Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

cherry-pick some bug fixes from master. #895

Merged
merged 3 commits into from
Dec 26, 2017
Merged

cherry-pick some bug fixes from master. #895

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
9ffa098
scheduler: check both add leader and remove leader when balance leader.
disksing Dec 25, 2017
bd9c52c
server: update limit using all operators
disksing Dec 25, 2017
8ea591c
server: fix panic that cluster status is nil
disksing Dec 25, 2017
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 9 additions & 0 deletions server/balancer_test.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -386,6 +386,15 @@ func (s *testBalanceLeaderSchedulerSuite) TestBalanceSelector(c *C) {
s.tc.updateLeaderCount(3, 9)
// Average leader is 7. Select store 1 as target.
checkTransferLeader(c, s.schedule(), 3, 1)

// Stores: 1 2 3 4
// Leaders: 1 2 15 16
// Region1: - F F L
// Region2: - F L F
s.tc.addLeaderRegion(2, 3, 2, 4)
s.tc.addLeaderStore(2, 2)
// Unable to find a region in store 1. Transfer a leader out of store 4 instead.
checkTransferLeader(c, s.schedule(), 4, 2)
}

var _ = Suite(&testBalanceRegionSchedulerSuite{})
Expand Down
23 changes: 6 additions & 17 deletions server/cluster.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -62,8 +62,6 @@ type RaftCluster struct {

wg sync.WaitGroup
quit chan struct{}

status *ClusterStatus
}

// ClusterStatus saves some state information
Expand All @@ -80,22 +78,19 @@ func newRaftCluster(s *Server, clusterID uint64) *RaftCluster {
}
}

func (c *RaftCluster) loadClusterStatus() error {
func (c *RaftCluster) loadClusterStatus() (*ClusterStatus, error) {
data, err := c.s.kv.Load((c.s.kv.ClusterStatePath("raft_bootstrap_time")))
if err != nil {
return errors.Trace(err)
return nil, errors.Trace(err)
}
if len(data) == 0 {
return nil
return &ClusterStatus{}, nil
}
t, err := parseTimestamp([]byte(data))
if err != nil {
return errors.Trace(err)
}
c.status = &ClusterStatus{
RaftBootstrapTime: t,
return nil, errors.Trace(err)
}
return nil
return &ClusterStatus{RaftBootstrapTime: t}, nil
}

func (c *RaftCluster) start() error {
Expand Down Expand Up @@ -217,13 +212,7 @@ func (s *Server) GetCluster() *metapb.Cluster {
func (s *Server) GetClusterStatus() (*ClusterStatus, error) {
s.cluster.Lock()
defer s.cluster.Unlock()
err := s.cluster.loadClusterStatus()
if err != nil {
return nil, errors.Trace(err)
}
clone := &ClusterStatus{}
*clone = *s.cluster.status
return clone, nil
return s.cluster.loadClusterStatus()
}

func (s *Server) createRaftCluster() error {
Expand Down
20 changes: 10 additions & 10 deletions server/coordinator.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -379,8 +379,8 @@ func (c *coordinator) addOperator(op *schedule.Operator) bool {
}

c.histories.Put(regionID, op)
c.limiter.addOperator(op)
c.operators[regionID] = op
c.limiter.updateCounts(c.operators)

if region := c.cluster.GetRegion(op.RegionID()); region != nil {
if step := op.Check(region); step != nil {
Expand Down Expand Up @@ -410,9 +410,8 @@ func (c *coordinator) removeOperator(op *schedule.Operator) {

func (c *coordinator) removeOperatorLocked(op *schedule.Operator) {
regionID := op.RegionID()
c.limiter.removeOperator(op)
delete(c.operators, regionID)

c.limiter.updateCounts(c.operators)
c.histories.Put(regionID, op)
operatorCounter.WithLabelValues(op.Desc(), "remove").Inc()
}
Expand Down Expand Up @@ -511,16 +510,17 @@ func newScheduleLimiter() *scheduleLimiter {
}
}

func (l *scheduleLimiter) addOperator(op *schedule.Operator) {
// updateCounts updates resouce counts using current pending operators.
func (l *scheduleLimiter) updateCounts(operators map[uint64]*schedule.Operator) {
l.Lock()
defer l.Unlock()
l.counts[op.ResourceKind()]++
}

func (l *scheduleLimiter) removeOperator(op *schedule.Operator) {
l.Lock()
defer l.Unlock()
l.counts[op.ResourceKind()]--
for k := range l.counts {
l.counts[k] = 0
}
for _, op := range operators {
l.counts[op.ResourceKind()]++
}
}

func (l *scheduleLimiter) operatorCount(kind core.ResourceKind) uint64 {
Expand Down
33 changes: 19 additions & 14 deletions server/coordinator_test.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -482,21 +482,26 @@ func (s *testScheduleLimiterSuite) TestOperatorCount(c *C) {
c.Assert(l.operatorCount(core.LeaderKind), Equals, uint64(0))
c.Assert(l.operatorCount(core.RegionKind), Equals, uint64(0))

leaderOP := newTestOperator(1, core.LeaderKind)
l.addOperator(leaderOP)
operators := make(map[uint64]*schedule.Operator)

operators[1] = newTestOperator(1, core.LeaderKind)
l.updateCounts(operators)
c.Assert(l.operatorCount(core.LeaderKind), Equals, uint64(1)) // 1:leader
operators[2] = newTestOperator(2, core.LeaderKind)
l.updateCounts(operators)
c.Assert(l.operatorCount(core.LeaderKind), Equals, uint64(2)) // 1:leader, 2:leader
delete(operators, 1)
l.updateCounts(operators)
c.Assert(l.operatorCount(core.LeaderKind), Equals, uint64(1)) // 2:leader

operators[1] = newTestOperator(1, core.RegionKind)
l.updateCounts(operators)
c.Assert(l.operatorCount(core.RegionKind), Equals, uint64(1)) // 1:region 2:leader
c.Assert(l.operatorCount(core.LeaderKind), Equals, uint64(1))
l.addOperator(leaderOP)
c.Assert(l.operatorCount(core.LeaderKind), Equals, uint64(2))
l.removeOperator(leaderOP)
c.Assert(l.operatorCount(core.LeaderKind), Equals, uint64(1))

regionOP := newTestOperator(1, core.RegionKind)
l.addOperator(regionOP)
c.Assert(l.operatorCount(core.RegionKind), Equals, uint64(1))
l.addOperator(regionOP)
c.Assert(l.operatorCount(core.RegionKind), Equals, uint64(2))
l.removeOperator(regionOP)
c.Assert(l.operatorCount(core.RegionKind), Equals, uint64(1))
operators[2] = newTestOperator(2, core.RegionKind)
l.updateCounts(operators)
c.Assert(l.operatorCount(core.RegionKind), Equals, uint64(2)) // 1:region 2:region
c.Assert(l.operatorCount(core.LeaderKind), Equals, uint64(0))
}

var _ = Suite(&testScheduleControllerSuite{})
Expand Down
45 changes: 30 additions & 15 deletions server/schedulers/utils.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,7 +25,7 @@ import (
)

// scheduleTransferLeader schedules a region to transfer leader to the peer.
func scheduleTransferLeader(cluster schedule.Cluster, schedulerName string, s schedule.Selector, filters ...schedule.Filter) (*core.RegionInfo, *metapb.Peer) {
func scheduleTransferLeader(cluster schedule.Cluster, schedulerName string, s schedule.Selector, filters ...schedule.Filter) (region *core.RegionInfo, peer *metapb.Peer) {
stores := cluster.GetStores()
if len(stores) == 0 {
schedulerCounter.WithLabelValues(schedulerName, "no_store").Inc()
Expand Down Expand Up @@ -53,29 +53,44 @@ func scheduleTransferLeader(cluster schedule.Cluster, schedulerName string, s sc
}

if mostLeaderDistance > leastLeaderDistance {
// Transfer a leader out of mostLeaderStore.
region := cluster.RandLeaderRegion(mostLeaderStore.GetId())
region, peer = scheduleRemoveLeader(cluster, schedulerName, mostLeaderStore.GetId(), s)
if region == nil {
schedulerCounter.WithLabelValues(schedulerName, "no_leader_region").Inc()
return nil, nil
region, peer = scheduleAddLeader(cluster, schedulerName, leastLeaderStore.GetId())
}
targetStores := cluster.GetFollowerStores(region)
target := s.SelectTarget(targetStores)
if target == nil {
schedulerCounter.WithLabelValues(schedulerName, "no_target_store").Inc()
return nil, nil
} else {
region, peer = scheduleAddLeader(cluster, schedulerName, leastLeaderStore.GetId())
if region == nil {
region, peer = scheduleRemoveLeader(cluster, schedulerName, mostLeaderStore.GetId(), s)
}

return region, region.GetStorePeer(target.GetId())
}
return region, peer
}

// Transfer a leader into leastLeaderStore.
region := cluster.RandFollowerRegion(leastLeaderStore.GetId())
// scheduleAddLeader transfers a leader into the store.
func scheduleAddLeader(cluster schedule.Cluster, schedulerName string, storeID uint64) (*core.RegionInfo, *metapb.Peer) {
region := cluster.RandFollowerRegion(storeID)
if region == nil {
schedulerCounter.WithLabelValues(schedulerName, "no_target_peer").Inc()
return nil, nil
}
return region, region.GetStorePeer(leastLeaderStore.GetId())
return region, region.GetStorePeer(storeID)
}

// scheduleRemoveLeader transfers a leader out of the store.
func scheduleRemoveLeader(cluster schedule.Cluster, schedulerName string, storeID uint64, s schedule.Selector) (*core.RegionInfo, *metapb.Peer) {
region := cluster.RandLeaderRegion(storeID)
if region == nil {
schedulerCounter.WithLabelValues(schedulerName, "no_leader_region").Inc()
return nil, nil
}
targetStores := cluster.GetFollowerStores(region)
target := s.SelectTarget(targetStores)
if target == nil {
schedulerCounter.WithLabelValues(schedulerName, "no_target_store").Inc()
return nil, nil
}

return region, region.GetStorePeer(target.GetId())
}

// scheduleRemovePeer schedules a region to remove the peer.
Expand Down