|
| 1 | +//! Parallel for each and map using tokio tasks. |
| 2 | +//! |
| 3 | +//! This avoid the problem of sleeping threads with mimalloc when using rayon in combination with |
| 4 | +//! tokio. It also avoid having multiple thread pools. |
| 5 | +//! see also https://pwy.io/posts/mimalloc-cigarette/ |
| 6 | +
|
| 7 | +use std::{sync::LazyLock, thread::available_parallelism}; |
| 8 | + |
| 9 | +use crate::{scope::scope_and_block, util::into_chunks}; |
| 10 | + |
| 11 | +/// Calculates a good chunk size for parallel processing based on the number of available threads. |
| 12 | +/// This is used to ensure that the workload is evenly distributed across the threads. |
| 13 | +fn good_chunk_size(len: usize) -> usize { |
| 14 | + static GOOD_CHUNK_COUNT: LazyLock<usize> = |
| 15 | + LazyLock::new(|| available_parallelism().map_or(16, |c| c.get() * 4)); |
| 16 | + let min_chunk_count = *GOOD_CHUNK_COUNT; |
| 17 | + len.div_ceil(min_chunk_count) |
| 18 | +} |
| 19 | + |
| 20 | +pub fn for_each<'l, T, F>(items: &'l [T], f: F) |
| 21 | +where |
| 22 | + T: Sync, |
| 23 | + F: Fn(&'l T) + Send + Sync, |
| 24 | +{ |
| 25 | + let len = items.len(); |
| 26 | + if len <= 1 { |
| 27 | + for item in items { |
| 28 | + f(item); |
| 29 | + } |
| 30 | + return; |
| 31 | + } |
| 32 | + let chunk_size = good_chunk_size(len); |
| 33 | + let f = &f; |
| 34 | + let _results = scope_and_block(len.div_ceil(chunk_size), |scope| { |
| 35 | + for chunk in items.chunks(chunk_size) { |
| 36 | + scope.spawn(async move { |
| 37 | + for item in chunk { |
| 38 | + f(item); |
| 39 | + } |
| 40 | + }) |
| 41 | + } |
| 42 | + }); |
| 43 | +} |
| 44 | + |
| 45 | +pub fn for_each_owned<T>(items: Vec<T>, f: impl Fn(T) + Send + Sync) |
| 46 | +where |
| 47 | + T: Send + Sync, |
| 48 | +{ |
| 49 | + let len = items.len(); |
| 50 | + if len <= 1 { |
| 51 | + for item in items { |
| 52 | + f(item); |
| 53 | + } |
| 54 | + return; |
| 55 | + } |
| 56 | + let chunk_size = good_chunk_size(len); |
| 57 | + let f = &f; |
| 58 | + let _results = scope_and_block(len.div_ceil(chunk_size), |scope| { |
| 59 | + for chunk in into_chunks(items, chunk_size) { |
| 60 | + scope.spawn(async move { |
| 61 | + // SAFETY: Even when f() panics we drop all items in the chunk. |
| 62 | + for item in chunk { |
| 63 | + f(item); |
| 64 | + } |
| 65 | + }) |
| 66 | + } |
| 67 | + }); |
| 68 | +} |
| 69 | + |
| 70 | +pub fn try_for_each<'l, T, E>( |
| 71 | + items: &'l [T], |
| 72 | + f: impl (Fn(&'l T) -> Result<(), E>) + Send + Sync, |
| 73 | +) -> Result<(), E> |
| 74 | +where |
| 75 | + T: Sync, |
| 76 | + E: Send + 'static, |
| 77 | +{ |
| 78 | + let len = items.len(); |
| 79 | + if len <= 1 { |
| 80 | + for item in items { |
| 81 | + f(item)?; |
| 82 | + } |
| 83 | + return Ok(()); |
| 84 | + } |
| 85 | + let chunk_size = good_chunk_size(len); |
| 86 | + let f = &f; |
| 87 | + scope_and_block(len.div_ceil(chunk_size), |scope| { |
| 88 | + for chunk in items.chunks(chunk_size) { |
| 89 | + scope.spawn(async move { |
| 90 | + for item in chunk { |
| 91 | + f(item)?; |
| 92 | + } |
| 93 | + Ok(()) |
| 94 | + }) |
| 95 | + } |
| 96 | + }) |
| 97 | + .collect::<Result<(), E>>() |
| 98 | +} |
| 99 | + |
| 100 | +pub fn try_for_each_mut<'l, T, E>( |
| 101 | + items: &'l mut [T], |
| 102 | + f: impl (Fn(&'l mut T) -> Result<(), E>) + Send + Sync, |
| 103 | +) -> Result<(), E> |
| 104 | +where |
| 105 | + T: Send + Sync, |
| 106 | + E: Send + 'static, |
| 107 | +{ |
| 108 | + let len = items.len(); |
| 109 | + if len <= 1 { |
| 110 | + for item in items { |
| 111 | + f(item)?; |
| 112 | + } |
| 113 | + return Ok(()); |
| 114 | + } |
| 115 | + let chunk_size = good_chunk_size(len); |
| 116 | + let f = &f; |
| 117 | + scope_and_block(len.div_ceil(chunk_size), |scope| { |
| 118 | + for chunk in items.chunks_mut(chunk_size) { |
| 119 | + scope.spawn(async move { |
| 120 | + for item in chunk { |
| 121 | + f(item)?; |
| 122 | + } |
| 123 | + Ok(()) |
| 124 | + }) |
| 125 | + } |
| 126 | + }) |
| 127 | + .collect::<Result<(), E>>() |
| 128 | +} |
| 129 | + |
| 130 | +pub fn try_for_each_owned<T, E>( |
| 131 | + items: Vec<T>, |
| 132 | + f: impl (Fn(T) -> Result<(), E>) + Send + Sync, |
| 133 | +) -> Result<(), E> |
| 134 | +where |
| 135 | + T: Send + Sync, |
| 136 | + E: Send + 'static, |
| 137 | +{ |
| 138 | + let len = items.len(); |
| 139 | + if len <= 1 { |
| 140 | + for item in items { |
| 141 | + f(item)?; |
| 142 | + } |
| 143 | + return Ok(()); |
| 144 | + } |
| 145 | + let chunk_size = good_chunk_size(len); |
| 146 | + let f = &f; |
| 147 | + scope_and_block(len.div_ceil(chunk_size), |scope| { |
| 148 | + for chunk in into_chunks(items, chunk_size) { |
| 149 | + scope.spawn(async move { |
| 150 | + for item in chunk { |
| 151 | + f(item)?; |
| 152 | + } |
| 153 | + Ok(()) |
| 154 | + }) |
| 155 | + } |
| 156 | + }) |
| 157 | + .collect::<Result<(), E>>() |
| 158 | +} |
| 159 | + |
| 160 | +pub fn map_collect<'l, Item, PerItemResult, Result>( |
| 161 | + items: &'l [Item], |
| 162 | + f: impl Fn(&'l Item) -> PerItemResult + Send + Sync, |
| 163 | +) -> Result |
| 164 | +where |
| 165 | + Item: Sync, |
| 166 | + PerItemResult: Send + Sync + 'l, |
| 167 | + Result: FromIterator<PerItemResult>, |
| 168 | +{ |
| 169 | + let len = items.len(); |
| 170 | + if len == 0 { |
| 171 | + return Result::from_iter(std::iter::empty()); // No items to process, return empty |
| 172 | + // collection |
| 173 | + } |
| 174 | + let chunk_size = good_chunk_size(len); |
| 175 | + let f = &f; |
| 176 | + scope_and_block(len.div_ceil(chunk_size), |scope| { |
| 177 | + for chunk in items.chunks(chunk_size) { |
| 178 | + scope.spawn(async move { chunk.iter().map(f).collect::<Vec<_>>() }) |
| 179 | + } |
| 180 | + }) |
| 181 | + .flatten() |
| 182 | + .collect() |
| 183 | +} |
| 184 | + |
| 185 | +pub fn map_collect_owned<'l, Item, PerItemResult, Result>( |
| 186 | + items: Vec<Item>, |
| 187 | + f: impl Fn(Item) -> PerItemResult + Send + Sync, |
| 188 | +) -> Result |
| 189 | +where |
| 190 | + Item: Send + Sync, |
| 191 | + PerItemResult: Send + Sync + 'l, |
| 192 | + Result: FromIterator<PerItemResult>, |
| 193 | +{ |
| 194 | + let len = items.len(); |
| 195 | + if len == 0 { |
| 196 | + return Result::from_iter(std::iter::empty()); // No items to process, return empty |
| 197 | + // collection; |
| 198 | + } |
| 199 | + let chunk_size = good_chunk_size(len); |
| 200 | + let f = &f; |
| 201 | + scope_and_block(len.div_ceil(chunk_size), |scope| { |
| 202 | + for chunk in into_chunks(items, chunk_size) { |
| 203 | + scope.spawn(async move { chunk.map(f).collect::<Vec<_>>() }) |
| 204 | + } |
| 205 | + }) |
| 206 | + .flatten() |
| 207 | + .collect() |
| 208 | +} |
| 209 | + |
| 210 | +#[cfg(test)] |
| 211 | +mod tests { |
| 212 | + use std::{ |
| 213 | + panic::{AssertUnwindSafe, catch_unwind}, |
| 214 | + sync::atomic::{AtomicI32, Ordering}, |
| 215 | + }; |
| 216 | + |
| 217 | + use super::*; |
| 218 | + |
| 219 | + #[tokio::test(flavor = "multi_thread")] |
| 220 | + async fn test_parallel_for_each() { |
| 221 | + let input = vec![1, 2, 3, 4, 5]; |
| 222 | + let sum = AtomicI32::new(0); |
| 223 | + for_each(&input, |&x| { |
| 224 | + sum.fetch_add(x, Ordering::SeqCst); |
| 225 | + }); |
| 226 | + assert_eq!(sum.load(Ordering::SeqCst), 15); |
| 227 | + } |
| 228 | + |
| 229 | + #[tokio::test(flavor = "multi_thread")] |
| 230 | + async fn test_parallel_try_for_each() { |
| 231 | + let input = vec![1, 2, 3, 4, 5]; |
| 232 | + let result = try_for_each(&input, |&x| { |
| 233 | + if x % 2 == 0 { |
| 234 | + Ok(()) |
| 235 | + } else { |
| 236 | + Err(format!("Odd number {x} encountered")) |
| 237 | + } |
| 238 | + }); |
| 239 | + assert!(result.is_err()); |
| 240 | + assert_eq!(result.unwrap_err(), "Odd number 1 encountered"); |
| 241 | + } |
| 242 | + |
| 243 | + #[tokio::test(flavor = "multi_thread")] |
| 244 | + async fn test_parallel_try_for_each_mut() { |
| 245 | + let mut input = vec![1, 2, 3, 4, 5]; |
| 246 | + let result = try_for_each_mut(&mut input, |x| { |
| 247 | + *x += 10; |
| 248 | + if *x % 2 == 0 { |
| 249 | + Ok(()) |
| 250 | + } else { |
| 251 | + Err(format!("Odd number {} encountered", *x)) |
| 252 | + } |
| 253 | + }); |
| 254 | + assert!(result.is_err()); |
| 255 | + assert_eq!(result.unwrap_err(), "Odd number 11 encountered"); |
| 256 | + assert_eq!(input, vec![11, 12, 13, 14, 15]); |
| 257 | + } |
| 258 | + |
| 259 | + #[tokio::test(flavor = "multi_thread")] |
| 260 | + async fn test_parallel_for_each_owned() { |
| 261 | + let input = vec![1, 2, 3, 4, 5]; |
| 262 | + let sum = AtomicI32::new(0); |
| 263 | + for_each_owned(input, |x| { |
| 264 | + sum.fetch_add(x, Ordering::SeqCst); |
| 265 | + }); |
| 266 | + assert_eq!(sum.load(Ordering::SeqCst), 15); |
| 267 | + } |
| 268 | + |
| 269 | + #[tokio::test(flavor = "multi_thread")] |
| 270 | + async fn test_parallel_map_collect() { |
| 271 | + let input = vec![1, 2, 3, 4, 5]; |
| 272 | + let result: Vec<_> = map_collect(&input, |&x| x * 2); |
| 273 | + assert_eq!(result, vec![2, 4, 6, 8, 10]); |
| 274 | + } |
| 275 | + |
| 276 | + #[tokio::test(flavor = "multi_thread")] |
| 277 | + async fn test_parallel_map_collect_owned() { |
| 278 | + let input = vec![1, 2, 3, 4, 5]; |
| 279 | + let result: Vec<_> = map_collect_owned(input, |x| x * 2); |
| 280 | + assert_eq!(result, vec![2, 4, 6, 8, 10]); |
| 281 | + } |
| 282 | + |
| 283 | + #[tokio::test(flavor = "multi_thread")] |
| 284 | + async fn test_parallel_map_collect_owned_many() { |
| 285 | + let input = vec![1; 1000]; |
| 286 | + let result: Vec<_> = map_collect_owned(input, |x| x * 2); |
| 287 | + assert_eq!(result, vec![2; 1000]); |
| 288 | + } |
| 289 | + |
| 290 | + #[tokio::test(flavor = "multi_thread")] |
| 291 | + async fn test_panic_in_scope() { |
| 292 | + let result = catch_unwind(AssertUnwindSafe(|| { |
| 293 | + let mut input = vec![1; 1000]; |
| 294 | + input[744] = 2; |
| 295 | + for_each(&input, |x| { |
| 296 | + if *x == 2 { |
| 297 | + panic!("Intentional panic"); |
| 298 | + } |
| 299 | + }); |
| 300 | + panic!("Should not get here") |
| 301 | + })); |
| 302 | + assert!(result.is_err()); |
| 303 | + assert_eq!( |
| 304 | + result.unwrap_err().downcast_ref::<&str>(), |
| 305 | + Some(&"Intentional panic") |
| 306 | + ); |
| 307 | + } |
| 308 | +} |
0 commit comments