Enable callers to control whether per-tenant usage metrics are included in calls to store.ServiceUsage (#20672)

* Enable callers to control whether per-tenant usage metrics are included in calls to store.ServiceUsage

* Add changelog
This commit is contained in:
Matt Keeler 2024-03-01 13:44:55 -05:00 committed by GitHub
parent a1c6181677
commit 5c936fba33
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
11 changed files with 422 additions and 21 deletions

3
.changelog/20672.txt Normal file
View File

@ -0,0 +1,3 @@
```release-note: improvement
xds: Improved the performance of xDS server side load balancing. Its slightly improved in Consul CE with drastic CPU usage reductions in Consul Enterprise.
```

View File

@ -54,7 +54,7 @@ func (op *Operator) Usage(args *structs.OperatorUsageRequest, reply *structs.Usa
&reply.QueryMeta,
func(ws memdb.WatchSet, state *state.Store) error {
// Get service usage.
index, serviceUsage, err := state.ServiceUsage(ws)
index, serviceUsage, err := state.ServiceUsage(ws, true)
if err != nil {
return err
}

View File

@ -0,0 +1,12 @@
# Copyright (c) HashiCorp, Inc.
# SPDX-License-Identifier: BUSL-1.1
with-expecter: true
all: true
recursive: false
mockname: "{{.InterfaceName}}"
dir: "reportingmock"
filename: "mock_{{.InterfaceName}}.go"
outpkg: "reportingmock"
packages:
github.com/hashicorp/consul/agent/consul/reporting:

View File

@ -36,7 +36,7 @@ type ServerDelegate interface {
type StateDelegate interface {
NodeUsage() (uint64, state.NodeUsage, error)
ServiceUsage(ws memdb.WatchSet) (uint64, structs.ServiceUsage, error)
ServiceUsage(ws memdb.WatchSet, tenantUsage bool) (uint64, structs.ServiceUsage, error)
}
func NewReportingManager(logger hclog.Logger, deps EntDeps, server ServerDelegate, stateProvider StateDelegate) *ReportingManager {

View File

@ -0,0 +1,168 @@
// Code generated by mockery v2.37.1. DO NOT EDIT.
package reportingmock
import mock "github.com/stretchr/testify/mock"
// ServerDelegate is an autogenerated mock type for the ServerDelegate type
type ServerDelegate struct {
mock.Mock
}
type ServerDelegate_Expecter struct {
mock *mock.Mock
}
func (_m *ServerDelegate) EXPECT() *ServerDelegate_Expecter {
return &ServerDelegate_Expecter{mock: &_m.Mock}
}
// GetSystemMetadata provides a mock function with given fields: key
func (_m *ServerDelegate) GetSystemMetadata(key string) (string, error) {
ret := _m.Called(key)
var r0 string
var r1 error
if rf, ok := ret.Get(0).(func(string) (string, error)); ok {
return rf(key)
}
if rf, ok := ret.Get(0).(func(string) string); ok {
r0 = rf(key)
} else {
r0 = ret.Get(0).(string)
}
if rf, ok := ret.Get(1).(func(string) error); ok {
r1 = rf(key)
} else {
r1 = ret.Error(1)
}
return r0, r1
}
// ServerDelegate_GetSystemMetadata_Call is a *mock.Call that shadows Run/Return methods with type explicit version for method 'GetSystemMetadata'
type ServerDelegate_GetSystemMetadata_Call struct {
*mock.Call
}
// GetSystemMetadata is a helper method to define mock.On call
// - key string
func (_e *ServerDelegate_Expecter) GetSystemMetadata(key interface{}) *ServerDelegate_GetSystemMetadata_Call {
return &ServerDelegate_GetSystemMetadata_Call{Call: _e.mock.On("GetSystemMetadata", key)}
}
func (_c *ServerDelegate_GetSystemMetadata_Call) Run(run func(key string)) *ServerDelegate_GetSystemMetadata_Call {
_c.Call.Run(func(args mock.Arguments) {
run(args[0].(string))
})
return _c
}
func (_c *ServerDelegate_GetSystemMetadata_Call) Return(_a0 string, _a1 error) *ServerDelegate_GetSystemMetadata_Call {
_c.Call.Return(_a0, _a1)
return _c
}
func (_c *ServerDelegate_GetSystemMetadata_Call) RunAndReturn(run func(string) (string, error)) *ServerDelegate_GetSystemMetadata_Call {
_c.Call.Return(run)
return _c
}
// IsLeader provides a mock function with given fields:
func (_m *ServerDelegate) IsLeader() bool {
ret := _m.Called()
var r0 bool
if rf, ok := ret.Get(0).(func() bool); ok {
r0 = rf()
} else {
r0 = ret.Get(0).(bool)
}
return r0
}
// ServerDelegate_IsLeader_Call is a *mock.Call that shadows Run/Return methods with type explicit version for method 'IsLeader'
type ServerDelegate_IsLeader_Call struct {
*mock.Call
}
// IsLeader is a helper method to define mock.On call
func (_e *ServerDelegate_Expecter) IsLeader() *ServerDelegate_IsLeader_Call {
return &ServerDelegate_IsLeader_Call{Call: _e.mock.On("IsLeader")}
}
func (_c *ServerDelegate_IsLeader_Call) Run(run func()) *ServerDelegate_IsLeader_Call {
_c.Call.Run(func(args mock.Arguments) {
run()
})
return _c
}
func (_c *ServerDelegate_IsLeader_Call) Return(_a0 bool) *ServerDelegate_IsLeader_Call {
_c.Call.Return(_a0)
return _c
}
func (_c *ServerDelegate_IsLeader_Call) RunAndReturn(run func() bool) *ServerDelegate_IsLeader_Call {
_c.Call.Return(run)
return _c
}
// SetSystemMetadataKey provides a mock function with given fields: key, val
func (_m *ServerDelegate) SetSystemMetadataKey(key string, val string) error {
ret := _m.Called(key, val)
var r0 error
if rf, ok := ret.Get(0).(func(string, string) error); ok {
r0 = rf(key, val)
} else {
r0 = ret.Error(0)
}
return r0
}
// ServerDelegate_SetSystemMetadataKey_Call is a *mock.Call that shadows Run/Return methods with type explicit version for method 'SetSystemMetadataKey'
type ServerDelegate_SetSystemMetadataKey_Call struct {
*mock.Call
}
// SetSystemMetadataKey is a helper method to define mock.On call
// - key string
// - val string
func (_e *ServerDelegate_Expecter) SetSystemMetadataKey(key interface{}, val interface{}) *ServerDelegate_SetSystemMetadataKey_Call {
return &ServerDelegate_SetSystemMetadataKey_Call{Call: _e.mock.On("SetSystemMetadataKey", key, val)}
}
func (_c *ServerDelegate_SetSystemMetadataKey_Call) Run(run func(key string, val string)) *ServerDelegate_SetSystemMetadataKey_Call {
_c.Call.Run(func(args mock.Arguments) {
run(args[0].(string), args[1].(string))
})
return _c
}
func (_c *ServerDelegate_SetSystemMetadataKey_Call) Return(_a0 error) *ServerDelegate_SetSystemMetadataKey_Call {
_c.Call.Return(_a0)
return _c
}
func (_c *ServerDelegate_SetSystemMetadataKey_Call) RunAndReturn(run func(string, string) error) *ServerDelegate_SetSystemMetadataKey_Call {
_c.Call.Return(run)
return _c
}
// NewServerDelegate creates a new instance of ServerDelegate. It also registers a testing interface on the mock and a cleanup function to assert the mocks expectations.
// The first argument is typically a *testing.T value.
func NewServerDelegate(t interface {
mock.TestingT
Cleanup(func())
}) *ServerDelegate {
mock := &ServerDelegate{}
mock.Mock.Test(t)
t.Cleanup(func() { mock.AssertExpectations(t) })
return mock
}

View File

@ -0,0 +1,157 @@
// Code generated by mockery v2.37.1. DO NOT EDIT.
package reportingmock
import (
memdb "github.com/hashicorp/go-memdb"
mock "github.com/stretchr/testify/mock"
state "github.com/hashicorp/consul/agent/consul/state"
structs "github.com/hashicorp/consul/agent/structs"
)
// StateDelegate is an autogenerated mock type for the StateDelegate type
type StateDelegate struct {
mock.Mock
}
type StateDelegate_Expecter struct {
mock *mock.Mock
}
func (_m *StateDelegate) EXPECT() *StateDelegate_Expecter {
return &StateDelegate_Expecter{mock: &_m.Mock}
}
// NodeUsage provides a mock function with given fields:
func (_m *StateDelegate) NodeUsage() (uint64, state.NodeUsage, error) {
ret := _m.Called()
var r0 uint64
var r1 state.NodeUsage
var r2 error
if rf, ok := ret.Get(0).(func() (uint64, state.NodeUsage, error)); ok {
return rf()
}
if rf, ok := ret.Get(0).(func() uint64); ok {
r0 = rf()
} else {
r0 = ret.Get(0).(uint64)
}
if rf, ok := ret.Get(1).(func() state.NodeUsage); ok {
r1 = rf()
} else {
r1 = ret.Get(1).(state.NodeUsage)
}
if rf, ok := ret.Get(2).(func() error); ok {
r2 = rf()
} else {
r2 = ret.Error(2)
}
return r0, r1, r2
}
// StateDelegate_NodeUsage_Call is a *mock.Call that shadows Run/Return methods with type explicit version for method 'NodeUsage'
type StateDelegate_NodeUsage_Call struct {
*mock.Call
}
// NodeUsage is a helper method to define mock.On call
func (_e *StateDelegate_Expecter) NodeUsage() *StateDelegate_NodeUsage_Call {
return &StateDelegate_NodeUsage_Call{Call: _e.mock.On("NodeUsage")}
}
func (_c *StateDelegate_NodeUsage_Call) Run(run func()) *StateDelegate_NodeUsage_Call {
_c.Call.Run(func(args mock.Arguments) {
run()
})
return _c
}
func (_c *StateDelegate_NodeUsage_Call) Return(_a0 uint64, _a1 state.NodeUsage, _a2 error) *StateDelegate_NodeUsage_Call {
_c.Call.Return(_a0, _a1, _a2)
return _c
}
func (_c *StateDelegate_NodeUsage_Call) RunAndReturn(run func() (uint64, state.NodeUsage, error)) *StateDelegate_NodeUsage_Call {
_c.Call.Return(run)
return _c
}
// ServiceUsage provides a mock function with given fields: ws, tenantUsage
func (_m *StateDelegate) ServiceUsage(ws memdb.WatchSet, tenantUsage bool) (uint64, structs.ServiceUsage, error) {
ret := _m.Called(ws, tenantUsage)
var r0 uint64
var r1 structs.ServiceUsage
var r2 error
if rf, ok := ret.Get(0).(func(memdb.WatchSet, bool) (uint64, structs.ServiceUsage, error)); ok {
return rf(ws, tenantUsage)
}
if rf, ok := ret.Get(0).(func(memdb.WatchSet, bool) uint64); ok {
r0 = rf(ws, tenantUsage)
} else {
r0 = ret.Get(0).(uint64)
}
if rf, ok := ret.Get(1).(func(memdb.WatchSet, bool) structs.ServiceUsage); ok {
r1 = rf(ws, tenantUsage)
} else {
r1 = ret.Get(1).(structs.ServiceUsage)
}
if rf, ok := ret.Get(2).(func(memdb.WatchSet, bool) error); ok {
r2 = rf(ws, tenantUsage)
} else {
r2 = ret.Error(2)
}
return r0, r1, r2
}
// StateDelegate_ServiceUsage_Call is a *mock.Call that shadows Run/Return methods with type explicit version for method 'ServiceUsage'
type StateDelegate_ServiceUsage_Call struct {
*mock.Call
}
// ServiceUsage is a helper method to define mock.On call
// - ws memdb.WatchSet
// - tenantUsage bool
func (_e *StateDelegate_Expecter) ServiceUsage(ws interface{}, tenantUsage interface{}) *StateDelegate_ServiceUsage_Call {
return &StateDelegate_ServiceUsage_Call{Call: _e.mock.On("ServiceUsage", ws, tenantUsage)}
}
func (_c *StateDelegate_ServiceUsage_Call) Run(run func(ws memdb.WatchSet, tenantUsage bool)) *StateDelegate_ServiceUsage_Call {
_c.Call.Run(func(args mock.Arguments) {
run(args[0].(memdb.WatchSet), args[1].(bool))
})
return _c
}
func (_c *StateDelegate_ServiceUsage_Call) Return(_a0 uint64, _a1 structs.ServiceUsage, _a2 error) *StateDelegate_ServiceUsage_Call {
_c.Call.Return(_a0, _a1, _a2)
return _c
}
func (_c *StateDelegate_ServiceUsage_Call) RunAndReturn(run func(memdb.WatchSet, bool) (uint64, structs.ServiceUsage, error)) *StateDelegate_ServiceUsage_Call {
_c.Call.Return(run)
return _c
}
// NewStateDelegate creates a new instance of StateDelegate. It also registers a testing interface on the mock and a cleanup function to assert the mocks expectations.
// The first argument is typically a *testing.T value.
func NewStateDelegate(t interface {
mock.TestingT
Cleanup(func())
}) *StateDelegate {
mock := &StateDelegate{}
mock.Mock.Test(t)
t.Cleanup(func() { mock.AssertExpectations(t) })
return mock
}

View File

@ -410,7 +410,7 @@ func (s *Store) PeeringUsage() (uint64, PeeringUsage, error) {
// ServiceUsage returns the latest seen Raft index, a compiled set of service
// usage data, and any errors.
func (s *Store) ServiceUsage(ws memdb.WatchSet) (uint64, structs.ServiceUsage, error) {
func (s *Store) ServiceUsage(ws memdb.WatchSet, tenantUsage bool) (uint64, structs.ServiceUsage, error) {
tx := s.db.ReadTxn()
defer tx.Abort()
@ -450,6 +450,12 @@ func (s *Store) ServiceUsage(ws memdb.WatchSet) (uint64, structs.ServiceUsage, e
BillableServiceInstances: billableServiceInstances.Count,
Nodes: nodes.Count,
}
// Unless we need to gather per-tenant usage go ahead and return what we have
if !tenantUsage {
return serviceInstances.Index, usage, nil
}
results, err := compileEnterpriseServiceUsage(ws, tx, usage)
if err != nil {
return 0, structs.ServiceUsage{}, fmt.Errorf("failed services lookup: %s", err)

View File

@ -155,7 +155,7 @@ func TestStateStore_Usage_ServiceUsageEmpty(t *testing.T) {
s := testStateStore(t)
// No services have been registered, and thus no usage entry exists
idx, usage, err := s.ServiceUsage(nil)
idx, usage, err := s.ServiceUsage(nil, true)
require.NoError(t, err)
require.Equal(t, idx, uint64(0))
require.Equal(t, usage.Services, 0)
@ -186,7 +186,7 @@ func TestStateStore_Usage_ServiceUsage(t *testing.T) {
testRegisterAPIService(t, s, 20, "node2", "api")
ws := memdb.NewWatchSet()
idx, usage, err := s.ServiceUsage(ws)
idx, usage, err := s.ServiceUsage(ws, true)
require.NoError(t, err)
require.Equal(t, idx, uint64(20))
require.Equal(t, 9, usage.Services)
@ -232,7 +232,7 @@ func TestStateStore_Usage_ServiceUsage_DeleteNode(t *testing.T) {
testRegisterSidecarProxy(t, s, 3, "node1", "service2")
testRegisterConnectNativeService(t, s, 4, "node1", "service-connect")
idx, usage, err := s.ServiceUsage(nil)
idx, usage, err := s.ServiceUsage(nil, true)
require.NoError(t, err)
require.Equal(t, idx, uint64(4))
require.Equal(t, 3, usage.Services)
@ -243,7 +243,7 @@ func TestStateStore_Usage_ServiceUsage_DeleteNode(t *testing.T) {
require.NoError(t, s.DeleteNode(4, "node1", nil, ""))
idx, usage, err = s.ServiceUsage(nil)
idx, usage, err = s.ServiceUsage(nil, true)
require.NoError(t, err)
require.Equal(t, idx, uint64(4))
require.Equal(t, usage.Services, 0)
@ -272,7 +272,7 @@ func TestStateStore_Usage_ServiceUsagePeering(t *testing.T) {
testRegisterConnectNativeService(t, s, 7, "node2", "service-native")
testutil.RunStep(t, "writes", func(t *testing.T) {
idx, usage, err := s.ServiceUsage(nil)
idx, usage, err := s.ServiceUsage(nil, true)
require.NoError(t, err)
require.Equal(t, uint64(7), idx)
require.Equal(t, 3, usage.Services)
@ -285,7 +285,7 @@ func TestStateStore_Usage_ServiceUsagePeering(t *testing.T) {
testutil.RunStep(t, "deletes", func(t *testing.T) {
require.NoError(t, s.DeleteNode(7, "node1", nil, peerName))
require.NoError(t, s.DeleteNode(8, "node2", nil, ""))
idx, usage, err := s.ServiceUsage(nil)
idx, usage, err := s.ServiceUsage(nil, true)
require.NoError(t, err)
require.Equal(t, uint64(8), idx)
require.Equal(t, 0, usage.Services)
@ -324,7 +324,7 @@ func TestStateStore_Usage_Restore(t *testing.T) {
require.Equal(t, idx, uint64(9))
require.Equal(t, nodeUsage.Nodes, 1)
idx, usage, err := s.ServiceUsage(nil)
idx, usage, err := s.ServiceUsage(nil, true)
require.NoError(t, err)
require.Equal(t, idx, uint64(9))
require.Equal(t, usage.Services, 1)
@ -425,7 +425,7 @@ func TestStateStore_Usage_ServiceUsage_updatingService(t *testing.T) {
require.NoError(t, s.EnsureService(2, "node1", svc))
// We renamed a service with a single instance, so we maintain 1 service.
idx, usage, err := s.ServiceUsage(nil)
idx, usage, err := s.ServiceUsage(nil, true)
require.NoError(t, err)
require.Equal(t, idx, uint64(2))
require.Equal(t, usage.Services, 1)
@ -446,7 +446,7 @@ func TestStateStore_Usage_ServiceUsage_updatingService(t *testing.T) {
require.NoError(t, s.EnsureService(3, "node1", svc))
// We renamed a service with a single instance, so we maintain 1 service.
idx, usage, err := s.ServiceUsage(nil)
idx, usage, err := s.ServiceUsage(nil, true)
require.NoError(t, err)
require.Equal(t, idx, uint64(3))
require.Equal(t, usage.Services, 1)
@ -468,7 +468,7 @@ func TestStateStore_Usage_ServiceUsage_updatingService(t *testing.T) {
require.NoError(t, s.EnsureService(4, "node1", svc))
// We renamed a service with a single instance, so we maintain 1 service.
idx, usage, err := s.ServiceUsage(nil)
idx, usage, err := s.ServiceUsage(nil, true)
require.NoError(t, err)
require.Equal(t, idx, uint64(4))
require.Equal(t, usage.Services, 1)
@ -500,7 +500,7 @@ func TestStateStore_Usage_ServiceUsage_updatingService(t *testing.T) {
}
require.NoError(t, s.EnsureService(6, "node1", svc3))
idx, usage, err := s.ServiceUsage(nil)
idx, usage, err := s.ServiceUsage(nil, true)
require.NoError(t, err)
require.Equal(t, idx, uint64(6))
require.Equal(t, usage.Services, 2)
@ -519,7 +519,7 @@ func TestStateStore_Usage_ServiceUsage_updatingService(t *testing.T) {
}
require.NoError(t, s.EnsureService(7, "node1", update))
idx, usage, err = s.ServiceUsage(nil)
idx, usage, err = s.ServiceUsage(nil, true)
require.NoError(t, err)
require.Equal(t, idx, uint64(7))
require.Equal(t, usage.Services, 3)
@ -546,7 +546,7 @@ func TestStateStore_Usage_ServiceUsage_updatingConnectProxy(t *testing.T) {
require.NoError(t, s.EnsureService(2, "node1", svc))
// We renamed a service with a single instance, so we maintain 1 service.
idx, usage, err := s.ServiceUsage(nil)
idx, usage, err := s.ServiceUsage(nil, true)
require.NoError(t, err)
require.Equal(t, idx, uint64(2))
require.Equal(t, usage.Services, 1)
@ -573,7 +573,7 @@ func TestStateStore_Usage_ServiceUsage_updatingConnectProxy(t *testing.T) {
}
require.NoError(t, s.EnsureService(4, "node1", svc3))
idx, usage, err := s.ServiceUsage(nil)
idx, usage, err := s.ServiceUsage(nil, true)
require.NoError(t, err)
require.Equal(t, idx, uint64(4))
require.Equal(t, usage.Services, 2)
@ -589,7 +589,7 @@ func TestStateStore_Usage_ServiceUsage_updatingConnectProxy(t *testing.T) {
}
require.NoError(t, s.EnsureService(5, "node1", update))
idx, usage, err = s.ServiceUsage(nil)
idx, usage, err = s.ServiceUsage(nil, true)
require.NoError(t, err)
require.Equal(t, idx, uint64(5))
require.Equal(t, usage.Services, 3)

View File

@ -226,7 +226,7 @@ func (u *UsageMetricsReporter) runOnce() {
u.emitPeeringUsage(peeringUsage)
_, serviceUsage, err := state.ServiceUsage(nil)
_, serviceUsage, err := state.ServiceUsage(nil, true)
if err != nil {
u.logger.Warn("failed to retrieve services from state store", "error", err)
}

View File

@ -187,7 +187,9 @@ func (c *Controller) countProxies(ctx context.Context) (<-chan error, uint32, er
ws.Add(store.AbandonCh())
var count uint32
_, usage, err := store.ServiceUsage(ws)
// we don't care about the per-tenant counts so avoid excessive cpu utilization
// and don't aggregate that information
_, usage, err := store.ServiceUsage(ws, false)
// Query failed? Wait for a while, and then go to the top of the loop to
// retry (unless the context is cancelled).
@ -209,5 +211,5 @@ func (c *Controller) countProxies(ctx context.Context) (<-chan error, uint32, er
type Store interface {
AbandonCh() <-chan struct{}
ServiceUsage(ws memdb.WatchSet) (uint64, structs.ServiceUsage, error)
ServiceUsage(ws memdb.WatchSet, tenantUsage bool) (uint64, structs.ServiceUsage, error)
}

View File

@ -137,3 +137,56 @@ func TestCalcRateLimit(t *testing.T) {
require.Equalf(t, out, calcRateLimit(in), "calcRateLimit(%d)", in)
}
}
func BenchmarkCountProxies(b *testing.B) {
const index = 123
store := state.NewStateStore(nil)
// This loop generates:
//
// 4 (service kind) * 100 (service) * 5 * (node) = 2000 proxy services. And 500 non-proxy services.
for _, kind := range []structs.ServiceKind{
// These will be included in the count.
structs.ServiceKindConnectProxy,
structs.ServiceKindIngressGateway,
structs.ServiceKindTerminatingGateway,
structs.ServiceKindMeshGateway,
// This one will not.
structs.ServiceKindTypical,
} {
for i := 0; i < 100; i++ {
serviceName := fmt.Sprintf("%s-%d", kind, i)
for j := 0; j < 5; j++ {
nodeName := fmt.Sprintf("%s-node-%d", serviceName, j)
require.NoError(b, store.EnsureRegistration(index, &structs.RegisterRequest{
Node: nodeName,
Service: &structs.NodeService{
ID: serviceName,
Service: serviceName,
Kind: kind,
},
}))
}
}
}
ctx := testutil.TestContext(b)
c := &Controller{
cfg: Config{
GetStore: func() Store { return store },
},
}
b.ResetTimer()
for i := 0; i < b.N; i++ {
if _, numProxies, err := c.countProxies(ctx); err != nil {
b.Fatalf("encountered unexpected error: %v", err)
} else if numProxies != 2000 {
b.Fatalf("unexpected count: %d", numProxies)
}
}
}