reorg.go

// Copyright 2015 PingCAP, Inc.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// See the License for the specific language governing permissions and
// limitations under the License.

package ddl

import (
	"sync/atomic"
	"time"

	"github.com/juju/errors"
	"github.com/ngaut/log"
	"github.com/pingcap/tidb/context"
	"github.com/pingcap/tidb/kv"
	"github.com/pingcap/tidb/meta"
	"github.com/pingcap/tidb/model"
	"github.com/pingcap/tidb/mysql"
	"github.com/pingcap/tidb/terror"
	"github.com/pingcap/tidb/util/mock"
)

// newContext gets a context. It is only used for adding column in reorganization state.
func (d *ddl) newContext() context.Context {
	c := mock.NewContext()
	c.Store = d.store
	c.GetSessionVars().SetStatusFlag(mysql.ServerStatusAutocommit, false)
	return c
}

const waitReorgTimeout = 10 * time.Second

func (d *ddl) setReorgRowCount(count int64) {
	atomic.StoreInt64(&d.reorgRowCount, count)
}

func (d *ddl) getReorgRowCount() int64 {
	return atomic.LoadInt64(&d.reorgRowCount)
}

func (d *ddl) runReorgJob(job *model.Job, f func() error) error {
	if d.reorgDoneCh == nil {
		// start a reorganization job
		d.wait.Add(1)
		d.reorgDoneCh = make(chan error, 1)
		go func() {
			defer d.wait.Done()
			d.reorgDoneCh <- f()
		}()
	}

	waitTimeout := waitReorgTimeout
	// if d.lease is 0, we are using a local storage,
	// and we can wait the reorganization to be done here.
	// if d.lease > 0, we don't need to wait here because
	// we will wait 2 * lease outer and try checking again,
	// so we use a very little timeout here.
	if d.lease > 0 {
		waitTimeout = 1 * time.Millisecond
	}

	// wait reorganization job done or timeout
	select {
	case err := <-d.reorgDoneCh:
		log.Info("[ddl] run reorg job done")
		d.reorgDoneCh = nil
		// Update a job's RowCount.
		job.SetRowCount(d.getReorgRowCount())
		d.setReorgRowCount(0)
		return errors.Trace(err)
	case <-d.quitCh:
		log.Info("[ddl] run reorg job ddl quit")
		d.setReorgRowCount(0)
		// We return errWaitReorgTimeout here too, so that outer loop will break.
		return errWaitReorgTimeout
	case <-time.After(waitTimeout):
		log.Infof("[ddl] run reorg job wait timeout %v", waitTimeout)
		// Update a job's RowCount.
		job.SetRowCount(d.getReorgRowCount())
		// If timeout, we will return, check the owner and retry to wait job done again.
		return errWaitReorgTimeout
	}
}

func (d *ddl) isReorgRunnable(txn kv.Transaction, flag JobType) error {
	if d.isClosed() {
		// worker is closed, can't run reorganization.
		return errInvalidWorker.Gen("worker is closed")
	}

	if !d.isOwner(flag) {
		// If it's not the owner, we will try later, so here just returns an error.
		log.Infof("[ddl] the %s not the %s job owner, txnTS:%d", d.uuid, flag, txn.StartTS())
		return errors.Trace(errNotOwner)
	}
	return nil
}

// delKeysWithStartKey deletes keys with start key in a limited number. If limit < 0, deletes all keys.
// It returns the number of rows deleted, next start key and the error.
func (d *ddl) delKeysWithStartKey(prefix, startKey kv.Key, jobType JobType, job *model.Job, limit int) (int, kv.Key, error) {
	limitedDel := limit >= 0

	var count int
	total := job.GetRowCount()
	keys := make([]kv.Key, 0, defaultBatchCnt)
	for {
		if limitedDel && count >= limit {
			break
		}
		batch := defaultBatchCnt
		if limitedDel && count+batch > limit {
			batch = limit - count
		}
		startTS := time.Now()
		err := kv.RunInNewTxn(d.store, true, func(txn kv.Transaction) error {
			if err1 := d.isReorgRunnable(txn, jobType); err1 != nil {
				return errors.Trace(err1)
			}

			iter, err := txn.Seek(startKey)
			if err != nil {
				return errors.Trace(err)
			}
			defer iter.Close()

			for i := 0; i < batch; i++ {
				if iter.Valid() && iter.Key().HasPrefix(prefix) {
					keys = append(keys, iter.Key().Clone())
					err = iter.Next()
					if err != nil {
						return errors.Trace(err)
					}
				} else {
					break
				}
			}

			for _, key := range keys {
				err := txn.Delete(key)
				// must skip ErrNotExist
				// if key doesn't exist, skip this error.
				if err != nil && !terror.ErrorEqual(err, kv.ErrNotExist) {
					return errors.Trace(err)
				}
			}

			count += len(keys)
			total += int64(len(keys))
			return nil
		})
		sub := time.Since(startTS).Seconds()
		if err != nil {
			log.Warnf("[ddl] deleted %d keys failed, take time %v, deleted %d keys in total", len(keys), sub, total)
			return 0, startKey, errors.Trace(err)
		}

		// Update the background job's RowCount.
		job.SetRowCount(total)
		d.setReorgRowCount(total)
		batchHandleDataHistogram.WithLabelValues(batchDelData).Observe(sub)
		log.Infof("[ddl] deleted %d keys take time %v, deleted %d keys in total", len(keys), sub, total)

		if len(keys) > 0 {
			startKey = keys[len(keys)-1]
		}

		if noMoreKeysToDelete := len(keys) < batch; noMoreKeysToDelete {
			break
		}

		keys = keys[:0]
	}

	return count, startKey, nil
}

type reorgInfo struct {
	*model.Job
	Handle int64
	d      *ddl
	first  bool
}

func (d *ddl) getReorgInfo(t *meta.Meta, job *model.Job) (*reorgInfo, error) {
	var err error

	info := &reorgInfo{
		Job:   job,
		d:     d,
		first: job.SnapshotVer == 0,
	}

	if info.first {
		// get the current version for reorganization if we don't have
		var ver kv.Version
		ver, err = d.store.CurrentVersion()
		if err != nil {
			return nil, errors.Trace(err)
		} else if ver.Ver <= 0 {
			return nil, errInvalidStoreVer.Gen("invalid storage current version %d", ver.Ver)
		}

		job.SnapshotVer = ver.Ver
	} else {
		info.Handle, err = t.GetDDLReorgHandle(job)
		if err != nil {
			return nil, errors.Trace(err)
		}
	}

	if info.Handle > 0 {
		// we have already handled this handle, so use next
		info.Handle++
	}

	return info, errors.Trace(err)
}

func (r *reorgInfo) UpdateHandle(txn kv.Transaction, handle int64) error {
	t := meta.NewMeta(txn)
	return errors.Trace(t.UpdateDDLReorgHandle(r.Job, handle))
}