From 4d5d6edcbe433853bf41d16a2d1a61c2d97223ce Mon Sep 17 00:00:00 2001
From: Dmitrii Stepanov <d.stepanov@yadro.com>
Date: Wed, 5 Mar 2025 15:44:15 +0300
Subject: [PATCH] [#1653] treeSvc: Add operations by IO tag metric
Signed-off-by: Dmitrii Stepanov <d.stepanov@yadro.com>
---
internal/metrics/treeservice.go | 15 +++++++++++++++
pkg/services/tree/metrics.go | 2 ++
pkg/services/tree/service.go | 9 +++++++++
3 files changed, 26 insertions(+)
diff --git a/internal/metrics/treeservice.go b/internal/metrics/treeservice.go
index 6702aa83c..e192c4398 100644
--- a/internal/metrics/treeservice.go
+++ b/internal/metrics/treeservice.go
@@ -12,12 +12,14 @@ type TreeMetricsRegister interface {
AddReplicateTaskDuration(time.Duration, bool)
AddReplicateWaitDuration(time.Duration, bool)
AddSyncDuration(time.Duration, bool)
+ AddOperation(string, string)
}
type treeServiceMetrics struct {
replicateTaskDuration *prometheus.HistogramVec
replicateWaitDuration *prometheus.HistogramVec
syncOpDuration *prometheus.HistogramVec
+ ioTagOpsCounter *prometheus.CounterVec
}
var _ TreeMetricsRegister = (*treeServiceMetrics)(nil)
@@ -42,6 +44,12 @@ func newTreeServiceMetrics() *treeServiceMetrics {
Name: "sync_duration_seconds",
Help: "Duration of synchronization operations",
}, []string{successLabel}),
+ ioTagOpsCounter: metrics.NewCounterVec(prometheus.CounterOpts{
+ Namespace: namespace,
+ Subsystem: treeServiceSubsystem,
+ Name: "requests_total",
+ Help: "Count of requests for each IO tag",
+ }, []string{methodLabel, ioTagLabel}),
}
}
@@ -62,3 +70,10 @@ func (m *treeServiceMetrics) AddSyncDuration(d time.Duration, success bool) {
successLabel: strconv.FormatBool(success),
}).Observe(d.Seconds())
}
+
+func (m *treeServiceMetrics) AddOperation(op string, ioTag string) {
+ m.ioTagOpsCounter.With(prometheus.Labels{
+ ioTagLabel: ioTag,
+ methodLabel: op,
+ }).Inc()
+}
diff --git a/pkg/services/tree/metrics.go b/pkg/services/tree/metrics.go
index 0f0e4ee57..07503f8c3 100644
--- a/pkg/services/tree/metrics.go
+++ b/pkg/services/tree/metrics.go
@@ -6,6 +6,7 @@ type MetricsRegister interface {
AddReplicateTaskDuration(time.Duration, bool)
AddReplicateWaitDuration(time.Duration, bool)
AddSyncDuration(time.Duration, bool)
+ AddOperation(string, string)
}
type defaultMetricsRegister struct{}
@@ -13,3 +14,4 @@ type defaultMetricsRegister struct{}
func (defaultMetricsRegister) AddReplicateTaskDuration(time.Duration, bool) {}
func (defaultMetricsRegister) AddReplicateWaitDuration(time.Duration, bool) {}
func (defaultMetricsRegister) AddSyncDuration(time.Duration, bool) {}
+func (defaultMetricsRegister) AddOperation(string, string) {}
diff --git a/pkg/services/tree/service.go b/pkg/services/tree/service.go
index 2e9722e79..f9b7395e7 100644
--- a/pkg/services/tree/service.go
+++ b/pkg/services/tree/service.go
@@ -105,6 +105,7 @@ func (s *Service) Shutdown() {
}
func (s *Service) Add(ctx context.Context, req *AddRequest) (*AddResponse, error) {
+ defer s.metrics.AddOperation("Add", qos.IOTagFromContext(ctx))
if !s.initialSyncDone.Load() {
return nil, ErrAlreadySyncing
}
@@ -148,6 +149,7 @@ func (s *Service) Add(ctx context.Context, req *AddRequest) (*AddResponse, error
}
func (s *Service) AddByPath(ctx context.Context, req *AddByPathRequest) (*AddByPathResponse, error) {
+ defer s.metrics.AddOperation("AddByPath", qos.IOTagFromContext(ctx))
if !s.initialSyncDone.Load() {
return nil, ErrAlreadySyncing
}
@@ -203,6 +205,7 @@ func (s *Service) AddByPath(ctx context.Context, req *AddByPathRequest) (*AddByP
}
func (s *Service) Remove(ctx context.Context, req *RemoveRequest) (*RemoveResponse, error) {
+ defer s.metrics.AddOperation("Remove", qos.IOTagFromContext(ctx))
if !s.initialSyncDone.Load() {
return nil, ErrAlreadySyncing
}
@@ -247,6 +250,7 @@ func (s *Service) Remove(ctx context.Context, req *RemoveRequest) (*RemoveRespon
// Move applies client operation to the specified tree and pushes in queue
// for replication on other nodes.
func (s *Service) Move(ctx context.Context, req *MoveRequest) (*MoveResponse, error) {
+ defer s.metrics.AddOperation("Move", qos.IOTagFromContext(ctx))
if !s.initialSyncDone.Load() {
return nil, ErrAlreadySyncing
}
@@ -290,6 +294,7 @@ func (s *Service) Move(ctx context.Context, req *MoveRequest) (*MoveResponse, er
}
func (s *Service) GetNodeByPath(ctx context.Context, req *GetNodeByPathRequest) (*GetNodeByPathResponse, error) {
+ defer s.metrics.AddOperation("GetNodeByPath", qos.IOTagFromContext(ctx))
if !s.initialSyncDone.Load() {
return nil, ErrAlreadySyncing
}
@@ -363,6 +368,7 @@ func (s *Service) GetNodeByPath(ctx context.Context, req *GetNodeByPathRequest)
}
func (s *Service) GetSubTree(req *GetSubTreeRequest, srv TreeService_GetSubTreeServer) error {
+ defer s.metrics.AddOperation("GetSubTree", qos.IOTagFromContext(srv.Context()))
if !s.initialSyncDone.Load() {
return ErrAlreadySyncing
}
@@ -590,6 +596,7 @@ func sortByFilename(nodes []pilorama.NodeInfo, d GetSubTreeRequest_Body_Order_Di
// Apply locally applies operation from the remote node to the tree.
func (s *Service) Apply(ctx context.Context, req *ApplyRequest) (*ApplyResponse, error) {
+ defer s.metrics.AddOperation("Apply", qos.IOTagFromContext(ctx))
err := verifyMessage(req)
if err != nil {
return nil, err
@@ -633,6 +640,7 @@ func (s *Service) Apply(ctx context.Context, req *ApplyRequest) (*ApplyResponse,
}
func (s *Service) GetOpLog(req *GetOpLogRequest, srv TreeService_GetOpLogServer) error {
+ defer s.metrics.AddOperation("GetOpLog", qos.IOTagFromContext(srv.Context()))
if !s.initialSyncDone.Load() {
return ErrAlreadySyncing
}
@@ -697,6 +705,7 @@ func (s *Service) GetOpLog(req *GetOpLogRequest, srv TreeService_GetOpLogServer)
}
func (s *Service) TreeList(ctx context.Context, req *TreeListRequest) (*TreeListResponse, error) {
+ defer s.metrics.AddOperation("TreeList", qos.IOTagFromContext(ctx))
if !s.initialSyncDone.Load() {
return nil, ErrAlreadySyncing
}