efa08d4081
the relatively small batch size of 128 was chosen so that if we have a set of keys like a/1 a/2 ... a/100000 b c list operations would not have to walk 100k keys inside of a/ before skipping to b. unfortunately, iteration is also used by the metainfo loop. in that case, it's doing a recursive listing, and so there's no need to skip large prefixes. thus, we can use a bigger batch size when recursive listing is requested. Change-Id: I87cf1ba385b6eb2928c5b7cc5e0f7a8c7bd126d9
281 lines
6.6 KiB
Go
281 lines
6.6 KiB
Go
// Copyright (C) 2019 Storj Labs, Inc.
|
|
// See LICENSE for copying information.
|
|
|
|
package cockroachkv
|
|
|
|
import (
|
|
"bytes"
|
|
"context"
|
|
"database/sql"
|
|
|
|
"github.com/cockroachdb/cockroach-go/crdb"
|
|
"github.com/lib/pq"
|
|
"github.com/zeebo/errs"
|
|
"gopkg.in/spacemonkeygo/monkit.v2"
|
|
|
|
"storj.io/storj/private/dbutil"
|
|
"storj.io/storj/storage"
|
|
"storj.io/storj/storage/cockroachkv/schema"
|
|
)
|
|
|
|
const (
|
|
defaultBatchSize = 128
|
|
defaultRecursiveBatchSize = 10000
|
|
)
|
|
|
|
var (
|
|
mon = monkit.Package()
|
|
)
|
|
|
|
// Client is the entrypoint into a cockroachkv data store
|
|
type Client struct {
|
|
db *sql.DB
|
|
}
|
|
|
|
// New instantiates a new cockroachkv client given db URL
|
|
func New(dbURL string) (*Client, error) {
|
|
db, err := sql.Open("cockroach", dbURL)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
dbutil.Configure(db, mon)
|
|
|
|
err = schema.PrepareDB(db)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
return NewWith(db), nil
|
|
}
|
|
|
|
// NewWith instantiates a new postgreskv client given db.
|
|
func NewWith(db *sql.DB) *Client {
|
|
return &Client{db: db}
|
|
}
|
|
|
|
// Close closes the client
|
|
func (client *Client) Close() error {
|
|
return client.db.Close()
|
|
}
|
|
|
|
// Put sets the value for the provided key.
|
|
func (client *Client) Put(ctx context.Context, key storage.Key, value storage.Value) (err error) {
|
|
defer mon.Task()(&ctx)(&err)
|
|
|
|
if key.IsZero() {
|
|
return storage.ErrEmptyKey.New("")
|
|
}
|
|
|
|
q := `
|
|
INSERT INTO pathdata (fullpath, metadata)
|
|
VALUES ($1:::BYTEA, $2:::BYTEA)
|
|
ON CONFLICT (fullpath) DO UPDATE SET metadata = EXCLUDED.metadata
|
|
`
|
|
|
|
_, err = client.db.Exec(q, []byte(key), []byte(value))
|
|
return Error.Wrap(err)
|
|
}
|
|
|
|
// Get looks up the provided key and returns its value (or an error).
|
|
func (client *Client) Get(ctx context.Context, key storage.Key) (_ storage.Value, err error) {
|
|
defer mon.Task()(&ctx)(&err)
|
|
|
|
if key.IsZero() {
|
|
return nil, storage.ErrEmptyKey.New("")
|
|
}
|
|
|
|
q := "SELECT metadata FROM pathdata WHERE fullpath = $1:::BYTEA"
|
|
row := client.db.QueryRow(q, []byte(key))
|
|
|
|
var val []byte
|
|
err = row.Scan(&val)
|
|
if err == sql.ErrNoRows {
|
|
return nil, storage.ErrKeyNotFound.New("%q", key)
|
|
}
|
|
|
|
return val, Error.Wrap(err)
|
|
}
|
|
|
|
// GetAll finds all values for the provided keys (up to storage.LookupLimit).
|
|
// If more keys are provided than the maximum, an error will be returned.
|
|
func (client *Client) GetAll(ctx context.Context, keys storage.Keys) (_ storage.Values, err error) {
|
|
defer mon.Task()(&ctx)(&err)
|
|
|
|
if len(keys) > storage.LookupLimit {
|
|
return nil, storage.ErrLimitExceeded
|
|
}
|
|
|
|
q := `
|
|
SELECT metadata
|
|
FROM pathdata pd
|
|
RIGHT JOIN
|
|
unnest($1:::BYTEA[]) WITH ORDINALITY pk(request, ord)
|
|
ON (pd.fullpath = pk.request)
|
|
ORDER BY pk.ord
|
|
`
|
|
rows, err := client.db.Query(q, pq.ByteaArray(keys.ByteSlices()))
|
|
if err != nil {
|
|
return nil, errs.Wrap(err)
|
|
}
|
|
defer func() { err = errs.Combine(err, Error.Wrap(rows.Close())) }()
|
|
|
|
values := make([]storage.Value, 0, len(keys))
|
|
for rows.Next() {
|
|
var value []byte
|
|
if err := rows.Scan(&value); err != nil {
|
|
return nil, Error.Wrap(err)
|
|
}
|
|
values = append(values, storage.Value(value))
|
|
}
|
|
|
|
return values, Error.Wrap(rows.Err())
|
|
}
|
|
|
|
// Delete deletes the given key and its associated value.
|
|
func (client *Client) Delete(ctx context.Context, key storage.Key) (err error) {
|
|
defer mon.Task()(&ctx)(&err)
|
|
|
|
if key.IsZero() {
|
|
return storage.ErrEmptyKey.New("")
|
|
}
|
|
|
|
q := "DELETE FROM pathdata WHERE fullpath = $1:::BYTEA"
|
|
result, err := client.db.Exec(q, []byte(key))
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
numRows, err := result.RowsAffected()
|
|
if err != nil {
|
|
return err
|
|
}
|
|
if numRows == 0 {
|
|
return storage.ErrKeyNotFound.New("%q", key)
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// List returns either a list of known keys, in order, or an error.
|
|
func (client *Client) List(ctx context.Context, first storage.Key, limit int) (_ storage.Keys, err error) {
|
|
defer mon.Task()(&ctx)(&err)
|
|
return storage.ListKeys(ctx, client, first, limit)
|
|
}
|
|
|
|
// Iterate calls the callback with an iterator over the keys.
|
|
func (client *Client) Iterate(ctx context.Context, opts storage.IterateOptions, fn func(context.Context, storage.Iterator) error) (err error) {
|
|
defer mon.Task()(&ctx)(&err)
|
|
|
|
batchSize := defaultBatchSize
|
|
if opts.Recurse {
|
|
batchSize = defaultRecursiveBatchSize
|
|
}
|
|
|
|
opi, err := newOrderedCockroachIterator(ctx, client, opts, batchSize)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
defer func() {
|
|
err = errs.Combine(err, opi.Close())
|
|
}()
|
|
|
|
return fn(ctx, opi)
|
|
}
|
|
|
|
// CompareAndSwap atomically compares and swaps oldValue with newValue
|
|
func (client *Client) CompareAndSwap(ctx context.Context, key storage.Key, oldValue, newValue storage.Value) (err error) {
|
|
defer mon.Task()(&ctx)(&err)
|
|
|
|
if key.IsZero() {
|
|
return storage.ErrEmptyKey.New("")
|
|
}
|
|
|
|
if oldValue == nil && newValue == nil {
|
|
q := "SELECT metadata FROM pathdata WHERE fullpath = $1:::BYTEA"
|
|
row := client.db.QueryRow(q, []byte(key))
|
|
|
|
var val []byte
|
|
err = row.Scan(&val)
|
|
if err == sql.ErrNoRows {
|
|
return nil
|
|
}
|
|
|
|
if err != nil {
|
|
return Error.Wrap(err)
|
|
}
|
|
return storage.ErrValueChanged.New("%q", key)
|
|
}
|
|
|
|
if oldValue == nil {
|
|
q := `
|
|
INSERT INTO pathdata (fullpath, metadata) VALUES ($1:::BYTEA, $2:::BYTEA)
|
|
ON CONFLICT DO NOTHING
|
|
RETURNING 1
|
|
`
|
|
row := client.db.QueryRow(q, []byte(key), []byte(newValue))
|
|
|
|
var val []byte
|
|
err = row.Scan(&val)
|
|
if err == sql.ErrNoRows {
|
|
return storage.ErrValueChanged.New("%q", key)
|
|
}
|
|
return Error.Wrap(err)
|
|
}
|
|
|
|
return crdb.ExecuteTx(ctx, client.db, nil, func(txn *sql.Tx) error {
|
|
q := "SELECT metadata FROM pathdata WHERE fullpath = $1:::BYTEA;"
|
|
row := txn.QueryRowContext(ctx, q, []byte(key))
|
|
|
|
var metadata []byte
|
|
err = row.Scan(&metadata)
|
|
if err == sql.ErrNoRows {
|
|
// Row not found for this fullpath.
|
|
// Potentially because another concurrent transaction changed the row.
|
|
return storage.ErrKeyNotFound.New("%q", key)
|
|
}
|
|
if err != nil {
|
|
return Error.Wrap(err)
|
|
}
|
|
|
|
if equal := bytes.Compare(metadata, oldValue); equal != 0 {
|
|
// If the row is found but the metadata has been already changed
|
|
// we can't continue to delete it.
|
|
return storage.ErrValueChanged.New("%q", key)
|
|
}
|
|
|
|
var res sql.Result
|
|
if newValue == nil {
|
|
q = `
|
|
DELETE FROM pathdata
|
|
WHERE pathdata.fullpath = $1:::BYTEA
|
|
AND pathdata.metadata = $2:::BYTEA
|
|
`
|
|
|
|
res, err = txn.ExecContext(ctx, q, []byte(key), []byte(oldValue))
|
|
} else {
|
|
q = `
|
|
UPDATE pathdata
|
|
SET metadata = $3:::BYTEA
|
|
WHERE pathdata.fullpath = $1:::BYTEA
|
|
AND pathdata.metadata = $2:::BYTEA
|
|
`
|
|
res, err = txn.ExecContext(ctx, q, []byte(key), []byte(oldValue), []byte(newValue))
|
|
}
|
|
|
|
if err != nil {
|
|
return Error.Wrap(err)
|
|
}
|
|
|
|
affected, err := res.RowsAffected()
|
|
if err != nil {
|
|
return Error.Wrap(err)
|
|
}
|
|
|
|
if affected != 1 {
|
|
return storage.ErrValueChanged.New("%q", key)
|
|
}
|
|
|
|
return nil
|
|
})
|
|
}
|