|
| 1 | +use std::sync::Arc; |
| 2 | + |
| 3 | +use crate::utils::test_with_3_node_cluster; |
| 4 | +use crate::utils::{setup_tracing, unique_keyspace_name, PerformDDL}; |
| 5 | +use scylla::batch::Batch; |
| 6 | +use scylla::batch::BatchType; |
| 7 | +use scylla::client::caching_session::CachingSession; |
| 8 | +use scylla_proxy::RequestOpcode; |
| 9 | +use scylla_proxy::RequestReaction; |
| 10 | +use scylla_proxy::RequestRule; |
| 11 | +use scylla_proxy::ShardAwareness; |
| 12 | +use scylla_proxy::{Condition, ProxyError, Reaction, RequestFrame, TargetShard, WorkerError}; |
| 13 | +use tokio::sync::mpsc; |
| 14 | + |
| 15 | +fn consume_current_feedbacks( |
| 16 | + rx: &mut mpsc::UnboundedReceiver<(RequestFrame, Option<TargetShard>)>, |
| 17 | +) -> usize { |
| 18 | + std::iter::from_fn(|| rx.try_recv().ok()).count() |
| 19 | +} |
| 20 | + |
| 21 | +#[tokio::test] |
| 22 | +#[cfg(not(scylla_cloud_tests))] |
| 23 | +async fn ensure_cache_is_used() { |
| 24 | + use scylla::client::execution_profile::ExecutionProfile; |
| 25 | + |
| 26 | + use crate::utils::SingleTargetLBP; |
| 27 | + |
| 28 | + setup_tracing(); |
| 29 | + let res = test_with_3_node_cluster( |
| 30 | + ShardAwareness::QueryNode, |
| 31 | + |proxy_uris, translation_map, mut running_proxy| async move { |
| 32 | + let session = scylla::client::session_builder::SessionBuilder::new() |
| 33 | + .known_node(proxy_uris[0].as_str()) |
| 34 | + .address_translator(Arc::new(translation_map)) |
| 35 | + .build() |
| 36 | + .await |
| 37 | + .unwrap(); |
| 38 | + |
| 39 | + let cluster_size: usize = 3; |
| 40 | + let (feedback_txs, mut feedback_rxs): (Vec<_>, Vec<_>) = (0..cluster_size) |
| 41 | + .map(|_| mpsc::unbounded_channel::<(RequestFrame, Option<TargetShard>)>()) |
| 42 | + .unzip(); |
| 43 | + for (i, tx) in feedback_txs.iter().cloned().enumerate() { |
| 44 | + running_proxy.running_nodes[i].change_request_rules(Some(vec![RequestRule( |
| 45 | + Condition::and( |
| 46 | + Condition::RequestOpcode(RequestOpcode::Prepare), |
| 47 | + Condition::not(Condition::ConnectionRegisteredAnyEvent), |
| 48 | + ), |
| 49 | + RequestReaction::noop().with_feedback_when_performed(tx), |
| 50 | + )])); |
| 51 | + } |
| 52 | + |
| 53 | + let ks = unique_keyspace_name(); |
| 54 | + let rs = "{'class' : 'NetworkTopologyStrategy', 'replication_factor' : 1}"; |
| 55 | + session |
| 56 | + .ddl(format!( |
| 57 | + "CREATE KEYSPACE IF NOT EXISTS {} WITH REPLICATION = {}", |
| 58 | + ks, rs |
| 59 | + )) |
| 60 | + .await |
| 61 | + .unwrap(); |
| 62 | + session.use_keyspace(ks, false).await.unwrap(); |
| 63 | + session |
| 64 | + .ddl("CREATE TABLE IF NOT EXISTS tab (a int, b int, c int, primary key (a, b, c))") |
| 65 | + .await |
| 66 | + .unwrap(); |
| 67 | + // Assumption: all nodes have the same number of shards |
| 68 | + let nr_shards = session |
| 69 | + .get_cluster_state() |
| 70 | + .get_nodes_info() |
| 71 | + .first() |
| 72 | + .expect("No nodes information available") |
| 73 | + .sharder() |
| 74 | + .map(|sharder| sharder.nr_shards.get() as usize) |
| 75 | + .unwrap_or_else(|| 1); // If there is no sharder, assume 1 shard. |
| 76 | + |
| 77 | + // Consume all feedbacks so far to ensure we will not count something unrelated. |
| 78 | + let _feedbacks = feedback_rxs |
| 79 | + .iter_mut() |
| 80 | + .map(consume_current_feedbacks) |
| 81 | + .sum::<usize>(); |
| 82 | + |
| 83 | + let caching_session: CachingSession = CachingSession::from(session, 100); |
| 84 | + |
| 85 | + let batch_size: usize = 4; |
| 86 | + let mut batch = Batch::new(BatchType::Logged); |
| 87 | + for i in 1..=batch_size { |
| 88 | + let insert_b_c = format!("INSERT INTO tab (a, b, c) VALUES ({}, ?, ?)", i); |
| 89 | + batch.append_statement(insert_b_c.as_str()); |
| 90 | + } |
| 91 | + let batch_values: Vec<(i32, i32)> = (1..=batch_size as i32).map(|i| (i, i)).collect(); |
| 92 | + |
| 93 | + // First batch that should generate prepares for each shard. |
| 94 | + caching_session |
| 95 | + .batch(&batch, batch_values.clone()) |
| 96 | + .await |
| 97 | + .unwrap(); |
| 98 | + let feedbacks: usize = feedback_rxs.iter_mut().map(consume_current_feedbacks).sum(); |
| 99 | + assert_eq!(feedbacks, batch_size * nr_shards * cluster_size); |
| 100 | + |
| 101 | + // Few extra runs. Those batches should not result in any prepares being sent. |
| 102 | + for _ in 0..4 { |
| 103 | + caching_session |
| 104 | + .batch(&batch, batch_values.clone()) |
| 105 | + .await |
| 106 | + .unwrap(); |
| 107 | + let feedbacks: usize = feedback_rxs.iter_mut().map(consume_current_feedbacks).sum(); |
| 108 | + assert_eq!(feedbacks, 0); |
| 109 | + } |
| 110 | + |
| 111 | + let prepared_batch_res_rows: Vec<(i32, i32, i32)> = caching_session |
| 112 | + .execute_unpaged("SELECT * FROM tab", &[]) |
| 113 | + .await |
| 114 | + .unwrap() |
| 115 | + .into_rows_result() |
| 116 | + .unwrap() |
| 117 | + .rows() |
| 118 | + .unwrap() |
| 119 | + .collect::<Result<_, _>>() |
| 120 | + .unwrap(); |
| 121 | + |
| 122 | + // Select should have been prepared on all shards |
| 123 | + let feedbacks: usize = feedback_rxs.iter_mut().map(consume_current_feedbacks).sum(); |
| 124 | + assert_eq!(feedbacks, nr_shards * cluster_size); |
| 125 | + |
| 126 | + // Verify the data from inserts |
| 127 | + let mut prepared_batch_res_rows = prepared_batch_res_rows; |
| 128 | + prepared_batch_res_rows.sort(); |
| 129 | + let expected_rows: Vec<(i32, i32, i32)> = |
| 130 | + (1..=batch_size as i32).map(|i| (i, i, i)).collect(); |
| 131 | + assert_eq!(prepared_batch_res_rows, expected_rows); |
| 132 | + |
| 133 | + // Run some alters to invalidate the server side cache, similarly to scylla/src/session_test.rs |
| 134 | + caching_session |
| 135 | + .ddl("ALTER TABLE tab RENAME c to tmp") |
| 136 | + .await |
| 137 | + .unwrap(); |
| 138 | + caching_session |
| 139 | + .ddl("ALTER TABLE tab RENAME b to c") |
| 140 | + .await |
| 141 | + .unwrap(); |
| 142 | + caching_session |
| 143 | + .ddl("ALTER TABLE tab RENAME tmp to b") |
| 144 | + .await |
| 145 | + .unwrap(); |
| 146 | + |
| 147 | + // execute_unpageds caused by alters likely resulted in some prepares being sent. |
| 148 | + // Consume those frames. |
| 149 | + feedback_rxs |
| 150 | + .iter_mut() |
| 151 | + .map(consume_current_feedbacks) |
| 152 | + .sum::<usize>(); |
| 153 | + |
| 154 | + // Run batch for each shard. The server cache should be updated on the first mismatch, |
| 155 | + // therefore only first contacted shard will request reprepare due to mismatch. |
| 156 | + for node_info in caching_session |
| 157 | + .get_session() |
| 158 | + .get_cluster_state() |
| 159 | + .get_nodes_info() |
| 160 | + .iter() |
| 161 | + { |
| 162 | + for shard_id in 0..nr_shards { |
| 163 | + let policy = SingleTargetLBP { |
| 164 | + target: (node_info.clone(), Some(shard_id as u32)), |
| 165 | + }; |
| 166 | + let execution_profile = ExecutionProfile::builder() |
| 167 | + .load_balancing_policy(Arc::new(policy)) |
| 168 | + .build(); |
| 169 | + batch.set_execution_profile_handle(Some(execution_profile.into_handle())); |
| 170 | + caching_session |
| 171 | + .batch(&batch, batch_values.clone()) |
| 172 | + .await |
| 173 | + .unwrap(); |
| 174 | + let feedbacks: usize = |
| 175 | + feedback_rxs.iter_mut().map(consume_current_feedbacks).sum(); |
| 176 | + let expected_feedbacks = if shard_id == 0 { batch_size } else { 0 }; |
| 177 | + assert_eq!( |
| 178 | + feedbacks, expected_feedbacks, |
| 179 | + "Mismatch in feedbacks on execution for node: {:?}, shard: {}", |
| 180 | + node_info, shard_id |
| 181 | + ); |
| 182 | + } |
| 183 | + } |
| 184 | + running_proxy |
| 185 | + }, |
| 186 | + ) |
| 187 | + .await; |
| 188 | + match res { |
| 189 | + Ok(()) => (), |
| 190 | + Err(ProxyError::Worker(WorkerError::DriverDisconnected(_))) => (), |
| 191 | + Err(err) => panic!("{}", err), |
| 192 | + } |
| 193 | +} |
0 commit comments