bolt/tx.go

package bolt

import (
	"errors"
	"fmt"
	"sort"
	"time"
	"unsafe"
)

var (
	// ErrTxNotWritable is returned when performing a write operation on a
	// read-only transaction.
	ErrTxNotWritable = errors.New("tx not writable")

	// ErrTxClosed is returned when committing or rolling back a transaction
	// that has already been committed or rolled back.
	ErrTxClosed = errors.New("tx closed")
)

// txid represents the internal transaction identifier.
type txid uint64

// Tx represents a read-only or read/write transaction on the database.
// Read-only transactions can be used for retrieving values for keys and creating cursors.
// Read/write transactions can create and remove buckets and create and remove keys.
//
// IMPORTANT: You must commit or rollback transactions when you are done with
// them. Pages can not be reclaimed by the writer until no more transactions
// are using them. A long running read transaction can cause the database to
// quickly grow.
type Tx struct {
	writable       bool
	managed        bool
	db             *DB
	meta           *meta
	root           Bucket
	pages          map[pgid]*page
	stats          TxStats
	commitHandlers []func()
}

// init initializes the transaction.
func (tx *Tx) init(db *DB) {
	tx.db = db
	tx.pages = nil

	// Copy the meta page since it can be changed by the writer.
	tx.meta = &meta{}
	db.meta().copy(tx.meta)

	// Copy over the root bucket.
	tx.root = newBucket(tx)
	tx.root.bucket = &bucket{}
	*tx.root.bucket = tx.meta.root

	// Increment the transaction id and add a page cache for writable transactions.
	if tx.writable {
		tx.pages = make(map[pgid]*page)
		tx.meta.txid += txid(1)
	}
}

// id returns the transaction id.
func (tx *Tx) id() txid {
	return tx.meta.txid
}

// DB returns a reference to the database that created the transaction.
func (tx *Tx) DB() *DB {
	return tx.db
}

// Writable returns whether the transaction can perform write operations.
func (tx *Tx) Writable() bool {
	return tx.writable
}

// Cursor creates a cursor associated with the root bucket.
// All items in the cursor will return a nil value because all root bucket keys point to buckets.
// The cursor is only valid as long as the transaction is open.
// Do not use a cursor after the transaction is closed.
func (tx *Tx) Cursor() *Cursor {
	return tx.root.Cursor()
}

// Stats retrieves a copy of the current transaction statistics.
func (tx *Tx) Stats() TxStats {
	return tx.stats
}

// Bucket retrieves a bucket by name.
// Returns nil if the bucket does not exist.
func (tx *Tx) Bucket(name []byte) *Bucket {
	return tx.root.Bucket(name)
}

// CreateBucket creates a new bucket.
// Returns an error if the bucket already exists, if the bucket name is blank, or if the bucket name is too long.
func (tx *Tx) CreateBucket(name []byte) (*Bucket, error) {
	return tx.root.CreateBucket(name)
}

// CreateBucketIfNotExists creates a new bucket if it doesn't already exist.
// Returns an error if the bucket name is blank, or if the bucket name is too long.
func (tx *Tx) CreateBucketIfNotExists(name []byte) (*Bucket, error) {
	return tx.root.CreateBucketIfNotExists(name)
}

// DeleteBucket deletes a bucket.
// Returns an error if the bucket cannot be found or if the key represents a non-bucket value.
func (tx *Tx) DeleteBucket(name []byte) error {
	return tx.root.DeleteBucket(name)
}

// ForEach executes a function for each bucket in the root.
// If the provided function returns an error then the iteration is stopped and
// the error is returned to the caller.
func (tx *Tx) ForEach(fn func(name []byte, b *Bucket) error) error {
	return tx.root.ForEach(func(k, v []byte) error {
		if err := fn(k, tx.root.Bucket(k)); err != nil {
			return err
		}
		return nil
	})
}

// OnCommit adds a handler function to be executed after the transaction successfully commits.
func (tx *Tx) OnCommit(fn func()) {
	tx.commitHandlers = append(tx.commitHandlers, fn)
}

// Commit writes all changes to disk and updates the meta page.
// Returns an error if a disk write error occurs.
func (tx *Tx) Commit() error {
	_assert(!tx.managed, "managed tx commit not allowed")
	if tx.db == nil {
		return ErrTxClosed
	} else if !tx.writable {
		return ErrTxNotWritable
	}

	// TODO(benbjohnson): Use vectorized I/O to write out dirty pages.

	// Rebalance nodes which have had deletions.
	var startTime = time.Now()
	tx.root.rebalance()
	tx.stats.RebalanceTime += time.Since(startTime)

	// spill data onto dirty pages.
	startTime = time.Now()
	if err := tx.root.spill(); err != nil {
		tx.close()
		return err
	}
	tx.stats.SpillTime += time.Since(startTime)

	// Free the old root bucket.
	tx.meta.root.root = tx.root.root

	// Free the freelist and allocate new pages for it. This will overestimate
	// the size of the freelist but not underestimate the size (which would be bad).
	tx.db.freelist.free(tx.id(), tx.db.page(tx.meta.freelist))
	p, err := tx.allocate((tx.db.freelist.size() / tx.db.pageSize) + 1)
	if err != nil {
		tx.close()
		return err
	}
	tx.db.freelist.write(p)
	tx.meta.freelist = p.id

	// Write dirty pages to disk.
	startTime = time.Now()
	if err := tx.write(); err != nil {
		tx.close()
		return err
	}

	// If strict mode is enabled then perform a consistency check.
	if tx.db.StrictMode {
		if err := tx.Check(); err != nil {
			err := err.(ErrorList)
			panic("check fail: " + err.Error() + ": " + err.join("; "))
		}
	}

	// Write meta to disk.
	if err := tx.writeMeta(); err != nil {
		tx.close()
		return err
	}
	tx.stats.WriteTime += time.Since(startTime)

	// Finalize the transaction.
	tx.close()

	// Execute commit handlers now that the locks have been removed.
	for _, fn := range tx.commitHandlers {
		fn()
	}

	return nil
}

// Rollback closes the transaction and ignores all previous updates.
func (tx *Tx) Rollback() error {
	_assert(!tx.managed, "managed tx rollback not allowed")
	if tx.db == nil {
		return ErrTxClosed
	}
	tx.close()
	return nil
}

func (tx *Tx) close() {
	if tx.writable {
		// Remove writer lock.
		tx.db.rwlock.Unlock()

		// Merge statistics.
		tx.db.statlock.Lock()
		tx.db.stats.TxStats.add(&tx.stats)
		tx.db.statlock.Unlock()
	} else {
		tx.db.removeTx(tx)
	}
	tx.db = nil
}

// Check performs several consistency checks on the database for this transaction.
// An error is returned if any inconsistency is found or if executed on a read-only transaction.
func (tx *Tx) Check() error {
	if !tx.writable {
		return ErrTxNotWritable
	}

	var errors ErrorList

	// Check if any pages are double freed.
	freed := make(map[pgid]bool)
	for _, id := range tx.db.freelist.all() {
		if freed[id] {
			errors = append(errors, fmt.Errorf("page %d: already freed", id))
		}
		freed[id] = true
	}

	// Track every reachable page.
	reachable := make(map[pgid]*page)
	reachable[0] = tx.page(0) // meta0
	reachable[1] = tx.page(1) // meta1
	for i := uint32(0); i <= tx.page(tx.meta.freelist).overflow; i++ {
		reachable[tx.meta.freelist+pgid(i)] = tx.page(tx.meta.freelist)
	}

	// Recursively check buckets.
	tx.checkBucket(&tx.root, reachable, &errors)

	// Ensure all pages below high water mark are either reachable or freed.
	for i := pgid(0); i < tx.meta.pgid; i++ {
		_, isReachable := reachable[i]
		if !isReachable && !freed[i] {
			errors = append(errors, fmt.Errorf("page %d: unreachable unfreed", int(i)))
		} else if isReachable && freed[i] {
			errors = append(errors, fmt.Errorf("page %d: reachable freed", int(i)))
		}
	}

	if len(errors) > 0 {
		return errors
	}

	return nil
}

func (tx *Tx) checkBucket(b *Bucket, reachable map[pgid]*page, errors *ErrorList) {
	// Ignore inline buckets.
	if b.root == 0 {
		return
	}

	// Check every page used by this bucket.
	b.tx.forEachPage(b.root, 0, func(p *page, _ int) {
		// Ensure each page is only referenced once.
		for i := pgid(0); i <= pgid(p.overflow); i++ {
			var id = p.id + i
			if _, ok := reachable[id]; ok {
				*errors = append(*errors, fmt.Errorf("page %d: multiple references", int(id)))
			}
			reachable[id] = p
		}

		// Retrieve page info.
		info, err := b.tx.Page(int(p.id))
		if err != nil {
			*errors = append(*errors, err)
		} else if info == nil {
			*errors = append(*errors, fmt.Errorf("page %d: out of bounds: %d", int(p.id), int(b.tx.meta.pgid)))
		} else if info.Type != "branch" && info.Type != "leaf" {
			*errors = append(*errors, fmt.Errorf("page %d: invalid type: %s", int(p.id), info.Type))
		}
	})

	// Check each bucket within this bucket.
	_ = b.ForEach(func(k, v []byte) error {
		if child := b.Bucket(k); child != nil {
			tx.checkBucket(child, reachable, errors)
		}
		return nil
	})
}

// allocate returns a contiguous block of memory starting at a given page.
func (tx *Tx) allocate(count int) (*page, error) {
	p, err := tx.db.allocate(count)
	if err != nil {
		return nil, err
	}

	// Save to our page cache.
	tx.pages[p.id] = p

	// Update statistics.
	tx.stats.PageCount++
	tx.stats.PageAlloc += count * tx.db.pageSize

	return p, nil
}

// write writes any dirty pages to disk.
func (tx *Tx) write() error {
	// Sort pages by id.
	pages := make(pages, 0, len(tx.pages))
	for _, p := range tx.pages {
		pages = append(pages, p)
	}
	sort.Sort(pages)

	// Write pages to disk in order.
	for _, p := range pages {
		size := (int(p.overflow) + 1) * tx.db.pageSize
		buf := (*[maxAllocSize]byte)(unsafe.Pointer(p))[:size]
		offset := int64(p.id) * int64(tx.db.pageSize)
		if _, err := tx.db.ops.writeAt(buf, offset); err != nil {
			return err
		}

		// Update statistics.
		tx.stats.Write++
	}
	if err := fdatasync(tx.db.file); err != nil {
		return err
	}

	// Clear out page cache.
	tx.pages = make(map[pgid]*page)

	return nil
}

// writeMeta writes the meta to the disk.
func (tx *Tx) writeMeta() error {
	// Create a temporary buffer for the meta page.
	buf := make([]byte, tx.db.pageSize)
	p := tx.db.pageInBuffer(buf, 0)
	tx.meta.write(p)

	// Write the meta page to file.
	if _, err := tx.db.ops.writeAt(buf, int64(p.id)*int64(tx.db.pageSize)); err != nil {
		return err
	}
	if err := fdatasync(tx.db.file); err != nil {
		return err
	}

	// Update statistics.
	tx.stats.Write++

	return nil
}

// page returns a reference to the page with a given id.
// If page has been written to then a temporary bufferred page is returned.
func (tx *Tx) page(id pgid) *page {
	// Check the dirty pages first.
	if tx.pages != nil {
		if p, ok := tx.pages[id]; ok {
			return p
		}
	}

	// Otherwise return directly from the mmap.
	return tx.db.page(id)
}

// forEachPage iterates over every page within a given page and executes a function.
func (tx *Tx) forEachPage(pgid pgid, depth int, fn func(*page, int)) {
	p := tx.page(pgid)

	// Execute function.
	fn(p, depth)

	// Recursively loop over children.
	if (p.flags & branchPageFlag) != 0 {
		for i := 0; i < int(p.count); i++ {
			elem := p.branchPageElement(uint16(i))
			tx.forEachPage(elem.pgid, depth+1, fn)
		}
	}
}

// Page returns page information for a given page number.
// This is only available from writable transactions.
func (tx *Tx) Page(id int) (*PageInfo, error) {
	if tx.db == nil {
		return nil, ErrTxClosed
	} else if !tx.writable {
		return nil, ErrTxNotWritable
	} else if pgid(id) >= tx.meta.pgid {
		return nil, nil
	}

	// Build the page info.
	p := tx.db.page(pgid(id))
	info := &PageInfo{
		ID:            id,
		Count:         int(p.count),
		OverflowCount: int(p.overflow),
	}

	// Determine the type (or if it's free).
	if tx.db.freelist.isFree(pgid(id)) {
		info.Type = "free"
	} else {
		info.Type = p.typ()
	}

	return info, nil
}

// TxStats represents statistics about the actions performed by the transaction.
type TxStats struct {
	// Page statistics.
	PageCount int // number of page allocations
	PageAlloc int // total bytes allocated

	// Cursor statistics.
	CursorCount int // number of cursors created

	// Node statistics
	NodeCount int // number of node allocations
	NodeDeref int // number of node dereferences

	// Rebalance statistics.
	Rebalance     int           // number of node rebalances
	RebalanceTime time.Duration // total time spent rebalancing

	// Split/Spill statistics.
	Split     int           // number of nodes split
	Spill     int           // number of nodes spilled
	SpillTime time.Duration // total time spent spilling

	// Write statistics.
	Write     int           // number of writes performed
	WriteTime time.Duration // total time spent writing to disk
}

func (s *TxStats) add(other *TxStats) {
	s.PageCount += other.PageCount
	s.PageAlloc += other.PageAlloc
	s.CursorCount += other.CursorCount
	s.NodeCount += other.NodeCount
	s.NodeDeref += other.NodeDeref
	s.Rebalance += other.Rebalance
	s.RebalanceTime += other.RebalanceTime
	s.Split += other.Split
	s.Spill += other.Spill
	s.SpillTime += other.SpillTime
	s.Write += other.Write
	s.WriteTime += other.WriteTime
}

// Sub calculates and returns the difference between two sets of transaction stats.
// This is useful when obtaining stats at two different points and time and
// you need the performance counters that occurred within that time span.
func (s *TxStats) Sub(other *TxStats) TxStats {
	var diff TxStats
	diff.PageCount = s.PageCount - other.PageCount
	diff.PageAlloc = s.PageAlloc - other.PageAlloc
	diff.CursorCount = s.CursorCount - other.CursorCount
	diff.NodeCount = s.NodeCount - other.NodeCount
	diff.NodeDeref = s.NodeDeref - other.NodeDeref
	diff.Rebalance = s.Rebalance - other.Rebalance
	diff.RebalanceTime = s.RebalanceTime - other.RebalanceTime
	diff.Split = s.Split - other.Split
	diff.Spill = s.Spill - other.Spill
	diff.SpillTime = s.SpillTime - other.SpillTime
	diff.Write = s.Write - other.Write
	diff.WriteTime = s.WriteTime - other.WriteTime
	return diff
}