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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679
|
package iradix
import (
"bytes"
"strings"
"github.com/hashicorp/golang-lru/v2/simplelru"
)
const (
// defaultModifiedCache is the default size of the modified node
// cache used per transaction. This is used to cache the updates
// to the nodes near the root, while the leaves do not need to be
// cached. This is important for very large transactions to prevent
// the modified cache from growing to be enormous. This is also used
// to set the max size of the mutation notify maps since those should
// also be bounded in a similar way.
defaultModifiedCache = 8192
)
// Tree implements an immutable radix tree. This can be treated as a
// Dictionary abstract data type. The main advantage over a standard
// hash map is prefix-based lookups and ordered iteration. The immutability
// means that it is safe to concurrently read from a Tree without any
// coordination.
type Tree[T any] struct {
root *Node[T]
size int
}
// New returns an empty Tree
func New[T any]() *Tree[T] {
t := &Tree[T]{
root: &Node[T]{
mutateCh: make(chan struct{}),
},
}
return t
}
// Len is used to return the number of elements in the tree
func (t *Tree[T]) Len() int {
return t.size
}
// Txn is a transaction on the tree. This transaction is applied
// atomically and returns a new tree when committed. A transaction
// is not thread safe, and should only be used by a single goroutine.
type Txn[T any] struct {
// root is the modified root for the transaction.
root *Node[T]
// snap is a snapshot of the root node for use if we have to run the
// slow notify algorithm.
snap *Node[T]
// size tracks the size of the tree as it is modified during the
// transaction.
size int
// writable is a cache of writable nodes that have been created during
// the course of the transaction. This allows us to re-use the same
// nodes for further writes and avoid unnecessary copies of nodes that
// have never been exposed outside the transaction. This will only hold
// up to defaultModifiedCache number of entries.
writable *simplelru.LRU[*Node[T], any]
// trackChannels is used to hold channels that need to be notified to
// signal mutation of the tree. This will only hold up to
// defaultModifiedCache number of entries, after which we will set the
// trackOverflow flag, which will cause us to use a more expensive
// algorithm to perform the notifications. Mutation tracking is only
// performed if trackMutate is true.
trackChannels map[chan struct{}]struct{}
trackOverflow bool
trackMutate bool
}
// Txn starts a new transaction that can be used to mutate the tree
func (t *Tree[T]) Txn() *Txn[T] {
txn := &Txn[T]{
root: t.root,
snap: t.root,
size: t.size,
}
return txn
}
// Clone makes an independent copy of the transaction. The new transaction
// does not track any nodes and has TrackMutate turned off. The cloned transaction will contain any uncommitted writes in the original transaction but further mutations to either will be independent and result in different radix trees on Commit. A cloned transaction may be passed to another goroutine and mutated there independently however each transaction may only be mutated in a single thread.
func (t *Txn[T]) Clone() *Txn[T] {
// reset the writable node cache to avoid leaking future writes into the clone
t.writable = nil
txn := &Txn[T]{
root: t.root,
snap: t.snap,
size: t.size,
}
return txn
}
// TrackMutate can be used to toggle if mutations are tracked. If this is enabled
// then notifications will be issued for affected internal nodes and leaves when
// the transaction is committed.
func (t *Txn[T]) TrackMutate(track bool) {
t.trackMutate = track
}
// trackChannel safely attempts to track the given mutation channel, setting the
// overflow flag if we can no longer track any more. This limits the amount of
// state that will accumulate during a transaction and we have a slower algorithm
// to switch to if we overflow.
func (t *Txn[T]) trackChannel(ch chan struct{}) {
// In overflow, make sure we don't store any more objects.
if t.trackOverflow {
return
}
// If this would overflow the state we reject it and set the flag (since
// we aren't tracking everything that's required any longer).
if len(t.trackChannels) >= defaultModifiedCache {
// Mark that we are in the overflow state
t.trackOverflow = true
// Clear the map so that the channels can be garbage collected. It is
// safe to do this since we have already overflowed and will be using
// the slow notify algorithm.
t.trackChannels = nil
return
}
// Create the map on the fly when we need it.
if t.trackChannels == nil {
t.trackChannels = make(map[chan struct{}]struct{})
}
// Otherwise we are good to track it.
t.trackChannels[ch] = struct{}{}
}
// writeNode returns a node to be modified, if the current node has already been
// modified during the course of the transaction, it is used in-place. Set
// forLeafUpdate to true if you are getting a write node to update the leaf,
// which will set leaf mutation tracking appropriately as well.
func (t *Txn[T]) writeNode(n *Node[T], forLeafUpdate bool) *Node[T] {
// Ensure the writable set exists.
if t.writable == nil {
lru, err := simplelru.NewLRU[*Node[T], any](defaultModifiedCache, nil)
if err != nil {
panic(err)
}
t.writable = lru
}
// If this node has already been modified, we can continue to use it
// during this transaction. We know that we don't need to track it for
// a node update since the node is writable, but if this is for a leaf
// update we track it, in case the initial write to this node didn't
// update the leaf.
if _, ok := t.writable.Get(n); ok {
if t.trackMutate && forLeafUpdate && n.leaf != nil {
t.trackChannel(n.leaf.mutateCh)
}
return n
}
// Mark this node as being mutated.
if t.trackMutate {
t.trackChannel(n.mutateCh)
}
// Mark its leaf as being mutated, if appropriate.
if t.trackMutate && forLeafUpdate && n.leaf != nil {
t.trackChannel(n.leaf.mutateCh)
}
// Copy the existing node. If you have set forLeafUpdate it will be
// safe to replace this leaf with another after you get your node for
// writing. You MUST replace it, because the channel associated with
// this leaf will be closed when this transaction is committed.
nc := &Node[T]{
mutateCh: make(chan struct{}),
leaf: n.leaf,
}
if n.prefix != nil {
nc.prefix = make([]byte, len(n.prefix))
copy(nc.prefix, n.prefix)
}
if len(n.edges) != 0 {
nc.edges = make([]edge[T], len(n.edges))
copy(nc.edges, n.edges)
}
// Mark this node as writable.
t.writable.Add(nc, nil)
return nc
}
// Visit all the nodes in the tree under n, and add their mutateChannels to the transaction
// Returns the size of the subtree visited
func (t *Txn[T]) trackChannelsAndCount(n *Node[T]) int {
// Count only leaf nodes
leaves := 0
if n.leaf != nil {
leaves = 1
}
// Mark this node as being mutated.
if t.trackMutate {
t.trackChannel(n.mutateCh)
}
// Mark its leaf as being mutated, if appropriate.
if t.trackMutate && n.leaf != nil {
t.trackChannel(n.leaf.mutateCh)
}
// Recurse on the children
for _, e := range n.edges {
leaves += t.trackChannelsAndCount(e.node)
}
return leaves
}
// mergeChild is called to collapse the given node with its child. This is only
// called when the given node is not a leaf and has a single edge.
func (t *Txn[T]) mergeChild(n *Node[T]) {
// Mark the child node as being mutated since we are about to abandon
// it. We don't need to mark the leaf since we are retaining it if it
// is there.
e := n.edges[0]
child := e.node
if t.trackMutate {
t.trackChannel(child.mutateCh)
}
// Merge the nodes.
n.prefix = concat(n.prefix, child.prefix)
n.leaf = child.leaf
if len(child.edges) != 0 {
n.edges = make([]edge[T], len(child.edges))
copy(n.edges, child.edges)
} else {
n.edges = nil
}
}
// insert does a recursive insertion
func (t *Txn[T]) insert(n *Node[T], k, search []byte, v T) (*Node[T], T, bool) {
var zero T
// Handle key exhaustion
if len(search) == 0 {
var oldVal T
didUpdate := false
if n.isLeaf() {
oldVal = n.leaf.val
didUpdate = true
}
nc := t.writeNode(n, true)
nc.leaf = &leafNode[T]{
mutateCh: make(chan struct{}),
key: k,
val: v,
}
return nc, oldVal, didUpdate
}
// Look for the edge
idx, child := n.getEdge(search[0])
// No edge, create one
if child == nil {
e := edge[T]{
label: search[0],
node: &Node[T]{
mutateCh: make(chan struct{}),
leaf: &leafNode[T]{
mutateCh: make(chan struct{}),
key: k,
val: v,
},
prefix: search,
},
}
nc := t.writeNode(n, false)
nc.addEdge(e)
return nc, zero, false
}
// Determine longest prefix of the search key on match
commonPrefix := longestPrefix(search, child.prefix)
if commonPrefix == len(child.prefix) {
search = search[commonPrefix:]
newChild, oldVal, didUpdate := t.insert(child, k, search, v)
if newChild != nil {
nc := t.writeNode(n, false)
nc.edges[idx].node = newChild
return nc, oldVal, didUpdate
}
return nil, oldVal, didUpdate
}
// Split the node
nc := t.writeNode(n, false)
splitNode := &Node[T]{
mutateCh: make(chan struct{}),
prefix: search[:commonPrefix],
}
nc.replaceEdge(edge[T]{
label: search[0],
node: splitNode,
})
// Restore the existing child node
modChild := t.writeNode(child, false)
splitNode.addEdge(edge[T]{
label: modChild.prefix[commonPrefix],
node: modChild,
})
modChild.prefix = modChild.prefix[commonPrefix:]
// Create a new leaf node
leaf := &leafNode[T]{
mutateCh: make(chan struct{}),
key: k,
val: v,
}
// If the new key is a subset, add to to this node
search = search[commonPrefix:]
if len(search) == 0 {
splitNode.leaf = leaf
return nc, zero, false
}
// Create a new edge for the node
splitNode.addEdge(edge[T]{
label: search[0],
node: &Node[T]{
mutateCh: make(chan struct{}),
leaf: leaf,
prefix: search,
},
})
return nc, zero, false
}
// delete does a recursive deletion
func (t *Txn[T]) delete(n *Node[T], search []byte) (*Node[T], *leafNode[T]) {
// Check for key exhaustion
if len(search) == 0 {
if !n.isLeaf() {
return nil, nil
}
// Copy the pointer in case we are in a transaction that already
// modified this node since the node will be reused. Any changes
// made to the node will not affect returning the original leaf
// value.
oldLeaf := n.leaf
// Remove the leaf node
nc := t.writeNode(n, true)
nc.leaf = nil
// Check if this node should be merged
if n != t.root && len(nc.edges) == 1 {
t.mergeChild(nc)
}
return nc, oldLeaf
}
// Look for an edge
label := search[0]
idx, child := n.getEdge(label)
if child == nil || !bytes.HasPrefix(search, child.prefix) {
return nil, nil
}
// Consume the search prefix
search = search[len(child.prefix):]
newChild, leaf := t.delete(child, search)
if newChild == nil {
return nil, nil
}
// Copy this node. WATCH OUT - it's safe to pass "false" here because we
// will only ADD a leaf via nc.mergeChild() if there isn't one due to
// the !nc.isLeaf() check in the logic just below. This is pretty subtle,
// so be careful if you change any of the logic here.
nc := t.writeNode(n, false)
// Delete the edge if the node has no edges
if newChild.leaf == nil && len(newChild.edges) == 0 {
nc.delEdge(label)
if n != t.root && len(nc.edges) == 1 && !nc.isLeaf() {
t.mergeChild(nc)
}
} else {
nc.edges[idx].node = newChild
}
return nc, leaf
}
// delete does a recursive deletion
func (t *Txn[T]) deletePrefix(n *Node[T], search []byte) (*Node[T], int) {
// Check for key exhaustion
if len(search) == 0 {
nc := t.writeNode(n, true)
if n.isLeaf() {
nc.leaf = nil
}
nc.edges = nil
return nc, t.trackChannelsAndCount(n)
}
// Look for an edge
label := search[0]
idx, child := n.getEdge(label)
// We make sure that either the child node's prefix starts with the search term, or the search term starts with the child node's prefix
// Need to do both so that we can delete prefixes that don't correspond to any node in the tree
if child == nil || (!bytes.HasPrefix(child.prefix, search) && !bytes.HasPrefix(search, child.prefix)) {
return nil, 0
}
// Consume the search prefix
if len(child.prefix) > len(search) {
search = []byte("")
} else {
search = search[len(child.prefix):]
}
newChild, numDeletions := t.deletePrefix(child, search)
if newChild == nil {
return nil, 0
}
// Copy this node. WATCH OUT - it's safe to pass "false" here because we
// will only ADD a leaf via nc.mergeChild() if there isn't one due to
// the !nc.isLeaf() check in the logic just below. This is pretty subtle,
// so be careful if you change any of the logic here.
nc := t.writeNode(n, false)
// Delete the edge if the node has no edges
if newChild.leaf == nil && len(newChild.edges) == 0 {
nc.delEdge(label)
if n != t.root && len(nc.edges) == 1 && !nc.isLeaf() {
t.mergeChild(nc)
}
} else {
nc.edges[idx].node = newChild
}
return nc, numDeletions
}
// Insert is used to add or update a given key. The return provides
// the previous value and a bool indicating if any was set.
func (t *Txn[T]) Insert(k []byte, v T) (T, bool) {
newRoot, oldVal, didUpdate := t.insert(t.root, k, k, v)
if newRoot != nil {
t.root = newRoot
}
if !didUpdate {
t.size++
}
return oldVal, didUpdate
}
// Delete is used to delete a given key. Returns the old value if any,
// and a bool indicating if the key was set.
func (t *Txn[T]) Delete(k []byte) (T, bool) {
var zero T
newRoot, leaf := t.delete(t.root, k)
if newRoot != nil {
t.root = newRoot
}
if leaf != nil {
t.size--
return leaf.val, true
}
return zero, false
}
// DeletePrefix is used to delete an entire subtree that matches the prefix
// This will delete all nodes under that prefix
func (t *Txn[T]) DeletePrefix(prefix []byte) bool {
newRoot, numDeletions := t.deletePrefix(t.root, prefix)
if newRoot != nil {
t.root = newRoot
t.size = t.size - numDeletions
return true
}
return false
}
// Root returns the current root of the radix tree within this
// transaction. The root is not safe across insert and delete operations,
// but can be used to read the current state during a transaction.
func (t *Txn[T]) Root() *Node[T] {
return t.root
}
// Get is used to lookup a specific key, returning
// the value and if it was found
func (t *Txn[T]) Get(k []byte) (T, bool) {
return t.root.Get(k)
}
// GetWatch is used to lookup a specific key, returning
// the watch channel, value and if it was found
func (t *Txn[T]) GetWatch(k []byte) (<-chan struct{}, T, bool) {
return t.root.GetWatch(k)
}
// Commit is used to finalize the transaction and return a new tree. If mutation
// tracking is turned on then notifications will also be issued.
func (t *Txn[T]) Commit() *Tree[T] {
nt := t.CommitOnly()
if t.trackMutate {
t.Notify()
}
return nt
}
// CommitOnly is used to finalize the transaction and return a new tree, but
// does not issue any notifications until Notify is called.
func (t *Txn[T]) CommitOnly() *Tree[T] {
nt := &Tree[T]{t.root, t.size}
t.writable = nil
return nt
}
// slowNotify does a complete comparison of the before and after trees in order
// to trigger notifications. This doesn't require any additional state but it
// is very expensive to compute.
func (t *Txn[T]) slowNotify() {
snapIter := t.snap.rawIterator()
rootIter := t.root.rawIterator()
for snapIter.Front() != nil || rootIter.Front() != nil {
// If we've exhausted the nodes in the old snapshot, we know
// there's nothing remaining to notify.
if snapIter.Front() == nil {
return
}
snapElem := snapIter.Front()
// If we've exhausted the nodes in the new root, we know we need
// to invalidate everything that remains in the old snapshot. We
// know from the loop condition there's something in the old
// snapshot.
if rootIter.Front() == nil {
close(snapElem.mutateCh)
if snapElem.isLeaf() {
close(snapElem.leaf.mutateCh)
}
snapIter.Next()
continue
}
// Do one string compare so we can check the various conditions
// below without repeating the compare.
cmp := strings.Compare(snapIter.Path(), rootIter.Path())
// If the snapshot is behind the root, then we must have deleted
// this node during the transaction.
if cmp < 0 {
close(snapElem.mutateCh)
if snapElem.isLeaf() {
close(snapElem.leaf.mutateCh)
}
snapIter.Next()
continue
}
// If the snapshot is ahead of the root, then we must have added
// this node during the transaction.
if cmp > 0 {
rootIter.Next()
continue
}
// If we have the same path, then we need to see if we mutated a
// node and possibly the leaf.
rootElem := rootIter.Front()
if snapElem != rootElem {
close(snapElem.mutateCh)
if snapElem.leaf != nil && (snapElem.leaf != rootElem.leaf) {
close(snapElem.leaf.mutateCh)
}
}
snapIter.Next()
rootIter.Next()
}
}
// Notify is used along with TrackMutate to trigger notifications. This must
// only be done once a transaction is committed via CommitOnly, and it is called
// automatically by Commit.
func (t *Txn[T]) Notify() {
if !t.trackMutate {
return
}
// If we've overflowed the tracking state we can't use it in any way and
// need to do a full tree compare.
if t.trackOverflow {
t.slowNotify()
} else {
for ch := range t.trackChannels {
close(ch)
}
}
// Clean up the tracking state so that a re-notify is safe (will trigger
// the else clause above which will be a no-op).
t.trackChannels = nil
t.trackOverflow = false
}
// Insert is used to add or update a given key. The return provides
// the new tree, previous value and a bool indicating if any was set.
func (t *Tree[T]) Insert(k []byte, v T) (*Tree[T], T, bool) {
txn := t.Txn()
old, ok := txn.Insert(k, v)
return txn.Commit(), old, ok
}
// Delete is used to delete a given key. Returns the new tree,
// old value if any, and a bool indicating if the key was set.
func (t *Tree[T]) Delete(k []byte) (*Tree[T], T, bool) {
txn := t.Txn()
old, ok := txn.Delete(k)
return txn.Commit(), old, ok
}
// DeletePrefix is used to delete all nodes starting with a given prefix. Returns the new tree,
// and a bool indicating if the prefix matched any nodes
func (t *Tree[T]) DeletePrefix(k []byte) (*Tree[T], bool) {
txn := t.Txn()
ok := txn.DeletePrefix(k)
return txn.Commit(), ok
}
// Root returns the root node of the tree which can be used for richer
// query operations.
func (t *Tree[T]) Root() *Node[T] {
return t.root
}
// Get is used to lookup a specific key, returning
// the value and if it was found
func (t *Tree[T]) Get(k []byte) (T, bool) {
return t.root.Get(k)
}
// longestPrefix finds the length of the shared prefix
// of two strings
func longestPrefix(k1, k2 []byte) int {
max := len(k1)
if l := len(k2); l < max {
max = l
}
var i int
for i = 0; i < max; i++ {
if k1[i] != k2[i] {
break
}
}
return i
}
// concat two byte slices, returning a third new copy
func concat(a, b []byte) []byte {
c := make([]byte, len(a)+len(b))
copy(c, a)
copy(c[len(a):], b)
return c
}
|