public/grafana
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Alerting: Add compressed protobuf-based alert state storage (#99193)

Browse Source
This commit is contained in:
Alexander Akhmetov
2025-01-27 18:47:33 +01:00
committed by GitHub
parent 6edd4f5a7c
commit cb43f4b696
24 changed files with 1437 additions and 134 deletions
Download Patch File Download Diff File Expand all files Collapse all files
pkg/services/ngalert/state/manager_test.go
+25 -25
View File
@@ -47,7 +47,7 @@ func TestWarmStateCache(t *testing.T) {
evaluationTime, err := time.Parse("2006-01-02", "2021-03-25")
require.NoError(t, err)
ctx := context.Background()
_, dbstore := tests.SetupTestEnv(t, 1)
ng, dbstore := tests.SetupTestEnv(t, 1)
const mainOrgID int64 = 1
rule := tests.CreateTestAlertRule(t, ctx, dbstore, 600, mainOrgID)
@@ -216,13 +216,13 @@ func TestWarmStateCache(t *testing.T) {
ResultFingerprint: data.Fingerprint(2).String(),
})
for _, instance := range instances {
_ = dbstore.SaveAlertInstance(ctx, instance)
_ = ng.InstanceStore.SaveAlertInstance(ctx, instance)
}
cfg := state.ManagerCfg{
Metrics: metrics.NewNGAlert(prometheus.NewPedanticRegistry()).GetStateMetrics(),
ExternalURL: nil,
InstanceStore: dbstore,
InstanceStore: ng.InstanceStore,
Images: &state.NoopImageService{},
Clock: clock.NewMock(),
Historian: &state.FakeHistorian{},
@@ -230,7 +230,7 @@ func TestWarmStateCache(t *testing.T) {
Log: log.New("ngalert.state.manager"),
}
st := state.NewManager(cfg, state.NewNoopPersister())
st.Warm(ctx, dbstore, dbstore)
st.Warm(ctx, dbstore, ng.InstanceStore)
t.Run("instance cache has expected entries", func(t *testing.T) {
for _, entry := range expectedEntries {
@@ -250,7 +250,7 @@ func TestDashboardAnnotations(t *testing.T) {
require.NoError(t, err)
ctx := context.Background()
_, dbstore := tests.SetupTestEnv(t, 1)
ng, dbstore := tests.SetupTestEnv(t, 1)
fakeAnnoRepo := annotationstest.NewFakeAnnotationsRepo()
historianMetrics := metrics.NewHistorianMetrics(prometheus.NewRegistry(), metrics.Subsystem)
@@ -261,7 +261,7 @@ func TestDashboardAnnotations(t *testing.T) {
cfg := state.ManagerCfg{
Metrics: metrics.NewNGAlert(prometheus.NewPedanticRegistry()).GetStateMetrics(),
ExternalURL: nil,
InstanceStore: dbstore,
InstanceStore: ng.InstanceStore,
Images: &state.NoopImageService{},
Clock: clock.New(),
Historian: hist,
@@ -277,7 +277,7 @@ func TestDashboardAnnotations(t *testing.T) {
"test2": "{{ $labels.instance_label }}",
})
st.Warm(ctx, dbstore, dbstore)
st.Warm(ctx, dbstore, ng.InstanceStore)
bValue := float64(42)
cValue := float64(1)
_ = st.ProcessEvalResults(ctx, evaluationTime, rule, eval.Results{{
@@ -1697,7 +1697,7 @@ func TestStaleResultsHandler(t *testing.T) {
interval := time.Minute
ctx := context.Background()
_, dbstore := tests.SetupTestEnv(t, 1)
ng, dbstore := tests.SetupTestEnv(t, 1)
const mainOrgID int64 = 1
rule := tests.CreateTestAlertRule(t, ctx, dbstore, int64(interval.Seconds()), mainOrgID)
@@ -1751,7 +1751,7 @@ func TestStaleResultsHandler(t *testing.T) {
}
for _, instance := range instances {
_ = dbstore.SaveAlertInstance(ctx, instance)
_ = ng.InstanceStore.SaveAlertInstance(ctx, instance)
}
testCases := []struct {
@@ -1805,7 +1805,7 @@ func TestStaleResultsHandler(t *testing.T) {
cfg := state.ManagerCfg{
Metrics: metrics.NewNGAlert(prometheus.NewPedanticRegistry()).GetStateMetrics(),
ExternalURL: nil,
InstanceStore: dbstore,
InstanceStore: ng.InstanceStore,
Images: &state.NoopImageService{},
Clock: clock.New(),
Historian: &state.FakeHistorian{},
@@ -1813,7 +1813,7 @@ func TestStaleResultsHandler(t *testing.T) {
Log: log.New("ngalert.state.manager"),
}
st := state.NewManager(cfg, state.NewNoopPersister())
st.Warm(ctx, dbstore, dbstore)
st.Warm(ctx, dbstore, ng.InstanceStore)
existingStatesForRule := st.GetStatesForRuleUID(rule.OrgID, rule.UID)
// We have loaded the expected number of entries from the db
@@ -1978,7 +1978,7 @@ func TestStaleResults(t *testing.T) {
func TestDeleteStateByRuleUID(t *testing.T) {
interval := time.Minute
ctx := context.Background()
_, dbstore := tests.SetupTestEnv(t, 1)
ng, dbstore := tests.SetupTestEnv(t, 1)
const mainOrgID int64 = 1
rule := tests.CreateTestAlertRule(t, ctx, dbstore, int64(interval.Seconds()), mainOrgID)
@@ -2009,7 +2009,7 @@ func TestDeleteStateByRuleUID(t *testing.T) {
}
for _, instance := range instances {
_ = dbstore.SaveAlertInstance(ctx, instance)
_ = ng.InstanceStore.SaveAlertInstance(ctx, instance)
}
testCases := []struct {
@@ -2025,7 +2025,7 @@ func TestDeleteStateByRuleUID(t *testing.T) {
}{
{
desc: "all states/instances are removed from cache and DB",
instanceStore: dbstore,
instanceStore: ng.InstanceStore,
expectedStates: []*state.State{
{
AlertRuleUID: rule.UID,
@@ -2065,7 +2065,7 @@ func TestDeleteStateByRuleUID(t *testing.T) {
cfg := state.ManagerCfg{
Metrics: metrics.NewNGAlert(prometheus.NewPedanticRegistry()).GetStateMetrics(),
ExternalURL: nil,
InstanceStore: dbstore,
InstanceStore: ng.InstanceStore,
Images: &state.NoopImageService{},
Clock: clk,
Historian: &state.FakeHistorian{},
@@ -2073,9 +2073,9 @@ func TestDeleteStateByRuleUID(t *testing.T) {
Log: log.New("ngalert.state.manager"),
}
st := state.NewManager(cfg, state.NewNoopPersister())
st.Warm(ctx, dbstore, dbstore)
st.Warm(ctx, dbstore, ng.InstanceStore)
q := &models.ListAlertInstancesQuery{RuleOrgID: rule.OrgID, RuleUID: rule.UID}
alerts, _ := dbstore.ListAlertInstances(ctx, q)
alerts, _ := ng.InstanceStore.ListAlertInstances(ctx, q)
existingStatesForRule := st.GetStatesForRuleUID(rule.OrgID, rule.UID)
// We have loaded the expected number of entries from the db
@@ -2107,7 +2107,7 @@ func TestDeleteStateByRuleUID(t *testing.T) {
}
q = &models.ListAlertInstancesQuery{RuleOrgID: rule.OrgID, RuleUID: rule.UID}
alertInstances, _ := dbstore.ListAlertInstances(ctx, q)
alertInstances, _ := ng.InstanceStore.ListAlertInstances(ctx, q)
existingStatesForRule = st.GetStatesForRuleUID(rule.OrgID, rule.UID)
// The expected number of state entries remains after states are deleted
@@ -2120,7 +2120,7 @@ func TestDeleteStateByRuleUID(t *testing.T) {
func TestResetStateByRuleUID(t *testing.T) {
interval := time.Minute
ctx := context.Background()
_, dbstore := tests.SetupTestEnv(t, 1)
ng, dbstore := tests.SetupTestEnv(t, 1)
const mainOrgID int64 = 1
rule := tests.CreateTestAlertRule(t, ctx, dbstore, int64(interval.Seconds()), mainOrgID)
@@ -2151,7 +2151,7 @@ func TestResetStateByRuleUID(t *testing.T) {
}
for _, instance := range instances {
_ = dbstore.SaveAlertInstance(ctx, instance)
_ = ng.InstanceStore.SaveAlertInstance(ctx, instance)
}
testCases := []struct {
@@ -2168,7 +2168,7 @@ func TestResetStateByRuleUID(t *testing.T) {
}{
{
desc: "all states/instances are removed from cache and DB and saved in historian",
instanceStore: dbstore,
instanceStore: ng.InstanceStore,
expectedStates: []*state.State{
{
AlertRuleUID: rule.UID,
@@ -2206,7 +2206,7 @@ func TestResetStateByRuleUID(t *testing.T) {
cfg := state.ManagerCfg{
Metrics: metrics.NewNGAlert(prometheus.NewPedanticRegistry()).GetStateMetrics(),
ExternalURL: nil,
InstanceStore: dbstore,
InstanceStore: ng.InstanceStore,
Images: &state.NoopImageService{},
Clock: clk,
Historian: fakeHistorian,
@@ -2214,9 +2214,9 @@ func TestResetStateByRuleUID(t *testing.T) {
Log: log.New("ngalert.state.manager"),
}
st := state.NewManager(cfg, state.NewNoopPersister())
st.Warm(ctx, dbstore, dbstore)
st.Warm(ctx, dbstore, ng.InstanceStore)
q := &models.ListAlertInstancesQuery{RuleOrgID: rule.OrgID, RuleUID: rule.UID}
alerts, _ := dbstore.ListAlertInstances(ctx, q)
alerts, _ := ng.InstanceStore.ListAlertInstances(ctx, q)
existingStatesForRule := st.GetStatesForRuleUID(rule.OrgID, rule.UID)
// We have loaded the expected number of entries from the db
@@ -2251,7 +2251,7 @@ func TestResetStateByRuleUID(t *testing.T) {
assert.Equal(t, transitions, fakeHistorian.StateTransitions)
q = &models.ListAlertInstancesQuery{RuleOrgID: rule.OrgID, RuleUID: rule.UID}
alertInstances, _ := dbstore.ListAlertInstances(ctx, q)
alertInstances, _ := ng.InstanceStore.ListAlertInstances(ctx, q)
existingStatesForRule = st.GetStatesForRuleUID(rule.OrgID, rule.UID)
// The expected number of state entries remains after states are deleted