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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829
|
package memdb
import (
"bytes"
"fmt"
"strings"
"sync/atomic"
"unsafe"
iradix "github.com/hashicorp/go-immutable-radix"
)
const (
id = "id"
)
var (
// ErrNotFound is returned when the requested item is not found
ErrNotFound = fmt.Errorf("not found")
)
// tableIndex is a tuple of (Table, Index) used for lookups
type tableIndex struct {
Table string
Index string
}
// Txn is a transaction against a MemDB.
// This can be a read or write transaction.
type Txn struct {
db *MemDB
write bool
rootTxn *iradix.Txn
after []func()
// changes is used to track the changes performed during the transaction. If
// it is nil at transaction start then changes are not tracked.
changes Changes
modified map[tableIndex]*iradix.Txn
}
// TrackChanges enables change tracking for the transaction. If called at any
// point before commit, subsequent mutations will be recorded and can be
// retrieved using ChangeSet. Once this has been called on a transaction it
// can't be unset. As with other Txn methods it's not safe to call this from a
// different goroutine than the one making mutations or committing the
// transaction.
func (txn *Txn) TrackChanges() {
if txn.changes == nil {
txn.changes = make(Changes, 0, 1)
}
}
// readableIndex returns a transaction usable for reading the given
// index in a table. If a write transaction is in progress, we may need
// to use an existing modified txn.
func (txn *Txn) readableIndex(table, index string) *iradix.Txn {
// Look for existing transaction
if txn.write && txn.modified != nil {
key := tableIndex{table, index}
exist, ok := txn.modified[key]
if ok {
return exist
}
}
// Create a read transaction
path := indexPath(table, index)
raw, _ := txn.rootTxn.Get(path)
indexTxn := raw.(*iradix.Tree).Txn()
return indexTxn
}
// writableIndex returns a transaction usable for modifying the
// given index in a table.
func (txn *Txn) writableIndex(table, index string) *iradix.Txn {
if txn.modified == nil {
txn.modified = make(map[tableIndex]*iradix.Txn)
}
// Look for existing transaction
key := tableIndex{table, index}
exist, ok := txn.modified[key]
if ok {
return exist
}
// Start a new transaction
path := indexPath(table, index)
raw, _ := txn.rootTxn.Get(path)
indexTxn := raw.(*iradix.Tree).Txn()
// If we are the primary DB, enable mutation tracking. Snapshots should
// not notify, otherwise we will trigger watches on the primary DB when
// the writes will not be visible.
indexTxn.TrackMutate(txn.db.primary)
// Keep this open for the duration of the txn
txn.modified[key] = indexTxn
return indexTxn
}
// Abort is used to cancel this transaction.
// This is a noop for read transactions.
func (txn *Txn) Abort() {
// Noop for a read transaction
if !txn.write {
return
}
// Check if already aborted or committed
if txn.rootTxn == nil {
return
}
// Clear the txn
txn.rootTxn = nil
txn.modified = nil
txn.changes = nil
// Release the writer lock since this is invalid
txn.db.writer.Unlock()
}
// Commit is used to finalize this transaction.
// This is a noop for read transactions.
func (txn *Txn) Commit() {
// Noop for a read transaction
if !txn.write {
return
}
// Check if already aborted or committed
if txn.rootTxn == nil {
return
}
// Commit each sub-transaction scoped to (table, index)
for key, subTxn := range txn.modified {
path := indexPath(key.Table, key.Index)
final := subTxn.CommitOnly()
txn.rootTxn.Insert(path, final)
}
// Update the root of the DB
newRoot := txn.rootTxn.CommitOnly()
atomic.StorePointer(&txn.db.root, unsafe.Pointer(newRoot))
// Now issue all of the mutation updates (this is safe to call
// even if mutation tracking isn't enabled); we do this after
// the root pointer is swapped so that waking responders will
// see the new state.
for _, subTxn := range txn.modified {
subTxn.Notify()
}
txn.rootTxn.Notify()
// Clear the txn
txn.rootTxn = nil
txn.modified = nil
// Release the writer lock since this is invalid
txn.db.writer.Unlock()
// Run the deferred functions, if any
for i := len(txn.after); i > 0; i-- {
fn := txn.after[i-1]
fn()
}
}
// Insert is used to add or update an object into the given table
func (txn *Txn) Insert(table string, obj interface{}) error {
if !txn.write {
return fmt.Errorf("cannot insert in read-only transaction")
}
// Get the table schema
tableSchema, ok := txn.db.schema.Tables[table]
if !ok {
return fmt.Errorf("invalid table '%s'", table)
}
// Get the primary ID of the object
idSchema := tableSchema.Indexes[id]
idIndexer := idSchema.Indexer.(SingleIndexer)
ok, idVal, err := idIndexer.FromObject(obj)
if err != nil {
return fmt.Errorf("failed to build primary index: %v", err)
}
if !ok {
return fmt.Errorf("object missing primary index")
}
// Lookup the object by ID first, to see if this is an update
idTxn := txn.writableIndex(table, id)
existing, update := idTxn.Get(idVal)
// On an update, there is an existing object with the given
// primary ID. We do the update by deleting the current object
// and inserting the new object.
for name, indexSchema := range tableSchema.Indexes {
indexTxn := txn.writableIndex(table, name)
// Determine the new index value
var (
ok bool
vals [][]byte
err error
)
switch indexer := indexSchema.Indexer.(type) {
case SingleIndexer:
var val []byte
ok, val, err = indexer.FromObject(obj)
vals = [][]byte{val}
case MultiIndexer:
ok, vals, err = indexer.FromObject(obj)
}
if err != nil {
return fmt.Errorf("failed to build index '%s': %v", name, err)
}
// Handle non-unique index by computing a unique index.
// This is done by appending the primary key which must
// be unique anyways.
if ok && !indexSchema.Unique {
for i := range vals {
vals[i] = append(vals[i], idVal...)
}
}
// Handle the update by deleting from the index first
if update {
var (
okExist bool
valsExist [][]byte
err error
)
switch indexer := indexSchema.Indexer.(type) {
case SingleIndexer:
var valExist []byte
okExist, valExist, err = indexer.FromObject(existing)
valsExist = [][]byte{valExist}
case MultiIndexer:
okExist, valsExist, err = indexer.FromObject(existing)
}
if err != nil {
return fmt.Errorf("failed to build index '%s': %v", name, err)
}
if okExist {
for i, valExist := range valsExist {
// Handle non-unique index by computing a unique index.
// This is done by appending the primary key which must
// be unique anyways.
if !indexSchema.Unique {
valExist = append(valExist, idVal...)
}
// If we are writing to the same index with the same value,
// we can avoid the delete as the insert will overwrite the
// value anyways.
if i >= len(vals) || !bytes.Equal(valExist, vals[i]) {
indexTxn.Delete(valExist)
}
}
}
}
// If there is no index value, either this is an error or an expected
// case and we can skip updating
if !ok {
if indexSchema.AllowMissing {
continue
} else {
return fmt.Errorf("missing value for index '%s'", name)
}
}
// Update the value of the index
for _, val := range vals {
indexTxn.Insert(val, obj)
}
}
if txn.changes != nil {
txn.changes = append(txn.changes, Change{
Table: table,
Before: existing, // might be nil on a create
After: obj,
primaryKey: idVal,
})
}
return nil
}
// Delete is used to delete a single object from the given table
// This object must already exist in the table
func (txn *Txn) Delete(table string, obj interface{}) error {
if !txn.write {
return fmt.Errorf("cannot delete in read-only transaction")
}
// Get the table schema
tableSchema, ok := txn.db.schema.Tables[table]
if !ok {
return fmt.Errorf("invalid table '%s'", table)
}
// Get the primary ID of the object
idSchema := tableSchema.Indexes[id]
idIndexer := idSchema.Indexer.(SingleIndexer)
ok, idVal, err := idIndexer.FromObject(obj)
if err != nil {
return fmt.Errorf("failed to build primary index: %v", err)
}
if !ok {
return fmt.Errorf("object missing primary index")
}
// Lookup the object by ID first, check fi we should continue
idTxn := txn.writableIndex(table, id)
existing, ok := idTxn.Get(idVal)
if !ok {
return ErrNotFound
}
// Remove the object from all the indexes
for name, indexSchema := range tableSchema.Indexes {
indexTxn := txn.writableIndex(table, name)
// Handle the update by deleting from the index first
var (
ok bool
vals [][]byte
err error
)
switch indexer := indexSchema.Indexer.(type) {
case SingleIndexer:
var val []byte
ok, val, err = indexer.FromObject(existing)
vals = [][]byte{val}
case MultiIndexer:
ok, vals, err = indexer.FromObject(existing)
}
if err != nil {
return fmt.Errorf("failed to build index '%s': %v", name, err)
}
if ok {
// Handle non-unique index by computing a unique index.
// This is done by appending the primary key which must
// be unique anyways.
for _, val := range vals {
if !indexSchema.Unique {
val = append(val, idVal...)
}
indexTxn.Delete(val)
}
}
}
if txn.changes != nil {
txn.changes = append(txn.changes, Change{
Table: table,
Before: existing,
After: nil, // Now nil indicates deletion
primaryKey: idVal,
})
}
return nil
}
// DeletePrefix is used to delete an entire subtree based on a prefix.
// The given index must be a prefix index, and will be used to perform a scan and enumerate the set of objects to delete.
// These will be removed from all other indexes, and then a special prefix operation will delete the objects from the given index in an efficient subtree delete operation.
// This is useful when you have a very large number of objects indexed by the given index, along with a much smaller number of entries in the other indexes for those objects.
func (txn *Txn) DeletePrefix(table string, prefix_index string, prefix string) (bool, error) {
if !txn.write {
return false, fmt.Errorf("cannot delete in read-only transaction")
}
if !strings.HasSuffix(prefix_index, "_prefix") {
return false, fmt.Errorf("Index name for DeletePrefix must be a prefix index, Got %v ", prefix_index)
}
deletePrefixIndex := strings.TrimSuffix(prefix_index, "_prefix")
// Get an iterator over all of the keys with the given prefix.
entries, err := txn.Get(table, prefix_index, prefix)
if err != nil {
return false, fmt.Errorf("failed kvs lookup: %s", err)
}
// Get the table schema
tableSchema, ok := txn.db.schema.Tables[table]
if !ok {
return false, fmt.Errorf("invalid table '%s'", table)
}
foundAny := false
for entry := entries.Next(); entry != nil; entry = entries.Next() {
if !foundAny {
foundAny = true
}
// Get the primary ID of the object
idSchema := tableSchema.Indexes[id]
idIndexer := idSchema.Indexer.(SingleIndexer)
ok, idVal, err := idIndexer.FromObject(entry)
if err != nil {
return false, fmt.Errorf("failed to build primary index: %v", err)
}
if !ok {
return false, fmt.Errorf("object missing primary index")
}
if txn.changes != nil {
// Record the deletion
idTxn := txn.writableIndex(table, id)
existing, ok := idTxn.Get(idVal)
if ok {
txn.changes = append(txn.changes, Change{
Table: table,
Before: existing,
After: nil, // Now nil indicates deletion
primaryKey: idVal,
})
}
}
// Remove the object from all the indexes except the given prefix index
for name, indexSchema := range tableSchema.Indexes {
if name == deletePrefixIndex {
continue
}
indexTxn := txn.writableIndex(table, name)
// Handle the update by deleting from the index first
var (
ok bool
vals [][]byte
err error
)
switch indexer := indexSchema.Indexer.(type) {
case SingleIndexer:
var val []byte
ok, val, err = indexer.FromObject(entry)
vals = [][]byte{val}
case MultiIndexer:
ok, vals, err = indexer.FromObject(entry)
}
if err != nil {
return false, fmt.Errorf("failed to build index '%s': %v", name, err)
}
if ok {
// Handle non-unique index by computing a unique index.
// This is done by appending the primary key which must
// be unique anyways.
for _, val := range vals {
if !indexSchema.Unique {
val = append(val, idVal...)
}
indexTxn.Delete(val)
}
}
}
}
if foundAny {
indexTxn := txn.writableIndex(table, deletePrefixIndex)
ok = indexTxn.DeletePrefix([]byte(prefix))
if !ok {
panic(fmt.Errorf("prefix %v matched some entries but DeletePrefix did not delete any ", prefix))
}
return true, nil
}
return false, nil
}
// DeleteAll is used to delete all the objects in a given table
// matching the constraints on the index
func (txn *Txn) DeleteAll(table, index string, args ...interface{}) (int, error) {
if !txn.write {
return 0, fmt.Errorf("cannot delete in read-only transaction")
}
// Get all the objects
iter, err := txn.Get(table, index, args...)
if err != nil {
return 0, err
}
// Put them into a slice so there are no safety concerns while actually
// performing the deletes
var objs []interface{}
for {
obj := iter.Next()
if obj == nil {
break
}
objs = append(objs, obj)
}
// Do the deletes
num := 0
for _, obj := range objs {
if err := txn.Delete(table, obj); err != nil {
return num, err
}
num++
}
return num, nil
}
// FirstWatch is used to return the first matching object for
// the given constraints on the index along with the watch channel
func (txn *Txn) FirstWatch(table, index string, args ...interface{}) (<-chan struct{}, interface{}, error) {
// Get the index value
indexSchema, val, err := txn.getIndexValue(table, index, args...)
if err != nil {
return nil, nil, err
}
// Get the index itself
indexTxn := txn.readableIndex(table, indexSchema.Name)
// Do an exact lookup
if indexSchema.Unique && val != nil && indexSchema.Name == index {
watch, obj, ok := indexTxn.GetWatch(val)
if !ok {
return watch, nil, nil
}
return watch, obj, nil
}
// Handle non-unique index by using an iterator and getting the first value
iter := indexTxn.Root().Iterator()
watch := iter.SeekPrefixWatch(val)
_, value, _ := iter.Next()
return watch, value, nil
}
// First is used to return the first matching object for
// the given constraints on the index
func (txn *Txn) First(table, index string, args ...interface{}) (interface{}, error) {
_, val, err := txn.FirstWatch(table, index, args...)
return val, err
}
// LongestPrefix is used to fetch the longest prefix match for the given
// constraints on the index. Note that this will not work with the memdb
// StringFieldIndex because it adds null terminators which prevent the
// algorithm from correctly finding a match (it will get to right before the
// null and fail to find a leaf node). This should only be used where the prefix
// given is capable of matching indexed entries directly, which typically only
// applies to a custom indexer. See the unit test for an example.
func (txn *Txn) LongestPrefix(table, index string, args ...interface{}) (interface{}, error) {
// Enforce that this only works on prefix indexes.
if !strings.HasSuffix(index, "_prefix") {
return nil, fmt.Errorf("must use '%s_prefix' on index", index)
}
// Get the index value.
indexSchema, val, err := txn.getIndexValue(table, index, args...)
if err != nil {
return nil, err
}
// This algorithm only makes sense against a unique index, otherwise the
// index keys will have the IDs appended to them.
if !indexSchema.Unique {
return nil, fmt.Errorf("index '%s' is not unique", index)
}
// Find the longest prefix match with the given index.
indexTxn := txn.readableIndex(table, indexSchema.Name)
if _, value, ok := indexTxn.Root().LongestPrefix(val); ok {
return value, nil
}
return nil, nil
}
// getIndexValue is used to get the IndexSchema and the value
// used to scan the index given the parameters. This handles prefix based
// scans when the index has the "_prefix" suffix. The index must support
// prefix iteration.
func (txn *Txn) getIndexValue(table, index string, args ...interface{}) (*IndexSchema, []byte, error) {
// Get the table schema
tableSchema, ok := txn.db.schema.Tables[table]
if !ok {
return nil, nil, fmt.Errorf("invalid table '%s'", table)
}
// Check for a prefix scan
prefixScan := false
if strings.HasSuffix(index, "_prefix") {
index = strings.TrimSuffix(index, "_prefix")
prefixScan = true
}
// Get the index schema
indexSchema, ok := tableSchema.Indexes[index]
if !ok {
return nil, nil, fmt.Errorf("invalid index '%s'", index)
}
// Hot-path for when there are no arguments
if len(args) == 0 {
return indexSchema, nil, nil
}
// Special case the prefix scanning
if prefixScan {
prefixIndexer, ok := indexSchema.Indexer.(PrefixIndexer)
if !ok {
return indexSchema, nil,
fmt.Errorf("index '%s' does not support prefix scanning", index)
}
val, err := prefixIndexer.PrefixFromArgs(args...)
if err != nil {
return indexSchema, nil, fmt.Errorf("index error: %v", err)
}
return indexSchema, val, err
}
// Get the exact match index
val, err := indexSchema.Indexer.FromArgs(args...)
if err != nil {
return indexSchema, nil, fmt.Errorf("index error: %v", err)
}
return indexSchema, val, err
}
// ResultIterator is used to iterate over a list of results
// from a Get query on a table.
type ResultIterator interface {
WatchCh() <-chan struct{}
Next() interface{}
}
// Get is used to construct a ResultIterator over all the
// rows that match the given constraints of an index.
func (txn *Txn) Get(table, index string, args ...interface{}) (ResultIterator, error) {
indexIter, val, err := txn.getIndexIterator(table, index, args...)
if err != nil {
return nil, err
}
// Seek the iterator to the appropriate sub-set
watchCh := indexIter.SeekPrefixWatch(val)
// Create an iterator
iter := &radixIterator{
iter: indexIter,
watchCh: watchCh,
}
return iter, nil
}
// LowerBound is used to construct a ResultIterator over all the the range of
// rows that have an index value greater than or equal to the provide args.
// Calling this then iterating until the rows are larger than required allows
// range scans within an index. It is not possible to watch the resulting
// iterator since the radix tree doesn't efficiently allow watching on lower
// bound changes. The WatchCh returned will be nill and so will block forever.
func (txn *Txn) LowerBound(table, index string, args ...interface{}) (ResultIterator, error) {
indexIter, val, err := txn.getIndexIterator(table, index, args...)
if err != nil {
return nil, err
}
// Seek the iterator to the appropriate sub-set
indexIter.SeekLowerBound(val)
// Create an iterator
iter := &radixIterator{
iter: indexIter,
}
return iter, nil
}
// objectID is a tuple of table name and the raw internal id byte slice
// converted to a string. It's only converted to a string to make it comparable
// so this struct can be used as a map index.
type objectID struct {
Table string
IndexVal string
}
// mutInfo stores metadata about mutations to allow collapsing multiple
// mutations to the same object into one.
type mutInfo struct {
firstBefore interface{}
lastIdx int
}
// Changes returns the set of object changes that have been made in the
// transaction so far. If change tracking is not enabled it wil always return
// nil. It can be called before or after Commit. If it is before Commit it will
// return all changes made so far which may not be the same as the final
// Changes. After abort it will always return nil. As with other Txn methods
// it's not safe to call this from a different goroutine than the one making
// mutations or committing the transaction. Mutations will appear in the order
// they were performed in the transaction but multiple operations to the same
// object will be collapsed so only the effective overall change to that object
// is present. If transaction operations are dependent (e.g. copy object X to Y
// then delete X) this might mean the set of mutations is incomplete to verify
// history, but it is complete in that the net effect is preserved (Y got a new
// value, X got removed).
func (txn *Txn) Changes() Changes {
if txn.changes == nil {
return nil
}
// De-duplicate mutations by key so all take effect at the point of the last
// write but we keep the mutations in order.
dups := make(map[objectID]mutInfo)
for i, m := range txn.changes {
oid := objectID{
Table: m.Table,
IndexVal: string(m.primaryKey),
}
// Store the latest mutation index for each key value
mi, ok := dups[oid]
if !ok {
// First entry for key, store the before value
mi.firstBefore = m.Before
}
mi.lastIdx = i
dups[oid] = mi
}
if len(dups) == len(txn.changes) {
// No duplicates found, fast path return it as is
return txn.changes
}
// Need to remove the duplicates
cs := make(Changes, 0, len(dups))
for i, m := range txn.changes {
oid := objectID{
Table: m.Table,
IndexVal: string(m.primaryKey),
}
mi := dups[oid]
if mi.lastIdx == i {
// This was the latest value for this key copy it with the before value in
// case it's different. Note that m is not a pointer so we are not
// modifying the txn.changeSet here - it's already a copy.
m.Before = mi.firstBefore
// Edge case - if the object was inserted and then eventually deleted in
// the same transaction, then the net affect on that key is a no-op. Don't
// emit a mutation with nil for before and after as it's meaningless and
// might violate expectations and cause a panic in code that assumes at
// least one must be set.
if m.Before == nil && m.After == nil {
continue
}
cs = append(cs, m)
}
}
// Store the de-duped version in case this is called again
txn.changes = cs
return cs
}
func (txn *Txn) getIndexIterator(table, index string, args ...interface{}) (*iradix.Iterator, []byte, error) {
// Get the index value to scan
indexSchema, val, err := txn.getIndexValue(table, index, args...)
if err != nil {
return nil, nil, err
}
// Get the index itself
indexTxn := txn.readableIndex(table, indexSchema.Name)
indexRoot := indexTxn.Root()
// Get an interator over the index
indexIter := indexRoot.Iterator()
return indexIter, val, nil
}
// Defer is used to push a new arbitrary function onto a stack which
// gets called when a transaction is committed and finished. Deferred
// functions are called in LIFO order, and only invoked at the end of
// write transactions.
func (txn *Txn) Defer(fn func()) {
txn.after = append(txn.after, fn)
}
// radixIterator is used to wrap an underlying iradix iterator.
// This is much more efficient than a sliceIterator as we are not
// materializing the entire view.
type radixIterator struct {
iter *iradix.Iterator
watchCh <-chan struct{}
}
func (r *radixIterator) WatchCh() <-chan struct{} {
return r.watchCh
}
func (r *radixIterator) Next() interface{} {
_, value, ok := r.iter.Next()
if !ok {
return nil
}
return value
}
// Snapshot creates a snapshot of the current state of the transaction.
// Returns a new read-only transaction or nil if the transaction is already
// aborted or committed.
func (txn *Txn) Snapshot() *Txn {
if txn.rootTxn == nil {
return nil
}
snapshot := &Txn{
db: txn.db,
rootTxn: txn.rootTxn.Clone(),
}
// Commit sub-transactions into the snapshot
for key, subTxn := range txn.modified {
path := indexPath(key.Table, key.Index)
final := subTxn.CommitOnly()
snapshot.rootTxn.Insert(path, final)
}
return snapshot
}
|