1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527
// Copyright Materialize, Inc. and contributors. All rights reserved.
//
// Use of this software is governed by the Business Source License
// included in the LICENSE file.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0.
//! In-process caches of [Blob].
use std::sync::{Arc, Mutex};
use async_trait::async_trait;
use bytes::Bytes;
use mz_dyncfg::{Config, ConfigSet};
use mz_ore::bytes::SegmentedBytes;
use mz_ore::cast::CastFrom;
use mz_persist::location::{Blob, BlobMetadata, ExternalError};
use crate::cfg::PersistConfig;
use crate::internal::metrics::Metrics;
// In-memory cache for [Blob].
#[derive(Debug)]
pub struct BlobMemCache {
cfg: Arc<ConfigSet>,
metrics: Arc<Metrics>,
cache: Mutex<lru::Lru<String, SegmentedBytes>>,
blob: Arc<dyn Blob>,
}
pub(crate) const BLOB_CACHE_MEM_LIMIT_BYTES: Config<usize> = Config::new(
"persist_blob_cache_mem_limit_bytes",
// This initial value was tuned via a one-time experiment that showed an
// environment running our demo "auction" source + mv got 90%+ cache hits
// with a 1 MiB cache. This doesn't scale up to prod data sizes and doesn't
// help with multi-process replicas, but the memory usage seems
// unobjectionable enough to have it for the cases that it does help.
1024 * 1024,
"Capacity of in-mem blob cache in bytes (Materialize).",
);
impl BlobMemCache {
pub fn new(cfg: &PersistConfig, metrics: Arc<Metrics>, blob: Arc<dyn Blob>) -> Arc<dyn Blob> {
let eviction_metrics = Arc::clone(&metrics);
let cache = lru::Lru::new(BLOB_CACHE_MEM_LIMIT_BYTES.get(cfg), move |_, _, _| {
eviction_metrics.blob_cache_mem.evictions.inc()
});
let blob = BlobMemCache {
cfg: Arc::clone(&cfg.configs),
metrics,
cache: Mutex::new(cache),
blob,
};
Arc::new(blob)
}
fn resize_and_update_size_metrics(&self, cache: &mut lru::Lru<String, SegmentedBytes>) {
cache.update_capacity(BLOB_CACHE_MEM_LIMIT_BYTES.get(&self.cfg));
self.metrics
.blob_cache_mem
.size_blobs
.set(u64::cast_from(cache.entry_count()));
self.metrics
.blob_cache_mem
.size_bytes
.set(u64::cast_from(cache.entry_weight()));
}
}
#[async_trait]
impl Blob for BlobMemCache {
async fn get(&self, key: &str) -> Result<Option<SegmentedBytes>, ExternalError> {
// First check if the blob is in the cache. If it is, return it. If not,
// fetch it and put it in the cache.
//
// Blobs are write-once modify-never, so we don't have to worry about
// any races or cache invalidations here. If the value is in the cache,
// any value in S3 is guaranteed to match (if not, then there's a
// horrible bug somewhere else).
if let Some((_, cached_value)) = self.cache.lock().expect("lock poisoned").get(key) {
self.metrics.blob_cache_mem.hits_blobs.inc();
self.metrics
.blob_cache_mem
.hits_bytes
.inc_by(u64::cast_from(cached_value.len()));
return Ok(Some(cached_value.clone()));
}
let res = self.blob.get(key).await?;
if let Some(blob) = res.as_ref() {
// TODO: It would likely be useful to allow a caller to opt out of
// adding the data to the cache (e.g. compaction inputs, perhaps
// some read handles).
let mut cache = self.cache.lock().expect("lock poisoned");
// If the weight of this single blob is greater than the capacity of
// the cache, it will push out everything in the cache and then
// immediately get evicted itself. So, skip adding it in that case.
if blob.len() <= cache.capacity() {
cache.insert(key.to_owned(), blob.clone(), blob.len());
self.resize_and_update_size_metrics(&mut cache);
}
}
Ok(res)
}
async fn list_keys_and_metadata(
&self,
key_prefix: &str,
f: &mut (dyn FnMut(BlobMetadata) + Send + Sync),
) -> Result<(), ExternalError> {
self.blob.list_keys_and_metadata(key_prefix, f).await
}
async fn set(&self, key: &str, value: Bytes) -> Result<(), ExternalError> {
let () = self.blob.set(key, value.clone()).await?;
let weight = value.len();
let mut cache = self.cache.lock().expect("lock poisoned");
// If the weight of this single blob is greater than the capacity of
// the cache, it will push out everything in the cache and then
// immediately get evicted itself. So, skip adding it in that case.
if weight <= cache.capacity() {
cache.insert(key.to_owned(), SegmentedBytes::from(value), weight);
self.resize_and_update_size_metrics(&mut cache);
}
Ok(())
}
async fn delete(&self, key: &str) -> Result<Option<usize>, ExternalError> {
let res = self.blob.delete(key).await;
let mut cache = self.cache.lock().expect("lock poisoned");
cache.remove(key);
self.resize_and_update_size_metrics(&mut cache);
res
}
async fn restore(&self, key: &str) -> Result<(), ExternalError> {
self.blob.restore(key).await
}
}
mod lru {
use std::borrow::Borrow;
use std::collections::BTreeMap;
use std::hash::Hash;
use mz_ore::collections::HashMap;
#[derive(Debug, PartialEq, Eq, PartialOrd, Ord)]
pub struct Weight(usize);
#[derive(Debug, Default, Clone, PartialEq, Eq, PartialOrd, Ord)]
pub struct Time(usize);
/// A weighted cache, evicting the least recently used entries.
///
/// This is reimplemented here, instead of using an existing crate, because
/// existing options seem to either not support weights or they use unsafe.
pub struct Lru<K, V> {
evict_fn: Box<dyn Fn(K, V, usize) + Send>,
capacity: Weight,
next_time: Time,
entries: HashMap<K, (V, Weight, Time)>,
by_time: BTreeMap<Time, K>,
total_weight: Weight,
}
impl<K: Hash + Eq + Clone, V> Lru<K, V> {
/// Returns a new [Lru] with the requested configuration.
///
/// `evict_fn` is called for every entry evicted by the least recently
/// used policy. It is not called for entries replaced by the same key
/// in `insert` nor entries explictly removed by `remove`.
pub fn new<F>(capacity: usize, evict_fn: F) -> Self
where
F: Fn(K, V, usize) + Send + 'static,
{
Lru {
evict_fn: Box::new(evict_fn),
capacity: Weight(capacity),
next_time: Time::default(),
entries: HashMap::new(),
by_time: BTreeMap::new(),
total_weight: Weight(0),
}
}
/// Returns the capacity of the cache.
pub fn capacity(&self) -> usize {
self.capacity.0
}
/// Returns the total number of entries in the cache.
pub fn entry_count(&self) -> usize {
debug_assert_eq!(self.entries.len(), self.by_time.len());
self.entries.len()
}
/// Returns the sum of weights of entries in the cache.
pub fn entry_weight(&self) -> usize {
self.total_weight.0
}
/// Changes the weighted capacity of the cache, evicting as necessary if
/// the new value is smaller.
pub fn update_capacity(&mut self, capacity: usize) {
self.capacity = Weight(capacity);
self.resize();
assert!(self.total_weight <= self.capacity);
// Intentionally not run with debug_assert, because validate is
// `O(n)` in the size of the cache.
#[cfg(test)]
self.validate();
}
/// Returns a reference to entry with the given key, if present, marking
/// it as most recently used.
pub fn get<Q>(&mut self, key: &Q) -> Option<(&K, &V)>
where
K: Borrow<Q>,
Q: Hash + Eq + ?Sized,
{
{
let (key, val, weight) = self.remove(key)?;
self.insert_not_exists(key, val, Weight(weight));
}
let (key, (val, _, _)) = self
.entries
.get_key_value(key)
.expect("internal lru invariant violated");
// Intentionally not run with debug_assert, because validate is
// `O(n)` in the size of the cache.
#[cfg(test)]
self.validate();
Some((key, val))
}
/// Inserts the given key and value into the cache, marking it as most
/// recently used.
///
/// If the key already exists in the cache, the existing value and
/// weight are first removed.
pub fn insert(&mut self, key: K, val: V, weight: usize) {
let _ = self.remove(&key);
self.insert_not_exists(key, val, Weight(weight));
// Intentionally not run with debug_assert, because validate is
// `O(n)` in the size of the cache.
#[cfg(test)]
self.validate();
}
/// Removes the entry with the given key from the cache, if present.
///
/// Returns None if the entry was not in the cache.
pub fn remove<Q>(&mut self, k: &Q) -> Option<(K, V, usize)>
where
K: Borrow<Q>,
Q: Hash + Eq + ?Sized,
{
let (_, _, time) = self.entries.get(k)?;
let (key, val, weight) = self.remove_exists(time.clone());
// Intentionally not run with debug_assert, because validate is
// `O(n)` in the size of the cache.
#[cfg(test)]
self.validate();
Some((key, val, weight.0))
}
/// Returns an iterator over the entries in the cache in order from most
/// recently used to least.
#[allow(dead_code)]
pub(crate) fn iter(&self) -> impl Iterator<Item = (&K, &V, usize)> {
self.by_time.iter().rev().map(|(_, key)| {
let (val, _, weight) = self
.entries
.get(key)
.expect("internal lru invariant violated");
(key, val, weight.0)
})
}
fn insert_not_exists(&mut self, key: K, val: V, weight: Weight) {
let time = self.next_time.clone();
self.next_time.0 += 1;
self.total_weight.0 = self
.total_weight
.0
.checked_add(weight.0)
.expect("weight overflow");
assert!(self
.entries
.insert(key.clone(), (val, weight, time.clone()))
.is_none());
assert!(self.by_time.insert(time, key).is_none());
self.resize();
}
fn remove_exists(&mut self, time: Time) -> (K, V, Weight) {
let key = self
.by_time
.remove(&time)
.expect("internal list invariant violated");
let (val, weight, _time) = self
.entries
.remove(&key)
.expect("internal list invariant violated");
self.total_weight.0 = self
.total_weight
.0
.checked_sub(weight.0)
.expect("internal lru invariant violated");
(key, val, weight)
}
fn resize(&mut self) {
while self.total_weight > self.capacity {
let (time, _) = self
.by_time
.first_key_value()
.expect("internal lru invariant violated");
let (key, val, weight) = self.remove_exists(time.clone());
(self.evict_fn)(key, val, weight.0);
}
}
/// Checks internal invariants.
///
/// TODO: Give this persist's usual `-> Result<(), String>` signature
/// instead of panic-ing.
#[cfg(test)]
pub(crate) fn validate(&self) {
assert!(self.total_weight <= self.capacity);
let mut count = 0;
let mut weight = 0;
for (time, key) in self.by_time.iter() {
let (_val, w, t) = self
.entries
.get(key)
.expect("internal lru invariant violated");
count += 1;
weight += w.0;
assert_eq!(time, t);
}
assert_eq!(count, self.by_time.len());
assert_eq!(weight, self.total_weight.0);
}
}
impl<K: std::fmt::Debug, V> std::fmt::Debug for Lru<K, V> {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
let Lru {
evict_fn: _,
capacity,
next_time,
entries: _,
by_time,
total_weight,
} = self;
f.debug_struct("Lru")
.field("capacity", &capacity)
.field("total_weight", &total_weight)
.field("next_time", &next_time)
.field("by_time", &by_time)
.finish_non_exhaustive()
}
}
}
#[cfg(test)]
mod tests {
use mz_ore::assert_none;
use proptest::arbitrary::any;
use proptest::proptest;
use proptest_derive::Arbitrary;
use super::lru::*;
#[derive(Debug, Arbitrary)]
enum LruOp {
Get { key: u8 },
Insert { key: u8, weight: u8 },
Remove { key: u8 },
}
fn prop_testcase(ops: Vec<LruOp>) {
// In the long run, we'd expect maybe 1/2s of the `u8::MAX` possible
// keys to be present and for the average weight to be `u8::MAX / 2`.
// Select a capacity that is somewhat less than that.
let capacity = usize::from(u8::MAX / 2) * usize::from(u8::MAX / 2) / 2;
let mut cache = Lru::new(capacity, |_, _, _| {});
for op in ops {
match op {
LruOp::Get { key } => {
let _ = cache.get(&key);
}
LruOp::Insert { key, weight } => {
cache.insert(key, (), usize::from(weight));
}
LruOp::Remove { key } => {
let _ = cache.remove(&key);
}
}
cache.validate();
}
}
#[mz_ore::test]
#[cfg_attr(miri, ignore)] // too slow
fn lru_cache_prop() {
proptest!(|(state in proptest::collection::vec(any::<LruOp>(), 0..100))| prop_testcase(state));
}
impl Lru<&'static str, ()> {
fn keys(&self) -> Vec<&'static str> {
self.iter().map(|(k, _, _)| *k).collect()
}
}
#[mz_ore::test]
#[cfg_attr(miri, ignore)]
fn lru_cache_usage() {
let mut cache = Lru::<&'static str, ()>::new(3, |_, _, _| {});
// Empty
assert_eq!(cache.entry_count(), 0);
assert_eq!(cache.entry_weight(), 0);
// Insert into empty.
cache.insert("a", (), 2);
assert_eq!(cache.entry_count(), 1);
assert_eq!(cache.entry_weight(), 2);
assert_eq!(cache.keys(), &["a"]);
// Insert and push out previous.
cache.insert("b", (), 2);
assert_eq!(cache.entry_count(), 1);
assert_eq!(cache.entry_weight(), 2);
assert_eq!(cache.keys(), &["b"]);
// Insert and don't push out previous.
cache.insert("c", (), 1);
assert_eq!(cache.entry_count(), 2);
assert_eq!(cache.entry_weight(), 3);
assert_eq!(cache.keys(), &["c", "b"]);
// More than two elements.
cache.insert("d", (), 1);
cache.insert("e", (), 1);
assert_eq!(cache.entry_count(), 3);
assert_eq!(cache.entry_weight(), 3);
assert_eq!(cache.keys(), &["e", "d", "c"]);
// Get the head.
cache.get("e");
assert_eq!(cache.entry_count(), 3);
assert_eq!(cache.entry_weight(), 3);
assert_eq!(cache.keys(), &["e", "d", "c"]);
// Get the tail.
cache.get("c");
assert_eq!(cache.entry_count(), 3);
assert_eq!(cache.entry_weight(), 3);
assert_eq!(cache.keys(), &["c", "e", "d"]);
// Get the mid.
cache.get("e");
assert_eq!(cache.entry_count(), 3);
assert_eq!(cache.entry_weight(), 3);
assert_eq!(cache.keys(), &["e", "c", "d"]);
// Get a non-existent element.
cache.get("f");
assert_eq!(cache.entry_count(), 3);
assert_eq!(cache.entry_weight(), 3);
assert_eq!(cache.keys(), &["e", "c", "d"]);
// Remove an element.
assert!(cache.remove("c").is_some());
assert_eq!(cache.entry_count(), 2);
assert_eq!(cache.entry_weight(), 2);
assert_eq!(cache.keys(), &["e", "d"]);
// Remove a non-existent element.
assert_none!(cache.remove("f"));
assert_eq!(cache.entry_count(), 2);
assert_eq!(cache.entry_weight(), 2);
assert_eq!(cache.keys(), &["e", "d"]);
// Push out everything with a big weight
cache.insert("f", (), 3);
assert_eq!(cache.entry_count(), 1);
assert_eq!(cache.entry_weight(), 3);
assert_eq!(cache.keys(), &["f"]);
// Push out everything with a weight so big it doesn't even fit. (Is
// this even the behavior we want?)
cache.insert("g", (), 4);
assert_eq!(cache.entry_count(), 0);
assert_eq!(cache.entry_weight(), 0);
// Resize up
cache.insert("h", (), 2);
cache.insert("i", (), 1);
cache.update_capacity(4);
cache.insert("j", (), 1);
assert_eq!(cache.entry_count(), 3);
assert_eq!(cache.entry_weight(), 4);
assert_eq!(cache.keys(), &["j", "i", "h"]);
// Resize down
cache.update_capacity(2);
assert_eq!(cache.entry_count(), 2);
assert_eq!(cache.entry_weight(), 2);
assert_eq!(cache.keys(), &["j", "i"]);
}
}