Merge pull request #944 from Lorak-mmk/shard-selecting-lb-v2

Shard selecting load balancing

docs/source/load-balancing/load-balancing.md

@@ -2,8 +2,8 @@
 
 ## Introduction
 
-The driver uses a load balancing policy to determine which node(s) to contact
-when executing a query. Load balancing policies implement the
+The driver uses a load balancing policy to determine which node(s) and shard(s)
+to contact when executing a query. Load balancing policies implement the
 `LoadBalancingPolicy` trait, which contains methods to generate a load
 balancing plan based on the query information and the state of the cluster.
 
@@ -12,12 +12,14 @@ being opened. For a node connection blacklist configuration refer to
 `scylla::transport::host_filter::HostFilter`, which can be set session-wide
 using `SessionBuilder::host_filter` method.
 
+In this chapter, "target" will refer to a pair `<node, optional shard>`.
+
 ## Plan
 
 When a query is prepared to be sent to the database, the load balancing policy
-constructs a load balancing plan. This plan is essentially a list of nodes to
+constructs a load balancing plan. This plan is essentially a list of targets to
 which the driver will try to send the query. The first elements of the plan are
-the nodes which are the best to contact (e.g. they might be replicas for the
+the targets which are the best to contact (e.g. they might be replicas for the
 requested data or have the best latency).
 
 ## Policy
@@ -84,17 +86,16 @@ first element of the load balancing plan is needed, so it's usually unnecessary
 to compute entire load balancing plan. To optimize this common case, the
 `LoadBalancingPolicy` trait provides two methods: `pick` and `fallback`.
 
-`pick` returns the first node to contact for a given query, which is usually
-the best based on a particular load balancing policy. If `pick` returns `None`,
-then `fallback` will not be called.
+`pick` returns the first target to contact for a given query, which is usually
+the best based on a particular load balancing policy.
 
-`fallback`, returns an iterator that provides the rest of the nodes in the load
-balancing plan. `fallback` is called only when using the initial picked node
-fails (or when executing speculatively).
+`fallback`, returns an iterator that provides the rest of the targets in the
+load balancing plan. `fallback` is called when using the initial picked
+target fails (or when executing speculatively) or when `pick` returned `None`.
 
-It's possible for the `fallback` method to include the same node that was
+It's possible for the `fallback` method to include the same target that was
 returned by the `pick` method. In such cases, the query execution layer filters
-out the picked node from the iterator returned by `fallback`.
+out the picked target from the iterator returned by `fallback`.
 
 ### `on_query_success` and `on_query_failure`:
 

examples/Cargo.toml

@@ -20,6 +20,7 @@ uuid = "1.0"
 tower = "0.4"
 stats_alloc = "0.1"
 clap = { version = "3.2.4", features = ["derive"] }
+rand = "0.8.5"
 
 [[example]]
 name = "auth"

examples/compare-tokens.rs

@@ -41,7 +41,7 @@ async fn main() -> Result<()> {
                 .get_cluster_data()
                 .get_token_endpoints("examples_ks", Token { value: t })
                 .iter()
-                .map(|n| n.address)
+                .map(|(node, _shard)| node.address)
                 .collect::<Vec<NodeAddr>>()
         );
 

examples/custom_load_balancing_policy.rs

@@ -1,23 +1,37 @@
 use anyhow::Result;
+use rand::thread_rng;
+use rand::Rng;
+use scylla::transport::NodeRef;
 use scylla::{
     load_balancing::{LoadBalancingPolicy, RoutingInfo},
+    routing::Shard,
     transport::{ClusterData, ExecutionProfile},
     Session, SessionBuilder,
 };
 use std::{env, sync::Arc};
 
 /// Example load balancing policy that prefers nodes from favorite datacenter
+/// This is, of course, very naive, as it is completely non token-aware.
+/// For more realistic implementation, see [`DefaultPolicy`](scylla::load_balancing::DefaultPolicy).
 #[derive(Debug)]
 struct CustomLoadBalancingPolicy {
     fav_datacenter_name: String,
 }
 
+fn with_random_shard(node: NodeRef) -> (NodeRef, Shard) {
+    let nr_shards = node
+        .sharder()
+        .map(|sharder| sharder.nr_shards.get())
+        .unwrap_or(1);
+    (node, thread_rng().gen_range(0..nr_shards) as Shard)
+}
+
 impl LoadBalancingPolicy for CustomLoadBalancingPolicy {
     fn pick<'a>(
         &'a self,
         _info: &'a RoutingInfo,
         cluster: &'a ClusterData,
-    ) -> Option<scylla::transport::NodeRef<'a>> {
+    ) -> Option<(NodeRef<'a>, Shard)> {
         self.fallback(_info, cluster).next()
     }
 
@@ -31,9 +45,9 @@ impl LoadBalancingPolicy for CustomLoadBalancingPolicy {
             .unique_nodes_in_datacenter_ring(&self.fav_datacenter_name);
 
         match fav_dc_nodes {
-            Some(nodes) => Box::new(nodes.iter()),
+            Some(nodes) => Box::new(nodes.iter().map(with_random_shard)),
             // If there is no dc with provided name, fallback to other datacenters
-            None => Box::new(cluster.get_nodes_info().iter()),
+            None => Box::new(cluster.get_nodes_info().iter().map(with_random_shard)),
         }
     }
 

scylla/src/transport/cluster.rs

@@ -1,7 +1,7 @@
 /// Cluster manages up to date information and connections to database nodes
 use crate::frame::response::event::{Event, StatusChangeEvent};
 use crate::prepared_statement::TokenCalculationError;
-use crate::routing::Token;
+use crate::routing::{Shard, Token};
 use crate::transport::host_filter::HostFilter;
 use crate::transport::{
     connection::{Connection, VerifiedKeyspaceName},
@@ -27,6 +27,7 @@ use tracing::{debug, warn};
 use uuid::Uuid;
 
 use super::node::{KnownNode, NodeAddr};
+use super::NodeRef;
 
 use super::locator::ReplicaLocator;
 use super::partitioner::calculate_token_for_partition_key;
@@ -408,17 +409,17 @@ impl ClusterData {
     }
 
     /// Access to replicas owning a given token
-    pub fn get_token_endpoints(&self, keyspace: &str, token: Token) -> Vec<Arc<Node>> {
+    pub fn get_token_endpoints(&self, keyspace: &str, token: Token) -> Vec<(Arc<Node>, Shard)> {
         self.get_token_endpoints_iter(keyspace, token)
-            .cloned()
+            .map(|(node, shard)| (node.clone(), shard))
             .collect()
     }
 
     pub(crate) fn get_token_endpoints_iter(
         &self,
         keyspace: &str,
         token: Token,
-    ) -> impl Iterator<Item = &Arc<Node>> {
+    ) -> impl Iterator<Item = (NodeRef<'_>, Shard)> {
         let keyspace = self.keyspaces.get(keyspace);
         let strategy = keyspace
             .map(|k| &k.strategy)
@@ -436,7 +437,7 @@ impl ClusterData {
         keyspace: &str,
         table: &str,
         partition_key: &SerializedValues,
-    ) -> Result<Vec<Arc<Node>>, BadQuery> {
+    ) -> Result<Vec<(Arc<Node>, Shard)>, BadQuery> {
         Ok(self.get_token_endpoints(
             keyspace,
             self.compute_token(keyspace, table, partition_key)?,

scylla/src/transport/connection_pool.rs

@@ -1,7 +1,7 @@
 #[cfg(feature = "cloud")]
 use crate::cloud::set_ssl_config_for_scylla_cloud_host;
 
-use crate::routing::{Shard, ShardCount, Sharder, Token};
+use crate::routing::{Shard, ShardCount, Sharder};
 use crate::transport::errors::QueryError;
 use crate::transport::{
     connection,
@@ -28,7 +28,7 @@ use std::time::Duration;
 
 use tokio::sync::{broadcast, mpsc, Notify};
 use tracing::instrument::WithSubscriber;
-use tracing::{debug, trace, warn};
+use tracing::{debug, error, trace, warn};
 
 /// The target size of a per-node connection pool.
 #[derive(Debug, Clone, Copy)]
@@ -235,22 +235,25 @@ impl NodeConnectionPool {
         .unwrap_or(None)
     }
 
-    pub(crate) fn connection_for_token(&self, token: Token) -> Result<Arc<Connection>, QueryError> {
-        trace!(token = token.value, "Selecting connection for token");
+    pub(crate) fn connection_for_shard(&self, shard: Shard) -> Result<Arc<Connection>, QueryError> {
+        trace!(shard = shard, "Selecting connection for shard");
         self.with_connections(|pool_conns| match pool_conns {
             PoolConnections::NotSharded(conns) => {
                 Self::choose_random_connection_from_slice(conns).unwrap()
             }
             PoolConnections::Sharded {
-                sharder,
                 connections,
+                sharder
             } => {
-                let shard: u16 = sharder
-                    .shard_of(token)
+                let shard = shard
                     .try_into()
-                    .expect("Shard number doesn't fit in u16");
-                trace!(shard = shard, "Selecting connection for token");
-                Self::connection_for_shard(shard, sharder.nr_shards, connections.as_slice())
+                    // It's safer to use 0 rather that panic here, as shards are returned by `LoadBalancingPolicy`
+                    // now, which can be implemented by a user in an arbitrary way.
+                    .unwrap_or_else(|_| {
+                        error!("The provided shard number: {} does not fit u16! Using 0 as the shard number. Check your LoadBalancingPolicy implementation.", shard);
+                        0
+                    });
+                Self::connection_for_shard_helper(shard, sharder.nr_shards, connections.as_slice())
             }
         })
     }
@@ -266,13 +269,13 @@ impl NodeConnectionPool {
                 connections,
             } => {
                 let shard: u16 = rand::thread_rng().gen_range(0..sharder.nr_shards.get());
-                Self::connection_for_shard(shard, sharder.nr_shards, connections.as_slice())
+                Self::connection_for_shard_helper(shard, sharder.nr_shards, connections.as_slice())
             }
         })
     }
 
     // Tries to get a connection to given shard, if it's broken returns any working connection
-    fn connection_for_shard(
+    fn connection_for_shard_helper(
         shard: u16,
         nr_shards: ShardCount,
         shard_conns: &[Vec<Arc<Connection>>],

scylla/src/transport/iterator.rs

@@ -35,7 +35,7 @@ use crate::transport::connection::{Connection, NonErrorQueryResponse, QueryRespo
 use crate::transport::load_balancing::{self, RoutingInfo};
 use crate::transport::metrics::Metrics;
 use crate::transport::retry_policy::{QueryInfo, RetryDecision, RetrySession};
-use crate::transport::{Node, NodeRef};
+use crate::transport::NodeRef;
 use tracing::{trace, trace_span, warn, Instrument};
 use uuid::Uuid;
 
@@ -160,8 +160,6 @@ impl RowIterator {
         let worker_task = async move {
             let query_ref = &query;
 
-            let choose_connection = |node: Arc<Node>| async move { node.random_connection().await };
-
             let page_query = |connection: Arc<Connection>,
                               consistency: Consistency,
                               paging_state: Option<Bytes>| {
@@ -187,7 +185,6 @@ impl RowIterator {
 
             let worker = RowIteratorWorker {
                 sender: sender.into(),
-                choose_connection,
                 page_query,
                 statement_info: routing_info,
                 query_is_idempotent: query.config.is_idempotent,
@@ -259,13 +256,6 @@ impl RowIterator {
                 is_confirmed_lwt: config.prepared.is_confirmed_lwt(),
             };
 
-            let choose_connection = |node: Arc<Node>| async move {
-                match token {
-                    Some(token) => node.connection_for_token(token).await,
-                    None => node.random_connection().await,
-                }
-            };
-
             let page_query = |connection: Arc<Connection>,
                               consistency: Consistency,
                               paging_state: Option<Bytes>| async move {
@@ -290,7 +280,7 @@ impl RowIterator {
                         config
                             .cluster_data
                             .get_token_endpoints_iter(keyspace, token)
-                            .cloned()
+                            .map(|(node, shard)| (node.clone(), shard))
                             .collect(),
                     )
                 } else {
@@ -311,7 +301,6 @@ impl RowIterator {
 
             let worker = RowIteratorWorker {
                 sender: sender.into(),
-                choose_connection,
                 page_query,
                 statement_info,
                 query_is_idempotent: config.prepared.config.is_idempotent,
@@ -496,13 +485,9 @@ type PageSendAttemptedProof = SendAttemptedProof<Result<ReceivedPage, QueryError
 
 // RowIteratorWorker works in the background to fetch pages
 // RowIterator receives them through a channel
-struct RowIteratorWorker<'a, ConnFunc, QueryFunc, SpanCreatorFunc> {
+struct RowIteratorWorker<'a, QueryFunc, SpanCreatorFunc> {
     sender: ProvingSender<Result<ReceivedPage, QueryError>>,
 
-    // Closure used to choose a connection from a node
-    // AsyncFn(Arc<Node>) -> Result<Arc<Connection>, QueryError>
-    choose_connection: ConnFunc,
-
     // Closure used to perform a single page query
     // AsyncFn(Arc<Connection>, Option<Bytes>) -> Result<QueryResponse, QueryError>
     page_query: QueryFunc,
@@ -524,11 +509,8 @@ struct RowIteratorWorker<'a, ConnFunc, QueryFunc, SpanCreatorFunc> {
     span_creator: SpanCreatorFunc,
 }
 
-impl<ConnFunc, ConnFut, QueryFunc, QueryFut, SpanCreator>
-    RowIteratorWorker<'_, ConnFunc, QueryFunc, SpanCreator>
+impl<QueryFunc, QueryFut, SpanCreator> RowIteratorWorker<'_, QueryFunc, SpanCreator>
 where
-    ConnFunc: Fn(Arc<Node>) -> ConnFut,
-    ConnFut: Future<Output = Result<Arc<Connection>, QueryError>>,
     QueryFunc: Fn(Arc<Connection>, Consistency, Option<Bytes>) -> QueryFut,
     QueryFut: Future<Output = Result<QueryResponse, QueryError>>,
     SpanCreator: Fn() -> RequestSpan,
@@ -546,12 +528,13 @@ where
 
         self.log_query_start();
 
-        'nodes_in_plan: for node in query_plan {
+        'nodes_in_plan: for (node, shard) in query_plan {
             let span =
                 trace_span!(parent: &self.parent_span, "Executing query", node = %node.address);
             // For each node in the plan choose a connection to use
             // This connection will be reused for same node retries to preserve paging cache on the shard
-            let connection: Arc<Connection> = match (self.choose_connection)(node.clone())
+            let connection: Arc<Connection> = match node
+                .connection_for_shard(shard)
                 .instrument(span.clone())
                 .await
             {