Refine QueryCoordV2 metrics (#20471)

Signed-off-by: yah01 <yang.cen@zilliz.com>

Signed-off-by: yah01 <yang.cen@zilliz.com>
This commit is contained in:
yah01 2022-11-10 17:09:06 +08:00 committed by GitHub
parent 14a544d631
commit ea45a2aa4b
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
5 changed files with 22 additions and 24 deletions

View File

@ -77,29 +77,12 @@ var (
Buckets: []float64{0, 5, 10, 20, 40, 100, 200, 400, 1000, 10000},
}, []string{})
QueryCoordNumChildTasks = prometheus.NewGaugeVec(
QueryCoordTaskNum = prometheus.NewGaugeVec(
prometheus.GaugeOpts{
Namespace: milvusNamespace,
Subsystem: typeutil.QueryCoordRole,
Name: "child_task_num",
Help: "number of child tasks in QueryCoord's queue",
}, []string{})
QueryCoordNumParentTasks = prometheus.NewGaugeVec(
prometheus.GaugeOpts{
Namespace: milvusNamespace,
Subsystem: typeutil.QueryCoordRole,
Name: "parent_task_num",
Help: "number of parent tasks in QueryCoord's queue",
}, []string{})
QueryCoordChildTaskLatency = prometheus.NewHistogramVec(
prometheus.HistogramOpts{
Namespace: milvusNamespace,
Subsystem: typeutil.QueryCoordRole,
Name: "child_task_latency",
Help: "latency of child tasks",
Buckets: buckets,
Name: "task_num",
Help: "the number of tasks in QueryCoord's scheduler",
}, []string{})
QueryCoordNumQueryNodes = prometheus.NewGaugeVec(
@ -119,8 +102,6 @@ func RegisterQueryCoord(registry *prometheus.Registry) {
registry.MustRegister(QueryCoordReleaseCount)
registry.MustRegister(QueryCoordLoadLatency)
registry.MustRegister(QueryCoordReleaseLatency)
registry.MustRegister(QueryCoordNumChildTasks)
registry.MustRegister(QueryCoordNumParentTasks)
registry.MustRegister(QueryCoordChildTaskLatency)
registry.MustRegister(QueryCoordTaskNum)
registry.MustRegister(QueryCoordNumQueryNodes)
}

View File

@ -22,6 +22,7 @@ import (
"time"
"github.com/milvus-io/milvus/internal/log"
"github.com/milvus-io/milvus/internal/metrics"
"github.com/milvus-io/milvus/internal/proto/querypb"
"github.com/milvus-io/milvus/internal/querycoordv2/meta"
"github.com/milvus-io/milvus/internal/querycoordv2/observers"
@ -235,6 +236,7 @@ func (job *LoadCollectionJob) Execute() error {
return utils.WrapError(msg, err)
}
metrics.QueryCoordNumCollections.WithLabelValues().Inc()
return nil
}
@ -299,6 +301,7 @@ func (job *ReleaseCollectionJob) Execute() error {
job.targetMgr.RemoveCollection(req.GetCollectionID())
waitCollectionReleased(job.dist, req.GetCollectionID())
metrics.QueryCoordNumCollections.WithLabelValues().Dec()
return nil
}
@ -443,6 +446,7 @@ func (job *LoadPartitionJob) Execute() error {
return utils.WrapError(msg, err)
}
metrics.QueryCoordNumCollections.WithLabelValues().Inc()
return nil
}
@ -538,5 +542,6 @@ func (job *ReleasePartitionJob) Execute() error {
}
waitCollectionReleased(job.dist, req.GetCollectionID(), toRelease...)
}
metrics.QueryCoordNumCollections.WithLabelValues().Dec()
return nil
}

View File

@ -27,6 +27,7 @@ import (
"syscall"
"time"
"github.com/milvus-io/milvus/internal/metrics"
"github.com/samber/lo"
clientv3 "go.etcd.io/etcd/client/v3"
"go.uber.org/zap"
@ -251,6 +252,8 @@ func (s *Server) initMeta() error {
log.Error("failed to recover collections")
return err
}
metrics.QueryCoordNumCollections.WithLabelValues().Set(float64(len(s.meta.GetAll())))
err = s.meta.ReplicaManager.Recover()
if err != nil {
log.Error("failed to recover replicas")

View File

@ -16,7 +16,11 @@
package session
import "sync"
import (
"sync"
"github.com/milvus-io/milvus/internal/metrics"
)
type Manager interface {
Add(node *NodeInfo)
@ -34,12 +38,14 @@ func (m *NodeManager) Add(node *NodeInfo) {
m.mu.Lock()
defer m.mu.Unlock()
m.nodes[node.ID()] = node
metrics.QueryCoordNumQueryNodes.WithLabelValues().Set(float64(len(m.nodes)))
}
func (m *NodeManager) Remove(nodeID int64) {
m.mu.Lock()
defer m.mu.Unlock()
delete(m.nodes, nodeID)
metrics.QueryCoordNumQueryNodes.WithLabelValues().Set(float64(len(m.nodes)))
}
func (m *NodeManager) Get(nodeID int64) *NodeInfo {

View File

@ -23,6 +23,7 @@ import (
"sync"
"github.com/milvus-io/milvus/internal/log"
"github.com/milvus-io/milvus/internal/metrics"
"github.com/milvus-io/milvus/internal/proto/querypb"
"github.com/milvus-io/milvus/internal/querycoordv2/meta"
"github.com/milvus-io/milvus/internal/querycoordv2/session"
@ -230,6 +231,7 @@ func (scheduler *taskScheduler) Add(task Task) error {
log.Warn("failed to add task", zap.String("task", task.String()))
return nil
}
metrics.QueryCoordTaskNum.WithLabelValues().Set(float64(scheduler.tasks.Len()))
log.Info("task added", zap.String("task", task.String()))
return nil
}
@ -574,6 +576,7 @@ func (scheduler *taskScheduler) remove(task Task) {
log = log.With(zap.String("channel", task.Channel()))
}
metrics.QueryCoordTaskNum.WithLabelValues().Set(float64(scheduler.tasks.Len()))
log.Debug("task removed")
}