| 1 |
// Copyright ©2015 The bíogo Authors. All rights reserved. |
| 2 |
// Use of this source code is governed by a BSD-style |
| 3 |
// license that can be found in the LICENSE file. |
| 4 |
|
| 5 |
// Package cache provides basic block cache types for the bgzf package. |
| 6 |
package cache |
| 7 |
|
| 8 |
import ( |
| 9 |
"sync" |
| 10 |
|
| 11 |
"github.com/biogo/hts/bgzf" |
| 12 |
) |
| 13 |
|
| 14 |
var ( |
| 15 |
_ Cache = (*LRU)(nil) |
| 16 |
_ Cache = (*FIFO)(nil) |
| 17 |
_ Cache = (*Random)(nil) |
| 18 |
) |
| 19 |
|
| 20 |
// Free attempts to drop as many blocks from c as needed allow |
| 21 |
// n successful Put calls on c. It returns a boolean indicating |
| 22 |
// whether n slots were made available. |
| 23 |
func Free(n int, c Cache) bool { |
| 24 |
empty := c.Cap() - c.Len() |
| 25 |
if n <= empty { |
| 26 |
return true |
| 27 |
} |
| 28 |
c.Drop(n - empty) |
| 29 |
return c.Cap()-c.Len() >= n |
| 30 |
} |
| 31 |
|
| 32 |
// Cache is an extension of bgzf.Cache that allows inspection |
| 33 |
// and manipulation of the cache. |
| 34 |
type Cache interface { |
| 35 |
bgzf.Cache |
| 36 |
|
| 37 |
// Len returns the number of elements held by |
| 38 |
// the cache. |
| 39 |
Len() int |
| 40 |
|
| 41 |
// Cap returns the maximum number of elements |
| 42 |
// that can be held by the cache. |
| 43 |
Cap() int |
| 44 |
|
| 45 |
// Resize changes the capacity of the cache to n, |
| 46 |
// dropping excess blocks if n is less than the |
| 47 |
// number of cached blocks. |
| 48 |
Resize(n int) |
| 49 |
|
| 50 |
// Drop evicts n elements from the cache according |
| 51 |
// to the cache eviction policy. |
| 52 |
Drop(n int) |
| 53 |
} |
| 54 |
|
| 55 |
func insertAfter(pos, n *node) { |
| 56 |
n.prev = pos |
| 57 |
pos.next, n.next, pos.next.prev = n, pos.next, n |
| 58 |
} |
| 59 |
|
| 60 |
func remove(n *node, table map[int64]*node) { |
| 61 |
delete(table, n.b.Base()) |
| 62 |
n.prev.next = n.next |
| 63 |
n.next.prev = n.prev |
| 64 |
n.next = nil |
| 65 |
n.prev = nil |
| 66 |
} |
| 67 |
|
| 68 |
// NewLRU returns an LRU cache with n slots. If n is less than 1 |
| 69 |
// a nil cache is returned. |
| 70 |
func NewLRU(n int) Cache { |
| 71 |
if n < 1 { |
| 72 |
return nil |
| 73 |
} |
| 74 |
c := LRU{ |
| 75 |
table: make(map[int64]*node, n), |
| 76 |
cap: n, |
| 77 |
} |
| 78 |
c.root.next = &c.root |
| 79 |
c.root.prev = &c.root |
| 80 |
return &c |
| 81 |
} |
| 82 |
|
| 83 |
// LRU satisfies the Cache interface with least recently used eviction |
| 84 |
// behavior where Unused Blocks are preferentially evicted. |
| 85 |
type LRU struct { |
| 86 |
mu sync.RWMutex |
| 87 |
root node |
| 88 |
table map[int64]*node |
| 89 |
cap int |
| 90 |
} |
| 91 |
|
| 92 |
type node struct { |
| 93 |
b bgzf.Block |
| 94 |
|
| 95 |
next, prev *node |
| 96 |
} |
| 97 |
|
| 98 |
// Len returns the number of elements held by the cache. |
| 99 |
func (c *LRU) Len() int { |
| 100 |
c.mu.RLock() |
| 101 |
defer c.mu.RUnlock() |
| 102 |
|
| 103 |
return len(c.table) |
| 104 |
} |
| 105 |
|
| 106 |
// Cap returns the maximum number of elements that can be held by the cache. |
| 107 |
func (c *LRU) Cap() int { |
| 108 |
c.mu.RLock() |
| 109 |
defer c.mu.RUnlock() |
| 110 |
|
| 111 |
return c.cap |
| 112 |
} |
| 113 |
|
| 114 |
// Resize changes the capacity of the cache to n, dropping excess blocks |
| 115 |
// if n is less than the number of cached blocks. |
| 116 |
func (c *LRU) Resize(n int) { |
| 117 |
c.mu.Lock() |
| 118 |
if n < len(c.table) { |
| 119 |
c.drop(len(c.table) - n) |
| 120 |
} |
| 121 |
c.cap = n |
| 122 |
c.mu.Unlock() |
| 123 |
} |
| 124 |
|
| 125 |
// Drop evicts n elements from the cache according to the cache eviction policy. |
| 126 |
func (c *LRU) Drop(n int) { |
| 127 |
c.mu.Lock() |
| 128 |
c.drop(n) |
| 129 |
c.mu.Unlock() |
| 130 |
} |
| 131 |
|
| 132 |
func (c *LRU) drop(n int) { |
| 133 |
for ; n > 0 && c.Len() > 0; n-- { |
| 134 |
remove(c.root.prev, c.table) |
| 135 |
} |
| 136 |
} |
| 137 |
|
| 138 |
// Get returns the Block in the Cache with the specified base or a nil Block |
| 139 |
// if it does not exist. |
| 140 |
func (c *LRU) Get(base int64) bgzf.Block { |
| 141 |
c.mu.Lock() |
| 142 |
defer c.mu.Unlock() |
| 143 |
|
| 144 |
n, ok := c.table[base] |
| 145 |
if !ok { |
| 146 |
return nil |
| 147 |
} |
| 148 |
remove(n, c.table) |
| 149 |
return n.b |
| 150 |
} |
| 151 |
|
| 152 |
// Peek returns a boolean indicating whether a Block exists in the Cache for |
| 153 |
// the given base offset and the expected offset for the subsequent Block in |
| 154 |
// the BGZF stream. |
| 155 |
func (c *LRU) Peek(base int64) (exist bool, next int64) { |
| 156 |
c.mu.RLock() |
| 157 |
defer c.mu.RUnlock() |
| 158 |
|
| 159 |
n, exist := c.table[base] |
| 160 |
if !exist { |
| 161 |
return false, -1 |
| 162 |
} |
| 163 |
next = n.b.NextBase() |
| 164 |
return exist, next |
| 165 |
} |
| 166 |
|
| 167 |
// Put inserts a Block into the Cache, returning the Block that was evicted or |
| 168 |
// nil if no eviction was necessary and the Block was retained. Unused Blocks |
| 169 |
// are not retained but are returned if the Cache is full. |
| 170 |
func (c *LRU) Put(b bgzf.Block) (evicted bgzf.Block, retained bool) { |
| 171 |
c.mu.Lock() |
| 172 |
defer c.mu.Unlock() |
| 173 |
|
| 174 |
var d bgzf.Block |
| 175 |
if _, ok := c.table[b.Base()]; ok { |
| 176 |
return b, false |
| 177 |
} |
| 178 |
used := b.Used() |
| 179 |
if len(c.table) == c.cap { |
| 180 |
if !used { |
| 181 |
return b, false |
| 182 |
} |
| 183 |
d = c.root.prev.b |
| 184 |
remove(c.root.prev, c.table) |
| 185 |
} |
| 186 |
n := &node{b: b} |
| 187 |
c.table[b.Base()] = n |
| 188 |
if used { |
| 189 |
insertAfter(&c.root, n) |
| 190 |
} else { |
| 191 |
insertAfter(c.root.prev, n) |
| 192 |
} |
| 193 |
return d, true |
| 194 |
} |
| 195 |
|
| 196 |
// NewFIFO returns a FIFO cache with n slots. If n is less than 1 |
| 197 |
// a nil cache is returned. |
| 198 |
func NewFIFO(n int) Cache { |
| 199 |
if n < 1 { |
| 200 |
return nil |
| 201 |
} |
| 202 |
c := FIFO{ |
| 203 |
table: make(map[int64]*node, n), |
| 204 |
cap: n, |
| 205 |
} |
| 206 |
c.root.next = &c.root |
| 207 |
c.root.prev = &c.root |
| 208 |
return &c |
| 209 |
} |
| 210 |
|
| 211 |
// FIFO satisfies the Cache interface with first in first out eviction |
| 212 |
// behavior where Unused Blocks are preferentially evicted. |
| 213 |
type FIFO struct { |
| 214 |
mu sync.RWMutex |
| 215 |
root node |
| 216 |
table map[int64]*node |
| 217 |
cap int |
| 218 |
} |
| 219 |
|
| 220 |
// Len returns the number of elements held by the cache. |
| 221 |
func (c *FIFO) Len() int { |
| 222 |
c.mu.RLock() |
| 223 |
defer c.mu.RUnlock() |
| 224 |
|
| 225 |
return len(c.table) |
| 226 |
} |
| 227 |
|
| 228 |
// Cap returns the maximum number of elements that can be held by the cache. |
| 229 |
func (c *FIFO) Cap() int { |
| 230 |
c.mu.RLock() |
| 231 |
defer c.mu.RUnlock() |
| 232 |
|
| 233 |
return c.cap |
| 234 |
} |
| 235 |
|
| 236 |
// Resize changes the capacity of the cache to n, dropping excess blocks |
| 237 |
// if n is less than the number of cached blocks. |
| 238 |
func (c *FIFO) Resize(n int) { |
| 239 |
c.mu.Lock() |
| 240 |
if n < len(c.table) { |
| 241 |
c.drop(len(c.table) - n) |
| 242 |
} |
| 243 |
c.cap = n |
| 244 |
c.mu.Unlock() |
| 245 |
} |
| 246 |
|
| 247 |
// Drop evicts n elements from the cache according to the cache eviction policy. |
| 248 |
func (c *FIFO) Drop(n int) { |
| 249 |
c.mu.Lock() |
| 250 |
c.drop(n) |
| 251 |
c.mu.Unlock() |
| 252 |
} |
| 253 |
|
| 254 |
func (c *FIFO) drop(n int) { |
| 255 |
for ; n > 0 && c.Len() > 0; n-- { |
| 256 |
remove(c.root.prev, c.table) |
| 257 |
} |
| 258 |
} |
| 259 |
|
| 260 |
// Get returns the Block in the Cache with the specified base or a nil Block |
| 261 |
// if it does not exist. |
| 262 |
func (c *FIFO) Get(base int64) bgzf.Block { |
| 263 |
c.mu.Lock() |
| 264 |
defer c.mu.Unlock() |
| 265 |
|
| 266 |
n, ok := c.table[base] |
| 267 |
if !ok { |
| 268 |
return nil |
| 269 |
} |
| 270 |
if !n.b.Used() { |
| 271 |
remove(n, c.table) |
| 272 |
} |
| 273 |
return n.b |
| 274 |
} |
| 275 |
|
| 276 |
// Peek returns a boolean indicating whether a Block exists in the Cache for |
| 277 |
// the given base offset and the expected offset for the subsequent Block in |
| 278 |
// the BGZF stream. |
| 279 |
func (c *FIFO) Peek(base int64) (exist bool, next int64) { |
| 280 |
c.mu.RLock() |
| 281 |
defer c.mu.RUnlock() |
| 282 |
|
| 283 |
n, exist := c.table[base] |
| 284 |
if !exist { |
| 285 |
return false, -1 |
| 286 |
} |
| 287 |
next = n.b.NextBase() |
| 288 |
return exist, next |
| 289 |
} |
| 290 |
|
| 291 |
// Put inserts a Block into the Cache, returning the Block that was evicted or |
| 292 |
// nil if no eviction was necessary and the Block was retained. Unused Blocks |
| 293 |
// are not retained but are returned if the Cache is full. |
| 294 |
func (c *FIFO) Put(b bgzf.Block) (evicted bgzf.Block, retained bool) { |
| 295 |
c.mu.Lock() |
| 296 |
defer c.mu.Unlock() |
| 297 |
|
| 298 |
var d bgzf.Block |
| 299 |
if _, ok := c.table[b.Base()]; ok { |
| 300 |
return b, false |
| 301 |
} |
| 302 |
used := b.Used() |
| 303 |
if len(c.table) == c.cap { |
| 304 |
if !used { |
| 305 |
return b, false |
| 306 |
} |
| 307 |
d = c.root.prev.b |
| 308 |
remove(c.root.prev, c.table) |
| 309 |
} |
| 310 |
n := &node{b: b} |
| 311 |
c.table[b.Base()] = n |
| 312 |
if used { |
| 313 |
insertAfter(&c.root, n) |
| 314 |
} else { |
| 315 |
insertAfter(c.root.prev, n) |
| 316 |
} |
| 317 |
return d, true |
| 318 |
} |
| 319 |
|
| 320 |
// NewRandom returns a random eviction cache with n slots. If n is less than 1 |
| 321 |
// a nil cache is returned. |
| 322 |
func NewRandom(n int) Cache { |
| 323 |
if n < 1 { |
| 324 |
return nil |
| 325 |
} |
| 326 |
return &Random{ |
| 327 |
table: make(map[int64]bgzf.Block, n), |
| 328 |
cap: n, |
| 329 |
} |
| 330 |
} |
| 331 |
|
| 332 |
// Random satisfies the Cache interface with random eviction behavior |
| 333 |
// where Unused Blocks are preferentially evicted. |
| 334 |
type Random struct { |
| 335 |
mu sync.RWMutex |
| 336 |
table map[int64]bgzf.Block |
| 337 |
cap int |
| 338 |
} |
| 339 |
|
| 340 |
// Len returns the number of elements held by the cache. |
| 341 |
func (c *Random) Len() int { |
| 342 |
c.mu.RLock() |
| 343 |
defer c.mu.RUnlock() |
| 344 |
|
| 345 |
return len(c.table) |
| 346 |
} |
| 347 |
|
| 348 |
// Cap returns the maximum number of elements that can be held by the cache. |
| 349 |
func (c *Random) Cap() int { |
| 350 |
c.mu.RLock() |
| 351 |
defer c.mu.RUnlock() |
| 352 |
|
| 353 |
return c.cap |
| 354 |
} |
| 355 |
|
| 356 |
// Resize changes the capacity of the cache to n, dropping excess blocks |
| 357 |
// if n is less than the number of cached blocks. |
| 358 |
func (c *Random) Resize(n int) { |
| 359 |
c.mu.Lock() |
| 360 |
if n < len(c.table) { |
| 361 |
c.drop(len(c.table) - n) |
| 362 |
} |
| 363 |
c.cap = n |
| 364 |
c.mu.Unlock() |
| 365 |
} |
| 366 |
|
| 367 |
// Drop evicts n elements from the cache according to the cache eviction policy. |
| 368 |
func (c *Random) Drop(n int) { |
| 369 |
c.mu.Lock() |
| 370 |
c.drop(n) |
| 371 |
c.mu.Unlock() |
| 372 |
} |
| 373 |
|
| 374 |
func (c *Random) drop(n int) { |
| 375 |
if n < 1 { |
| 376 |
return |
| 377 |
} |
| 378 |
for k, b := range c.table { |
| 379 |
if b.Used() { |
| 380 |
continue |
| 381 |
} |
| 382 |
delete(c.table, k) |
| 383 |
if n--; n == 0 { |
| 384 |
return |
| 385 |
} |
| 386 |
} |
| 387 |
for k := range c.table { |
| 388 |
delete(c.table, k) |
| 389 |
if n--; n == 0 { |
| 390 |
break |
| 391 |
} |
| 392 |
} |
| 393 |
} |
| 394 |
|
| 395 |
// Get returns the Block in the Cache with the specified base or a nil Block |
| 396 |
// if it does not exist. |
| 397 |
func (c *Random) Get(base int64) bgzf.Block { |
| 398 |
c.mu.Lock() |
| 399 |
defer c.mu.Unlock() |
| 400 |
|
| 401 |
b, ok := c.table[base] |
| 402 |
if !ok { |
| 403 |
return nil |
| 404 |
} |
| 405 |
delete(c.table, base) |
| 406 |
return b |
| 407 |
} |
| 408 |
|
| 409 |
// Peek returns a boolean indicating whether a Block exists in the Cache for |
| 410 |
// the given base offset and the expected offset for the subsequent Block in |
| 411 |
// the BGZF stream. |
| 412 |
func (c *Random) Peek(base int64) (exist bool, next int64) { |
| 413 |
c.mu.RLock() |
| 414 |
defer c.mu.RUnlock() |
| 415 |
|
| 416 |
n, exist := c.table[base] |
| 417 |
if !exist { |
| 418 |
return false, -1 |
| 419 |
} |
| 420 |
next = n.NextBase() |
| 421 |
return exist, next |
| 422 |
} |
| 423 |
|
| 424 |
// Put inserts a Block into the Cache, returning the Block that was evicted or |
| 425 |
// nil if no eviction was necessary and the Block was retained. Unused Blocks |
| 426 |
// are not retained but are returned if the Cache is full. |
| 427 |
func (c *Random) Put(b bgzf.Block) (evicted bgzf.Block, retained bool) { |
| 428 |
c.mu.Lock() |
| 429 |
defer c.mu.Unlock() |
| 430 |
|
| 431 |
var d bgzf.Block |
| 432 |
if _, ok := c.table[b.Base()]; ok { |
| 433 |
return b, false |
| 434 |
} |
| 435 |
if len(c.table) == c.cap { |
| 436 |
if !b.Used() { |
| 437 |
return b, false |
| 438 |
} |
| 439 |
for k, v := range c.table { |
| 440 |
if v.Used() { |
| 441 |
continue |
| 442 |
} |
| 443 |
delete(c.table, k) |
| 444 |
d = v |
| 445 |
goto done |
| 446 |
} |
| 447 |
for k, v := range c.table { |
| 448 |
delete(c.table, k) |
| 449 |
d = v |
| 450 |
break |
| 451 |
} |
| 452 |
done: |
| 453 |
} |
| 454 |
c.table[b.Base()] = b |
| 455 |
return d, true |
| 456 |
} |
| 457 |
|
| 458 |
// StatsRecorder allows a bgzf.Cache to capture cache statistics. |
| 459 |
type StatsRecorder struct { |
| 460 |
bgzf.Cache |
| 461 |
|
| 462 |
mu sync.RWMutex |
| 463 |
stats Stats |
| 464 |
} |
| 465 |
|
| 466 |
// Stats represents statistics of a bgzf.Cache. |
| 467 |
type Stats struct { |
| 468 |
Gets int // number of Get operations |
| 469 |
Misses int // number of cache misses |
| 470 |
Puts int // number of Put operations |
| 471 |
Retains int // number of times a Put has resulted in Block retention |
| 472 |
Evictions int // number of times a Put has resulted in a Block eviction |
| 473 |
} |
| 474 |
|
| 475 |
// Stats returns the current statistics for the cache. |
| 476 |
func (s *StatsRecorder) Stats() Stats { |
| 477 |
s.mu.RLock() |
| 478 |
defer s.mu.RUnlock() |
| 479 |
return s.stats |
| 480 |
} |
| 481 |
|
| 482 |
// Reset zeros the statistics kept by the StatsRecorder. |
| 483 |
func (s *StatsRecorder) Reset() { |
| 484 |
s.mu.Lock() |
| 485 |
s.stats = Stats{} |
| 486 |
s.mu.Unlock() |
| 487 |
} |
| 488 |
|
| 489 |
// Get returns the Block in the underlying Cache with the specified base or a nil |
| 490 |
// Block if it does not exist. It updates the gets and misses statistics. |
| 491 |
func (s *StatsRecorder) Get(base int64) bgzf.Block { |
| 492 |
s.mu.Lock() |
| 493 |
s.stats.Gets++ |
| 494 |
blk := s.Cache.Get(base) |
| 495 |
if blk == nil { |
| 496 |
s.stats.Misses++ |
| 497 |
} |
| 498 |
s.mu.Unlock() |
| 499 |
return blk |
| 500 |
} |
| 501 |
|
| 502 |
// Put inserts a Block into the underlying Cache, returning the Block and eviction |
| 503 |
// status according to the underlying cache behavior. It updates the puts, retains and |
| 504 |
// evictions statistics. |
| 505 |
func (s *StatsRecorder) Put(b bgzf.Block) (evicted bgzf.Block, retained bool) { |
| 506 |
s.mu.Lock() |
| 507 |
s.stats.Puts++ |
| 508 |
blk, retained := s.Cache.Put(b) |
| 509 |
if retained { |
| 510 |
s.stats.Retains++ |
| 511 |
if blk != nil { |
| 512 |
s.stats.Evictions++ |
| 513 |
} |
| 514 |
} |
| 515 |
s.mu.Unlock() |
| 516 |
return blk, retained |
| 517 |
} |