seaweedfs/weed/server/volume_grpc_vacuum.go

package weed_server

import (
	"context"
	"fmt"
	"strconv"
	"time"

	"github.com/seaweedfs/seaweedfs/weed/stats"

	"runtime"

	"github.com/prometheus/procfs"
	"github.com/seaweedfs/seaweedfs/weed/glog"
	"github.com/seaweedfs/seaweedfs/weed/pb/volume_server_pb"
	"github.com/seaweedfs/seaweedfs/weed/storage/needle"
	"google.golang.org/grpc/status"
)

var numCPU = runtime.NumCPU()

func (vs *VolumeServer) VacuumVolumeCheck(ctx context.Context, req *volume_server_pb.VacuumVolumeCheckRequest) (*volume_server_pb.VacuumVolumeCheckResponse, error) {

	resp := &volume_server_pb.VacuumVolumeCheckResponse{}

	garbageRatio, diskSpaceLow, err := vs.store.CheckCompactVolume(needle.VolumeId(req.VolumeId))

	resp.GarbageRatio = garbageRatio
	resp.DiskSpaceLow = diskSpaceLow

	if err != nil {
		glog.V(3).Infof("check volume %d: %v", req.VolumeId, err)
		return resp, volumeStatusError(err)
	}

	return resp, nil

}

func (vs *VolumeServer) VacuumVolumeCompact(req *volume_server_pb.VacuumVolumeCompactRequest, stream volume_server_pb.VolumeServer_VacuumVolumeCompactServer) error {
	if err := vs.checkGrpcAdminAuth(stream.Context()); err != nil {
		return err
	}
	if err := vs.CheckMaintenanceMode(); err != nil {
		return err
	}

	start := time.Now()
	defer func(start time.Time) {
		stats.VolumeServerVacuumingHistogram.WithLabelValues("compact").Observe(time.Since(start).Seconds())
	}(start)

	resp := &volume_server_pb.VacuumVolumeCompactResponse{}
	reportInterval := int64(1024 * 1024 * 128)
	nextReportTarget := reportInterval
	fs, fsErr := procfs.NewDefaultFS()
	var sendErr error
	err := vs.store.CompactVolume(needle.VolumeId(req.VolumeId), req.Preallocate, vs.compactionBytePerSecond, func(processed int64) bool {
		if stream.Context().Err() != nil {
			return false
		}
		if processed > nextReportTarget {
			resp.ProcessedBytes = processed
			if fsErr == nil && numCPU > 0 {
				if fsLa, err := fs.LoadAvg(); err == nil {
					resp.LoadAvg_1M = float32(fsLa.Load1 / float64(numCPU))
				}
			}
			if sendErr = stream.Send(resp); sendErr != nil {
				return false
			}
			nextReportTarget = processed + reportInterval
		}
		return true
	})

	stats.VolumeServerVacuumingCompactCounter.WithLabelValues(strconv.FormatBool(err == nil && sendErr == nil)).Inc()
	if err != nil {
		glog.Errorf("failed compact volume %d: %v", req.VolumeId, err)
		return volumeStatusError(fmt.Errorf("compact volume %d: %w", req.VolumeId, err))
	}
	if sendErr != nil {
		glog.Errorf("failed compact volume %d report progress: %v", req.VolumeId, sendErr)
		return sendErr
	}

	glog.V(1).Infof("compact volume %d", req.VolumeId)
	return nil

}

func (vs *VolumeServer) VacuumVolumeCommit(ctx context.Context, req *volume_server_pb.VacuumVolumeCommitRequest) (*volume_server_pb.VacuumVolumeCommitResponse, error) {
	if err := vs.checkGrpcAdminAuth(ctx); err != nil {
		return nil, err
	}
	if err := vs.CheckMaintenanceMode(); err != nil {
		return nil, err
	}

	release, err := vs.acquireVacuumCommitSlot(ctx)
	if err != nil {
		glog.Warningf("commit volume %d not started: %v", req.VolumeId, err)
		return nil, status.FromContextError(err).Err()
	}
	defer release()
	if err := vs.CheckMaintenanceMode(); err != nil {
		return nil, err
	}

	start := time.Now()
	defer func(start time.Time) {
		stats.VolumeServerVacuumingHistogram.WithLabelValues("commit").Observe(time.Since(start).Seconds())
	}(start)

	resp := &volume_server_pb.VacuumVolumeCommitResponse{}

	readOnly, volumeSize, err := vs.store.CommitCompactVolume(needle.VolumeId(req.VolumeId))

	stats.VolumeServerVacuumingCommitCounter.WithLabelValues(strconv.FormatBool(err == nil)).Inc()
	resp.IsReadOnly = readOnly
	resp.VolumeSize = uint64(volumeSize)
	if err != nil {
		glog.Errorf("failed commit volume %d: %v", req.VolumeId, err)
		return resp, volumeStatusError(fmt.Errorf("commit compact volume %d: %w", req.VolumeId, err))
	}
	glog.V(1).Infof("commit volume %d", req.VolumeId)
	return resp, nil

}

// acquireVacuumCommitSlot waits for a commit slot holding no volume lock; the
// wait counts against the caller's deadline.
func (vs *VolumeServer) acquireVacuumCommitSlot(ctx context.Context) (release func(), err error) {
	if vs.vacuumCommitSlots == nil {
		return func() {}, nil
	}
	vs.vacuumCommitsWaiting.Add(1)
	select {
	case vs.vacuumCommitSlots <- struct{}{}:
	case <-ctx.Done():
		err = ctx.Err()
	}
	vs.vacuumCommitsWaiting.Add(-1)
	if err != nil {
		return nil, err
	}
	release = func() { <-vs.vacuumCommitSlots }
	// select picks randomly when the slot and the deadline are ready together.
	if err := ctx.Err(); err != nil {
		release()
		return nil, err
	}
	return release, nil
}

func (vs *VolumeServer) VacuumVolumeCleanup(ctx context.Context, req *volume_server_pb.VacuumVolumeCleanupRequest) (*volume_server_pb.VacuumVolumeCleanupResponse, error) {
	if err := vs.checkGrpcAdminAuth(ctx); err != nil {
		return nil, err
	}
	if err := vs.CheckMaintenanceMode(); err != nil {
		return nil, err
	}

	resp := &volume_server_pb.VacuumVolumeCleanupResponse{}
	err := vs.store.CommitCleanupVolume(needle.VolumeId(req.VolumeId))

	if err != nil {
		glog.Errorf("failed cleanup volume %d: %v", req.VolumeId, err)
		return resp, volumeStatusError(fmt.Errorf("cleanup volume %d: %w", req.VolumeId, err))
	}
	glog.V(1).Infof("cleanup volume %d", req.VolumeId)

	return resp, nil

}