|
| 1 | +// Copyright 2021 Datafuse Labs |
| 2 | +// |
| 3 | +// Licensed under the Apache License, Version 2.0 (the "License"); |
| 4 | +// you may not use this file except in compliance with the License. |
| 5 | +// You may obtain a copy of the License at |
| 6 | +// |
| 7 | +// http://www.apache.org/licenses/LICENSE-2.0 |
| 8 | +// |
| 9 | +// Unless required by applicable law or agreed to in writing, software |
| 10 | +// distributed under the License is distributed on an "AS IS" BASIS, |
| 11 | +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 12 | +// See the License for the specific language governing permissions and |
| 13 | +// limitations under the License. |
| 14 | + |
| 15 | +use std::collections::BTreeMap; |
| 16 | +use std::collections::HashMap; |
| 17 | +use std::io::Write; |
| 18 | + |
| 19 | +use databend_common_catalog::BasicColumnStatistics; |
| 20 | +use databend_common_catalog::TableStatistics; |
| 21 | +use databend_common_catalog::table_context::TableContextSettings; |
| 22 | +use databend_common_exception::Result; |
| 23 | +use databend_common_expression::stat_distribution::NdvEstimate; |
| 24 | +use databend_common_sql::optimizer::OptimizerContext; |
| 25 | +use databend_common_sql::optimizer::OptimizerTraceEvent; |
| 26 | +use databend_common_sql::optimizer::optimizers::CascadesOptimizer; |
| 27 | +use databend_common_sql::optimizer::optimizers::CollectStatisticsOptimizer; |
| 28 | +use databend_common_sql::optimizer::optimizers::DPhpyOptimizer; |
| 29 | +use databend_common_sql::optimizer::optimizers::operator::PullUpFilterOptimizer; |
| 30 | +use databend_common_sql::optimizer::optimizers::operator::RuleStatsAggregateOptimizer; |
| 31 | +use databend_common_sql::optimizer::optimizers::operator::SubqueryDecorrelatorOptimizer; |
| 32 | +use databend_common_sql::optimizer::optimizers::recursive::RecursiveRuleOptimizer; |
| 33 | +use databend_common_sql::optimizer::optimizers::rule::DEFAULT_REWRITE_RULES; |
| 34 | +use databend_common_sql::optimizer::optimizers::rule::RuleID; |
| 35 | +use databend_common_sql::optimizer::pipeline::OptimizerPipeline; |
| 36 | +use databend_common_sql::plans::Plan; |
| 37 | +use databend_common_statistics::Datum; |
| 38 | + |
| 39 | +use crate::framework::LiteTableContext; |
| 40 | +use crate::framework::golden::open_golden_file; |
| 41 | +use crate::framework::golden::write_case_title; |
| 42 | + |
| 43 | +struct JoinMemoCase<'a> { |
| 44 | + name: &'a str, |
| 45 | + description: &'a str, |
| 46 | + cluster_by: &'a str, |
| 47 | + sql: &'a str, |
| 48 | +} |
| 49 | + |
| 50 | +fn table_statistics(rows: u64) -> TableStatistics { |
| 51 | + TableStatistics { |
| 52 | + num_rows: Some(rows), |
| 53 | + data_size: Some(rows.saturating_mul(24)), |
| 54 | + data_size_compressed: None, |
| 55 | + index_size: None, |
| 56 | + bloom_index_size: None, |
| 57 | + ngram_index_size: None, |
| 58 | + inverted_index_size: None, |
| 59 | + vector_index_size: None, |
| 60 | + virtual_column_size: None, |
| 61 | + number_of_blocks: Some(1), |
| 62 | + number_of_segments: Some(1), |
| 63 | + } |
| 64 | +} |
| 65 | + |
| 66 | +fn column_statistics(rows: u64) -> HashMap<String, BasicColumnStatistics> { |
| 67 | + ["k1", "k2", "v"] |
| 68 | + .into_iter() |
| 69 | + .map(|column| { |
| 70 | + (column.to_string(), BasicColumnStatistics { |
| 71 | + min: Some(Datum::Int(0)), |
| 72 | + max: Some(Datum::Int(rows as i64)), |
| 73 | + ndv: Some(NdvEstimate::exact(rows as f64)), |
| 74 | + null_count: 0, |
| 75 | + in_memory_size: rows.saturating_mul(8), |
| 76 | + }) |
| 77 | + }) |
| 78 | + .collect() |
| 79 | +} |
| 80 | + |
| 81 | +#[tokio::test(flavor = "multi_thread", worker_threads = 1)] |
| 82 | +async fn test_cluster_key_order_join_memo_golden() -> Result<()> { |
| 83 | + let mut file = open_golden_file("optimizer", "cluster_key_join_order.txt")?; |
| 84 | + |
| 85 | + for case in [ |
| 86 | + JoinMemoCase { |
| 87 | + name: "k1_k2_prefix", |
| 88 | + description: "Full memo output when the clustered probe can first match a.k1.", |
| 89 | + cluster_by: "CLUSTER BY (k1, k2)", |
| 90 | + sql: " |
| 91 | + SELECT * |
| 92 | + FROM a |
| 93 | + JOIN b ON a.k1 = b.k1 |
| 94 | + JOIN c ON a.k2 = c.k2 |
| 95 | + ", |
| 96 | + }, |
| 97 | + JoinMemoCase { |
| 98 | + name: "k2_k1_prefix", |
| 99 | + description: "Full memo output when the clustered probe can first match a.k2.", |
| 100 | + cluster_by: "CLUSTER BY (k2, k1)", |
| 101 | + sql: " |
| 102 | + SELECT * |
| 103 | + FROM a |
| 104 | + JOIN b ON a.k1 = b.k1 |
| 105 | + JOIN c ON a.k2 = c.k2 |
| 106 | + ", |
| 107 | + }, |
| 108 | + JoinMemoCase { |
| 109 | + name: "filter_preserves_cluster_keys", |
| 110 | + description: "Cluster keys still affect join order after a filter on the clustered table.", |
| 111 | + cluster_by: "CLUSTER BY (k1, k2)", |
| 112 | + sql: " |
| 113 | + SELECT * |
| 114 | + FROM (SELECT * FROM a WHERE v >= 0) a |
| 115 | + JOIN b ON a.k1 = b.k1 |
| 116 | + JOIN c ON a.k2 = c.k2 |
| 117 | + ", |
| 118 | + }, |
| 119 | + JoinMemoCase { |
| 120 | + name: "limit_and_join_preserve_cluster_keys", |
| 121 | + description: "Cluster keys still affect join order after a limit subquery and a partial join.", |
| 122 | + cluster_by: "CLUSTER BY (k1, k2)", |
| 123 | + sql: " |
| 124 | + SELECT * |
| 125 | + FROM (SELECT * FROM a LIMIT 1000) a |
| 126 | + JOIN b ON a.k1 = b.k1 |
| 127 | + JOIN c ON a.k2 = c.k2 |
| 128 | + ", |
| 129 | + }, |
| 130 | + ] { |
| 131 | + write_cluster_key_join_order_memo(&mut file, case).await?; |
| 132 | + } |
| 133 | + |
| 134 | + Ok(()) |
| 135 | +} |
| 136 | + |
| 137 | +async fn write_cluster_key_join_order_memo( |
| 138 | + file: &mut impl Write, |
| 139 | + case: JoinMemoCase<'_>, |
| 140 | +) -> Result<()> { |
| 141 | + let ctx = LiteTableContext::create().await?; |
| 142 | + ctx.configure_for_optimizer_case(true)?; |
| 143 | + ctx.set_cluster_node_num(1); |
| 144 | + |
| 145 | + write_case_title(file, case.name, case.description)?; |
| 146 | + for table in ["a", "b", "c"] { |
| 147 | + let table_cluster_by = if table == "a" { case.cluster_by } else { "" }; |
| 148 | + let setup_sql = match table_cluster_by { |
| 149 | + "" => format!("CREATE TABLE {table}(k1 BIGINT, k2 BIGINT, v BIGINT)"), |
| 150 | + _ => { |
| 151 | + format!("CREATE TABLE {table}(k1 BIGINT, k2 BIGINT, v BIGINT) {table_cluster_by}") |
| 152 | + } |
| 153 | + }; |
| 154 | + writeln!(file, "setup: {setup_sql}")?; |
| 155 | + ctx.register_table_sql_with_stats( |
| 156 | + &setup_sql, |
| 157 | + Some(table_statistics(1000)), |
| 158 | + column_statistics(1000), |
| 159 | + ) |
| 160 | + .await?; |
| 161 | + } |
| 162 | + |
| 163 | + let sql = case |
| 164 | + .sql |
| 165 | + .lines() |
| 166 | + .map(str::trim) |
| 167 | + .filter(|line| !line.is_empty()) |
| 168 | + .collect::<Vec<_>>() |
| 169 | + .join("\n"); |
| 170 | + writeln!(file, "sql: {sql}")?; |
| 171 | + writeln!(file, "memo:")?; |
| 172 | + writeln!(file, "{}", explain_memo(&ctx, &sql).await?)?; |
| 173 | + writeln!(file)?; |
| 174 | + Ok(()) |
| 175 | +} |
| 176 | + |
| 177 | +async fn explain_memo(ctx: &std::sync::Arc<LiteTableContext>, sql: &str) -> Result<String> { |
| 178 | + let Plan::Query { |
| 179 | + s_expr, metadata, .. |
| 180 | + } = ctx.bind_sql(sql).await? |
| 181 | + else { |
| 182 | + unreachable!("SELECT should bind to a query plan"); |
| 183 | + }; |
| 184 | + |
| 185 | + let settings = ctx.get_settings(); |
| 186 | + let opt_ctx = OptimizerContext::new(ctx.clone(), metadata) |
| 187 | + .with_settings(&settings)? |
| 188 | + .set_enable_distributed_optimization(true) |
| 189 | + .clone(); |
| 190 | + opt_ctx.set_flag("explain_memo", true); |
| 191 | + opt_ctx.clear_optimizer_trace(); |
| 192 | + |
| 193 | + let mut pipeline = OptimizerPipeline::new(opt_ctx.clone(), *s_expr) |
| 194 | + .await? |
| 195 | + .add(SubqueryDecorrelatorOptimizer::new(opt_ctx.clone(), None)) |
| 196 | + .add(RuleStatsAggregateOptimizer::new(opt_ctx.clone())) |
| 197 | + .add(CollectStatisticsOptimizer::new(opt_ctx.clone())) |
| 198 | + .add(PullUpFilterOptimizer::new(opt_ctx.clone())) |
| 199 | + .add(RecursiveRuleOptimizer::new( |
| 200 | + opt_ctx.clone(), |
| 201 | + &DEFAULT_REWRITE_RULES, |
| 202 | + )) |
| 203 | + .add(RecursiveRuleOptimizer::new(opt_ctx.clone(), &[ |
| 204 | + RuleID::SplitAggregate, |
| 205 | + ])) |
| 206 | + .add(DPhpyOptimizer::new(opt_ctx.clone())) |
| 207 | + .add(CascadesOptimizer::new(opt_ctx.clone())?); |
| 208 | + |
| 209 | + let _s_expr = pipeline.execute().await?; |
| 210 | + |
| 211 | + let mut sections = Vec::new(); |
| 212 | + sections.extend(format_optimizer_trace(opt_ctx.take_optimizer_trace())); |
| 213 | + sections.push(pipeline.memo().display()?); |
| 214 | + Ok(sections.join("\n\n")) |
| 215 | +} |
| 216 | + |
| 217 | +fn format_optimizer_trace(traces: Vec<OptimizerTraceEvent>) -> Vec<String> { |
| 218 | + let mut groups = BTreeMap::<String, BTreeMap<String, Vec<String>>>::new(); |
| 219 | + for trace in traces { |
| 220 | + groups |
| 221 | + .entry(trace.optimizer) |
| 222 | + .or_default() |
| 223 | + .entry(trace.event) |
| 224 | + .or_default() |
| 225 | + .push(trace.detail); |
| 226 | + } |
| 227 | + |
| 228 | + groups |
| 229 | + .into_iter() |
| 230 | + .map(|(optimizer, events)| { |
| 231 | + let mut lines = vec![format!("{optimizer}:")]; |
| 232 | + for (event, details) in events { |
| 233 | + lines.push(format!("{event}:")); |
| 234 | + lines.extend(details.into_iter().map(|detail| format!("- {detail}"))); |
| 235 | + } |
| 236 | + lines.join("\n") |
| 237 | + }) |
| 238 | + .collect() |
| 239 | +} |
0 commit comments