Skip to content

Commit

Permalink
scheduler: fix scheduler save config (#7108)
Browse files Browse the repository at this point in the history
close #6897

Signed-off-by: husharp <jinhao.hu@pingcap.com>

Co-authored-by: ti-chi-bot[bot] <108142056+ti-chi-bot[bot]@users.noreply.github.com>
  • Loading branch information
HuSharp and ti-chi-bot[bot] authored Sep 27, 2023
1 parent 8e844d8 commit 6752974
Show file tree
Hide file tree
Showing 5 changed files with 133 additions and 9 deletions.
2 changes: 1 addition & 1 deletion pkg/schedule/schedulers/evict_leader.go
Original file line number Diff line number Diff line change
Expand Up @@ -181,7 +181,7 @@ func newEvictLeaderScheduler(opController *operator.Controller, conf *evictLeade
}
}

// EvictStores returns the IDs of the evict-stores.
// EvictStoreIDs returns the IDs of the evict-stores.
func (s *evictLeaderScheduler) EvictStoreIDs() []uint64 {
return s.conf.getStores()
}
Expand Down
14 changes: 7 additions & 7 deletions pkg/schedule/schedulers/scheduler.go
Original file line number Diff line number Diff line change
Expand Up @@ -124,16 +124,16 @@ func CreateScheduler(typ string, oc *operator.Controller, storage endpoint.Confi
return nil, errs.ErrSchedulerCreateFuncNotRegistered.FastGenByArgs(typ)
}

s, err := fn(oc, storage, dec, removeSchedulerCb...)
if err != nil {
return nil, err
}
return fn(oc, storage, dec, removeSchedulerCb...)
}

// SaveSchedulerConfig saves the config of the specified scheduler.
func SaveSchedulerConfig(storage endpoint.ConfigStorage, s Scheduler) error {
data, err := s.EncodeConfig()
if err != nil {
return nil, err
return err
}
err = storage.SaveSchedulerConfig(s.GetName(), data)
return s, err
return storage.SaveSchedulerConfig(s.GetName(), data)
}

// FindSchedulerTypeByName finds the type of the specified name.
Expand Down
8 changes: 8 additions & 0 deletions pkg/schedule/schedulers/scheduler_controller.go
Original file line number Diff line number Diff line change
Expand Up @@ -138,6 +138,10 @@ func (c *Controller) AddSchedulerHandler(scheduler Scheduler, args ...string) er
}

c.schedulerHandlers[name] = scheduler
if err := SaveSchedulerConfig(c.storage, scheduler); err != nil {
log.Error("can not save HTTP scheduler config", zap.String("scheduler-name", scheduler.GetName()), errs.ZapError(err))
return err
}
c.cluster.GetSchedulerConfig().AddSchedulerCfg(scheduler.GetType(), args)
return nil
}
Expand Down Expand Up @@ -188,6 +192,10 @@ func (c *Controller) AddScheduler(scheduler Scheduler, args ...string) error {
c.wg.Add(1)
go c.runScheduler(s)
c.schedulers[s.Scheduler.GetName()] = s
if err := SaveSchedulerConfig(c.storage, scheduler); err != nil {
log.Error("can not save scheduler config", zap.String("scheduler-name", scheduler.GetName()), errs.ZapError(err))
return err
}
c.cluster.GetSchedulerConfig().AddSchedulerCfg(s.Scheduler.GetType(), args)
return nil
}
Expand Down
3 changes: 2 additions & 1 deletion server/cluster/cluster_test.go
Original file line number Diff line number Diff line change
Expand Up @@ -3111,8 +3111,9 @@ func TestPersistScheduler(t *testing.T) {
// whether the schedulers added or removed in dynamic way are recorded in opt
_, newOpt, err := newTestScheduleConfig()
re.NoError(err)
_, err = schedulers.CreateScheduler(schedulers.ShuffleRegionType, oc, storage, schedulers.ConfigJSONDecoder([]byte("null")))
shuffle, err := schedulers.CreateScheduler(schedulers.ShuffleRegionType, oc, storage, schedulers.ConfigJSONDecoder([]byte("null")))
re.NoError(err)
re.NoError(controller.AddScheduler(shuffle))
// suppose we add a new default enable scheduler
sc.DefaultSchedulers = append(sc.DefaultSchedulers, sc.SchedulerConfig{Type: "shuffle-region"})
defer func() {
Expand Down
115 changes: 115 additions & 0 deletions tests/server/cluster/cluster_test.go
Original file line number Diff line number Diff line change
Expand Up @@ -37,6 +37,7 @@ import (
"github.com/tikv/pd/pkg/mock/mockid"
sc "github.com/tikv/pd/pkg/schedule/config"
"github.com/tikv/pd/pkg/schedule/operator"
"github.com/tikv/pd/pkg/schedule/schedulers"
"github.com/tikv/pd/pkg/storage"
"github.com/tikv/pd/pkg/syncer"
"github.com/tikv/pd/pkg/tso"
Expand All @@ -47,6 +48,7 @@ import (
"github.com/tikv/pd/server/cluster"
"github.com/tikv/pd/server/config"
"github.com/tikv/pd/tests"
"github.com/tikv/pd/tests/server/api"
"google.golang.org/grpc/codes"
"google.golang.org/grpc/status"
)
Expand Down Expand Up @@ -1275,6 +1277,119 @@ func TestStaleTermHeartbeat(t *testing.T) {
re.NoError(err)
}

func TestTransferLeaderForScheduler(t *testing.T) {
re := require.New(t)
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
re.NoError(failpoint.Enable("github.com/tikv/pd/pkg/schedule/changeCoordinatorTicker", `return(true)`))
tc, err := tests.NewTestCluster(ctx, 2)
defer tc.Destroy()
re.NoError(err)
err = tc.RunInitialServers()
re.NoError(err)
tc.WaitLeader()
// start
leaderServer := tc.GetServer(tc.GetLeader())
re.NoError(leaderServer.BootstrapCluster())
rc := leaderServer.GetServer().GetRaftCluster()
re.NotNil(rc)

storesNum := 2
grpcPDClient := testutil.MustNewGrpcClient(re, leaderServer.GetAddr())
for i := 1; i <= storesNum; i++ {
store := &metapb.Store{
Id: uint64(i),
Address: "127.0.0.1:" + strconv.Itoa(i),
}
resp, err := putStore(grpcPDClient, leaderServer.GetClusterID(), store)
re.NoError(err)
re.Equal(pdpb.ErrorType_OK, resp.GetHeader().GetError().GetType())
}
// region heartbeat
id := leaderServer.GetAllocator()
putRegionWithLeader(re, rc, id, 1)

time.Sleep(time.Second)
re.True(leaderServer.GetRaftCluster().IsPrepared())
// Add evict leader scheduler
api.MustAddScheduler(re, leaderServer.GetAddr(), schedulers.EvictLeaderName, map[string]interface{}{
"store_id": 1,
})
api.MustAddScheduler(re, leaderServer.GetAddr(), schedulers.EvictLeaderName, map[string]interface{}{
"store_id": 2,
})
// Check scheduler updated.
schedulersController := rc.GetCoordinator().GetSchedulersController()
re.Len(schedulersController.GetSchedulerNames(), 6)
checkEvictLeaderSchedulerExist(re, schedulersController, true)
checkEvictLeaderStoreIDs(re, schedulersController, []uint64{1, 2})

// transfer PD leader to another PD
tc.ResignLeader()
rc.Stop()
tc.WaitLeader()
leaderServer = tc.GetServer(tc.GetLeader())
rc1 := leaderServer.GetServer().GetRaftCluster()
rc1.Start(leaderServer.GetServer())
re.NoError(err)
re.NotNil(rc1)
// region heartbeat
id = leaderServer.GetAllocator()
putRegionWithLeader(re, rc1, id, 1)
time.Sleep(time.Second)
re.True(leaderServer.GetRaftCluster().IsPrepared())
// Check scheduler updated.
schedulersController = rc1.GetCoordinator().GetSchedulersController()
re.Len(schedulersController.GetSchedulerNames(), 6)
checkEvictLeaderSchedulerExist(re, schedulersController, true)
checkEvictLeaderStoreIDs(re, schedulersController, []uint64{1, 2})

// transfer PD leader back to the previous PD
tc.ResignLeader()
rc1.Stop()
tc.WaitLeader()
leaderServer = tc.GetServer(tc.GetLeader())
rc = leaderServer.GetServer().GetRaftCluster()
rc.Start(leaderServer.GetServer())
re.NotNil(rc)
// region heartbeat
id = leaderServer.GetAllocator()
putRegionWithLeader(re, rc, id, 1)
time.Sleep(time.Second)
re.True(leaderServer.GetRaftCluster().IsPrepared())
// Check scheduler updated
schedulersController = rc.GetCoordinator().GetSchedulersController()
re.Len(schedulersController.GetSchedulerNames(), 6)
checkEvictLeaderSchedulerExist(re, schedulersController, true)
checkEvictLeaderStoreIDs(re, schedulersController, []uint64{1, 2})

re.NoError(failpoint.Disable("github.com/tikv/pd/pkg/schedule/changeCoordinatorTicker"))
}

func checkEvictLeaderSchedulerExist(re *require.Assertions, sc *schedulers.Controller, exist bool) {
testutil.Eventually(re, func() bool {
if !exist {
return sc.GetScheduler(schedulers.EvictLeaderName) == nil
}
return sc.GetScheduler(schedulers.EvictLeaderName) != nil
})
}

func checkEvictLeaderStoreIDs(re *require.Assertions, sc *schedulers.Controller, expected []uint64) {
handler, ok := sc.GetSchedulerHandlers()[schedulers.EvictLeaderName]
re.True(ok)
h, ok := handler.(interface {
EvictStoreIDs() []uint64
})
re.True(ok)
var evictStoreIDs []uint64
testutil.Eventually(re, func() bool {
evictStoreIDs = h.EvictStoreIDs()
return len(evictStoreIDs) == len(expected)
})
re.ElementsMatch(evictStoreIDs, expected)
}

func putRegionWithLeader(re *require.Assertions, rc *cluster.RaftCluster, id id.Allocator, storeID uint64) {
for i := 0; i < 3; i++ {
regionID, err := id.Alloc()
Expand Down

0 comments on commit 6752974

Please sign in to comment.