess_action: PreprocessAction::NoPreprocessing,
+        },
+    ]);
+    let mut reader = BufReader::new(first_stream);
+
+    let _ = get_server_response(&mut reader, message).await;
+    let message = AIServerQuery::from_queries(&[AIQuery::GetKey {
+        store: store_name,
+        keys: vec![store_input.clone()],
+    }]);
+
+    let mut expected = AIServerResult::with_capacity(1);
+
+    expected.push(Ok(AIServerResponse::Get(vec![(
+        Some(store_input),
+        HashMap::new(),
+    )])));
+
+    let mut reader = BufReader::new(second_stream);
+    let response = get_server_response(&mut reader, message).await;
+    assert!(response.len() == expected.len())
+}
+
+#[tokio::test]
+async fn test_list_clients_works() {
+    let address = provision_test_servers().await;
+    let _first_stream = TcpStream::connect(address).await.unwrap();
+    let second_stream = TcpStream::connect(address).await.unwrap();
+    let message = AIServerQuery::from_queries(&[AIQuery::ListClients]);
+    let mut reader = BufReader::new(second_stream);
+    let response = get_server_response(&mut reader, message).await;
+    let inner = response.into_inner();
+
+    // only two clients are connected
+    match inner.as_slice() {
+        [Ok(AIServerResponse::ClientList(connected_clients))] => {
+            assert!(connected_clients.len() == 2)
+        }
+        a => {
+            assert!(false, "Unexpected result for client list {:?}", a);
+        }
+    };
+}
+
 // TODO: Same issues with random storekeys, changing the order of expected response
 #[tokio::test]
 async fn test_ai_store_no_original() {
@@ -375,6 +447,7 @@ async fn test_ai_proxy_get_sim_n_succeeds() {
         condition: None,
         closest_n: NonZeroUsize::new(1).unwrap(),
         algorithm: Algorithm::DotProductSimilarity,
+        preprocess_action: PreprocessAction::ModelPreprocessing,
     }]);
 
     let mut expected = AIServerResult::with_capacity(1);

ahnlich/client/src/ai.rs

@@ -76,6 +76,7 @@ impl AIPipeline {
             condition: params.condition,
             closest_n: params.closest_n,
             algorithm: params.algorithm,
+            preprocess_action: params.preprocess_action,
         })
     }
 
@@ -234,6 +235,7 @@ impl AIClient {
                 condition: params.condition,
                 closest_n: params.closest_n,
                 algorithm: params.algorithm,
+                preprocess_action: params.preprocess_action,
             },
             params.tracing_id,
         )

ahnlich/client/src/builders/ai.rs

@@ -63,6 +63,8 @@ pub struct GetSimNParams {
 
     #[builder(default = None)]
     pub tracing_id: Option<String>,
+    #[builder(default = PreprocessAction::NoPreprocessing)]
+    pub preprocess_action: PreprocessAction,
 }
 
 #[derive(TypedBuilder)]

ahnlich/db/src/engine/predicate.rs

@@ -19,6 +19,29 @@ use std::collections::HashSet as StdHashSet;
 use std::mem::size_of_val;
 use utils::parallel;
 
+/// Predicates are essentially nested hashmaps that let us retrieve original keys that match a
+/// precise value. Take the following example
+///
+/// {
+///     "Country": {
+///         "Nigeria": [StoreKeyId(1), StoreKeyId(2)],
+///         "Australia": ..,
+///     },
+///     "Author": {
+///         ...
+///     }
+/// }
+///
+/// where `allowed_predicates` = ["Country", "Author"]
+///
+/// It takes less time to retrieve "where country = 'Nigeria'" by traversing the nested hashmap to
+/// obtain StoreKeyId(1) and StoreKeyId(2) than it would be to make a linear pass over an entire
+/// Store of size N comparing their metadata "country" along the way. Given that StoreKeyId is
+/// computed via blake hash, it is typically fast to compute and also of a fixed size which means
+/// predicate indices don't balloon with large metadata
+///
+/// Whichever key is not expressly included in `allowed_predicates` goes through the linear
+/// pass in order to obtain keys that satisfy the condition
 type InnerPredicateIndexVal = ConcurrentHashSet<StoreKeyId>;
 type InnerPredicateIndex = ConcurrentHashMap<MetadataValue, InnerPredicateIndexVal>;
 type InnerPredicateIndices = ConcurrentHashMap<MetadataKey, PredicateIndex>;

ahnlich/dsl/src/ai.rs

@@ -18,11 +18,11 @@ use pest::Parser;
 
 use crate::{error::DslError, predicate::parse_predicate_expression};
 
-fn parse_to_preprocess_action(input: &str) -> PreprocessAction {
+fn parse_to_preprocess_action(input: &str) -> Result<PreprocessAction, DslError> {
     match input.to_lowercase().trim() {
-        "nopreprocessing" => PreprocessAction::NoPreprocessing,
-        "modelpreprocessing" => PreprocessAction::ModelPreprocessing,
-        _ => panic!("Unexpected preprocess action"),
+        "nopreprocessing" => Ok(PreprocessAction::NoPreprocessing),
+        "modelpreprocessing" => Ok(PreprocessAction::ModelPreprocessing),
+        a => Err(DslError::UnsupportedPreprocessingMode(a.to_string())),
     }
 }
 
@@ -53,7 +53,7 @@ pub const COMMANDS: &[&str] = &[
     "dropnonlinearalgorithmindex",   // if exists (kdtree) in store_name
     "delkey",                        // ([input 1 text], [input 2 text]) in my_store
     "getpred",                       // ((author = dickens) or (country != Nigeria)) in my_store
-    "getsimn", // 4 with [random text inserted here] using cosinesimilarity in my_store where (author = dickens)
+    "getsimn", // 4 with [random text inserted here] using cosinesimilarity preprocessaction nopreprocessing in my_store where (author = dickens)
     "createstore", // if not exists my_store querymodel resnet-50 indexmodel resnet-50 predicates (author, country) nonlinearalgorithmindex (kdtree)
     "set", // (([This is the life of Haks paragraphed], {name: Haks, category: dev}), ([This is the life of Deven paragraphed], {name: Deven, category: dev})) in store
 ];
@@ -83,9 +83,9 @@ pub fn parse_ai_query(input: &str) -> Result<Vec<AIQuery>, DslError> {
                 let preprocess_action = parse_to_preprocess_action(
                     inner_pairs
                         .next()
-                        .ok_or(DslError::UnexpectedSpan((start_pos, end_pos)))?
-                        .as_str(),
-                );
+                        .map(|a| a.as_str())
+                        .unwrap_or("nopreprocessing"),
+                )?;
 
                 AIQuery::Set {
                     store: StoreName(store.to_string()),
@@ -175,6 +175,18 @@ pub fn parse_ai_query(input: &str) -> Result<Vec<AIQuery>, DslError> {
                         .ok_or(DslError::UnexpectedSpan((start_pos, end_pos)))?
                         .as_str(),
                 )?;
+                let mut preprocess_action = PreprocessAction::NoPreprocessing;
+                if let Some(next_pair) = inner_pairs.peek() {
+                    if next_pair.as_rule() == Rule::preprocess_optional {
+                        let mut pair = inner_pairs
+                            .next()
+                            .ok_or(DslError::UnexpectedSpan((start_pos, end_pos)))?
+                            .into_inner();
+                        preprocess_action = parse_to_preprocess_action(
+                            pair.next().map(|a| a.as_str()).unwrap_or("nopreprocessing"),
+                        )?;
+                    }
+                };
                 let store = inner_pairs
                     .next()
                     .ok_or(DslError::UnexpectedSpan((start_pos, end_pos)))?
@@ -190,6 +202,7 @@ pub fn parse_ai_query(input: &str) -> Result<Vec<AIQuery>, DslError> {
                     closest_n,
                     algorithm,
                     condition,
+                    preprocess_action,
                 }
             }
             Rule::get_pred => {

ahnlich/dsl/src/error.rs

@@ -19,4 +19,6 @@ pub enum DslError {
     UnsupportedAIModel(String),
     #[error("Unsupported rule used in parse fn {0:?}")]
     UnsupportedRule(Rule),
+    #[error("Unexpected preprocessing {0:?}")]
+    UnsupportedPreprocessingMode(String),
 }

ahnlich/dsl/src/syntax/syntax.pest

@@ -56,7 +56,7 @@ ai_del_key = { whitespace* ~ ^"delkey" ~ whitespace* ~ "(" ~ store_inputs ~ ")"
 get_pred = { whitespace* ~ ^"getpred" ~ whitespace* ~ predicate_condition ~ in_ignored ~ store_name }
 // GETSIMN 2 WITH store-key USING algorithm IN store (WHERE predicate_condition)
 get_sim_n = { whitespace* ~ ^"getsimn" ~ whitespace* ~ non_zero ~ whitespace* ~ ^"with" ~ whitespace* ~ f32_array ~ whitespace* ~ ^"using" ~ whitespace* ~ algorithm ~ whitespace* ~ in_ignored ~ whitespace* ~ store_name ~ whitespace* ~ (^"where" ~ whitespace* ~ predicate_condition)? }
-ai_get_sim_n = { whitespace* ~ ^"getsimn" ~ whitespace* ~ non_zero ~ whitespace* ~ ^"with" ~ whitespace* ~ "[" ~ whitespace* ~ metadata_value ~ whitespace* ~ "]" ~ whitespace* ~ ^"using" ~ whitespace* ~ algorithm ~ whitespace* ~ in_ignored ~ whitespace* ~ store_name ~ whitespace* ~ (^"where" ~ whitespace* ~ predicate_condition)? }
+ai_get_sim_n = { whitespace* ~ ^"getsimn" ~ whitespace* ~ non_zero ~ whitespace* ~ ^"with" ~ whitespace* ~ "[" ~ whitespace* ~ metadata_value ~ whitespace* ~ "]" ~ whitespace* ~ ^"using" ~ whitespace* ~ algorithm ~ whitespace* ~ (preprocess_optional)? ~ whitespace* ~ in_ignored ~ whitespace* ~ store_name ~ whitespace* ~ (^"where" ~ whitespace* ~ predicate_condition)? }
 // CREATESTORE IF NOT EXISTS store-name DIMENSION non-zero-size PREDICATES (key1, key2) NONLINEARALGORITHMINDEX (kdtree) 
 create_store = { whitespace* ~ ^"createstore" ~ whitespace* ~ (if_not_exists)? ~ whitespace* ~ store_name ~ whitespace* ~ ^"dimension" ~ whitespace* ~ non_zero ~ whitespace* ~ (^"predicates" ~ whitespace* ~ "(" ~ whitespace* ~ metadata_keys ~ whitespace* ~ ")" )? ~ (whitespace* ~ ^"nonlinearalgorithmindex" ~ whitespace* ~ "(" ~ whitespace* ~ non_linear_algorithms ~ whitespace* ~ ")")? }
 // CREATESTORE IF NOT EXISTS store-name QUERYMODEL model INDEXMODEL model PREDICATES (key1, key2) NONLINEARALGORITHMINDEX (kdtree) 
@@ -66,6 +66,7 @@ ai_set_in_store = { whitespace* ~ ^"set" ~ whitespace* ~ store_inputs_to_store_v
 
 if_exists = { whitespace* ~ ^"if" ~ whitespace* ~ ^"exists" ~ whitespace* }
 if_not_exists = { whitespace* ~ ^"if" ~ whitespace* ~ ^"not" ~ whitespace* ~ ^"exists" ~ whitespace* }
+preprocess_optional = { whitespace* ~ ^"preprocessaction" ~ whitespace* ~ preprocess_action}
 store_original = { whitespace* ~ ^"storeoriginal" ~ whitespace* }
 
 // stores and predicates can be alphanumeric

ahnlich/dsl/src/tests/ai.rs

@@ -199,15 +199,16 @@ fn test_get_sim_n_parse() {
         panic!("Unexpected error pattern found")
     };
     assert_eq!((start, end), (0, 68));
-    let input = r#"GETSIMN 5 with [hi my name is carter] using cosinesimilarity in random"#;
+    let input = r#"GETSIMN 5 with [hi my name is carter] using cosinesimilarity preprocessaction MODELPREPROCESSING in random"#;
     assert_eq!(
         parse_ai_query(input).expect("Could not parse query input"),
         vec![AIQuery::GetSimN {
             store: StoreName("random".to_string()),
             search_input: StoreInput::RawString("hi my name is carter".to_string()),
             closest_n: NonZeroUsize::new(5).unwrap(),
             algorithm: Algorithm::CosineSimilarity,
-            condition: None
+            condition: None,
+            preprocess_action: PreprocessAction::ModelPreprocessing,
         }]
     );
     let input = r#"GETSIMN 8 with [testing the limits of life] using euclideandistance in other where ((year != 2012) AND (month not in (december, october)))"#;
@@ -231,6 +232,7 @@ fn test_get_sim_n_parse() {
                     ]),
                 }))
             ),
+            preprocess_action: PreprocessAction::NoPreprocessing,
         }]
     );
 }