Remove limit_by_label_key from rate-limiting and concurrency-limiting…

… configurations
fluxninja · Jan 5, 2024 · 3de6e40 · 3de6e40
1 parent 4d3fd6a
commit 3de6e40
Show file tree

Hide file tree

Showing 28 changed files with 54 additions and 73 deletions.
diff --git a/blueprints/concurrency-limiting/base/config.libsonnet b/blueprints/concurrency-limiting/base/config.libsonnet
@@ -13,7 +13,6 @@ commonConfig {
       max_concurrency: '__REQUIRED_FIELD__',
       selectors: commonConfig.selectors_defaults,
       parameters: {
-        limit_by_label_key: 'limit_by_label_key',
         max_inflight_duration: '__REQUIRED_FIELD__',
       },
       request_parameters: {},

diff --git a/blueprints/concurrency-limiting/base/gen/definitions.json b/blueprints/concurrency-limiting/base/gen/definitions.json
@@ -67,7 +67,6 @@
             "parameters": {
               "description": "Parameters.",
               "default": {
-                "limit_by_label_key": "limit_by_label_key",
                 "max_inflight_duration": "__REQUIRED_FIELD__"
               },
               "type": "object",

diff --git a/blueprints/concurrency-limiting/base/gen/values.yaml b/blueprints/concurrency-limiting/base/gen/values.yaml
@@ -29,7 +29,6 @@ policy:
     # Type: aperture.spec.v1.ConcurrencyLimiterParameters
     # Required: True
     parameters:
-      limit_by_label_key: "limit_by_label_key"
       max_inflight_duration: __REQUIRED_FIELD__
     # Request Parameters.
     # Type: aperture.spec.v1.ConcurrencyLimiterRequestParameters

diff --git a/blueprints/concurrency-scheduling/base/config.libsonnet b/blueprints/concurrency-scheduling/base/config.libsonnet
@@ -13,14 +13,9 @@ commonConfig {
       max_concurrency: '__REQUIRED_FIELD__',
       selectors: commonConfig.selectors_defaults,
       concurrency_limiter: {
-        limit_by_label_key: 'limit_by_label_key',
         max_inflight_duration: '__REQUIRED_FIELD__',
       },
-      scheduler: {
-        tokens_label_key: 'tokens',
-        priority_label_key: 'priority',
-        workload_label_key: 'workload',
-      },
+      scheduler: {},
       alerter: {
         alert_name: 'Too many inflight requests',
       },

diff --git a/blueprints/concurrency-scheduling/base/gen/definitions.json b/blueprints/concurrency-scheduling/base/gen/definitions.json
@@ -61,7 +61,6 @@
             "concurrency_limiter": {
               "description": "Concurrency Limiter Parameters.",
               "default": {
-                "limit_by_label_key": "limit_by_label_key",
                 "max_inflight_duration": "__REQUIRED_FIELD__"
               },
               "type": "object",
@@ -75,11 +74,7 @@
             },
             "scheduler": {
               "description": "Scheduler configuration.",
-              "default": {
-                "priority_label_key": "priority",
-                "tokens_label_key": "tokens",
-                "workload_label_key": "workload"
-              },
+              "default": {},
               "type": "object",
               "$ref": "../../../gen/jsonschema/_definitions.json#/definitions/Scheduler"
             },

diff --git a/blueprints/concurrency-scheduling/base/gen/values.yaml b/blueprints/concurrency-scheduling/base/gen/values.yaml
@@ -25,18 +25,14 @@ policy:
     # Type: aperture.spec.v1.ConcurrencyLimiterParameters
     # Required: True
     concurrency_limiter:
-      limit_by_label_key: "limit_by_label_key"
       max_inflight_duration: __REQUIRED_FIELD__
     # Max concurrency.
     # Type: float64
     # Required: True
     max_concurrency: __REQUIRED_FIELD__
     # Scheduler configuration.
     # Type: aperture.spec.v1.Scheduler
-    scheduler:
-      priority_label_key: "priority"
-      tokens_label_key: "tokens"
-      workload_label_key: "workload"
+    scheduler: {}
     # Flow selectors to match requests against.
     # Type: []aperture.spec.v1.Selector
     # Required: True

diff --git a/blueprints/quota-scheduling/base/config.libsonnet b/blueprints/quota-scheduling/base/config.libsonnet
@@ -15,14 +15,9 @@ commonConfig {
       fill_amount: '__REQUIRED_FIELD__',
       selectors: commonConfig.selectors_defaults,
       rate_limiter: {
-        limit_by_label_key: 'limit_key',
         interval: '__REQUIRED_FIELD__',
       },
-      scheduler: {
-        tokens_label_key: 'tokens',
-        priority_label_key: 'priority',
-        workload_label_key: 'workload',
-      },
+      scheduler: {},
       alerter: {
         alert_name: 'More than 90% of requests are being rate limited',
       },

diff --git a/blueprints/quota-scheduling/base/gen/definitions.json b/blueprints/quota-scheduling/base/gen/definitions.json
@@ -78,19 +78,14 @@
             "rate_limiter": {
               "description": "Rate Limiter Parameters.",
               "default": {
-                "interval": "__REQUIRED_FIELD__",
-                "limit_by_label_key": "limit_key"
+                "interval": "__REQUIRED_FIELD__"
               },
               "type": "object",
               "$ref": "../../../gen/jsonschema/_definitions.json#/definitions/RateLimiterParameters"
             },
             "scheduler": {
               "description": "Scheduler configuration.",
-              "default": {
-                "priority_label_key": "priority",
-                "tokens_label_key": "tokens",
-                "workload_label_key": "workload"
-              },
+              "default": {},
               "type": "object",
               "$ref": "../../../gen/jsonschema/_definitions.json#/definitions/Scheduler"
             },

diff --git a/blueprints/quota-scheduling/base/gen/values.yaml b/blueprints/quota-scheduling/base/gen/values.yaml
@@ -34,13 +34,9 @@ policy:
     # Required: True
     rate_limiter:
       interval: __REQUIRED_FIELD__
-      limit_by_label_key: "limit_key"
     # Scheduler configuration.
     # Type: aperture.spec.v1.Scheduler
-    scheduler:
-      priority_label_key: "priority"
-      tokens_label_key: "tokens"
-      workload_label_key: "workload"
+    scheduler: {}
     # Flow selectors to match requests against.
     # Type: []aperture.spec.v1.Selector
     # Required: True

diff --git a/blueprints/rate-limiting/base/config.libsonnet b/blueprints/rate-limiting/base/config.libsonnet
@@ -15,7 +15,6 @@ commonConfig {
       fill_amount: '__REQUIRED_FIELD__',
       selectors: commonConfig.selectors_defaults,
       parameters: {
-        limit_by_label_key: 'limit_key',
         interval: '__REQUIRED_FIELD__',
       },
       request_parameters: {},

diff --git a/blueprints/rate-limiting/base/gen/definitions.json b/blueprints/rate-limiting/base/gen/definitions.json
@@ -78,8 +78,7 @@
             "parameters": {
               "description": "Parameters.",
               "default": {
-                "interval": "__REQUIRED_FIELD__",
-                "limit_by_label_key": "limit_key"
+                "interval": "__REQUIRED_FIELD__"
               },
               "type": "object",
               "$ref": "../../../gen/jsonschema/_definitions.json#/definitions/RateLimiterParameters"

diff --git a/blueprints/rate-limiting/base/gen/values.yaml b/blueprints/rate-limiting/base/gen/values.yaml
@@ -34,7 +34,6 @@ policy:
     # Required: True
     parameters:
       interval: __REQUIRED_FIELD__
-      limit_by_label_key: "limit_key"
     # Request Parameters.
     # Type: aperture.spec.v1.RateLimiterRequestParameters
     request_parameters: {}

diff --git a/...ture-for-infra/guides/api-quota-management/assets/inter-service-rate-limiting/policy.yaml b/...ture-for-infra/guides/api-quota-management/assets/inter-service-rate-limiting/policy.yaml
@@ -26,9 +26,6 @@ spec:
               num_sync: 4
             limit_by_label_key: api_key
           scheduler:
-            priority_label_key: priority
-            tokens_label_key: tokens
-            workload_label_key: workload
             workloads:
             - label_matcher:
                 match_labels:

diff --git a/docs/content/guides/assets/managing-quotas/policy.yaml b/docs/content/guides/assets/managing-quotas/policy.yaml
@@ -24,7 +24,6 @@ spec:
             limit_by_label_key: user_id
           scheduler:
             priority_label_key: priority
-            tokens_label_key: tokens
             workload_label_key: workload
           selectors:
           - control_point: quota-scheduling-feature

diff --git a/docs/content/guides/assets/openai/policy.yaml b/docs/content/guides/assets/openai/policy.yaml
@@ -24,8 +24,6 @@ spec:
             limit_by_label_key: api_key
           scheduler:
             priority_label_key: priority
-            tokens_label_key: tokens
-            workload_label_key: workload
           selectors:
           - control_point: openai
             label_matcher:

diff --git a/docs/content/reference/aperture-cli/policies/assets/raw_values.yaml b/docs/content/reference/aperture-cli/policies/assets/raw_values.yaml
@@ -34,7 +34,6 @@ policy:
     # Required: True
     parameters:
       interval: __REQUIRED_FIELD__
-      limit_by_label_key: "limit_key"
     # Request Parameters.
     # Type: aperture.spec.v1.RateLimiterRequestParameters
     request_parameters: {}

diff --git a/docs/content/reference/blueprints/concurrency-limiting/base.md b/docs/content/reference/blueprints/concurrency-limiting/base.md
@@ -121,7 +121,7 @@ href={`https://github.com/fluxninja/aperture/tree/${aver}/blueprints/concurrency
     description='Parameters.'
     type='Object (aperture.spec.v1.ConcurrencyLimiterParameters)'
     reference='../../configuration/spec#concurrency-limiter-parameters'
-    value='{"limit_by_label_key": "limit_by_label_key", "max_inflight_duration": "__REQUIRED_FIELD__"}'
+    value='{"max_inflight_duration": "__REQUIRED_FIELD__"}'
 />
 
 <!-- vale on -->

diff --git a/docs/content/reference/blueprints/concurrency-scheduling/base.md b/docs/content/reference/blueprints/concurrency-scheduling/base.md
@@ -107,7 +107,7 @@ href={`https://github.com/fluxninja/aperture/tree/${aver}/blueprints/concurrency
     description='Concurrency Limiter Parameters.'
     type='Object (aperture.spec.v1.ConcurrencyLimiterParameters)'
     reference='../../configuration/spec#concurrency-limiter-parameters'
-    value='{"limit_by_label_key": "limit_by_label_key", "max_inflight_duration": "__REQUIRED_FIELD__"}'
+    value='{"max_inflight_duration": "__REQUIRED_FIELD__"}'
 />
 
 <!-- vale on -->
@@ -135,7 +135,7 @@ href={`https://github.com/fluxninja/aperture/tree/${aver}/blueprints/concurrency
     description='Scheduler configuration.'
     type='Object (aperture.spec.v1.Scheduler)'
     reference='../../configuration/spec#scheduler'
-    value='{"priority_label_key": "priority", "tokens_label_key": "tokens", "workload_label_key": "workload"}'
+    value='{}'
 />
 
 <!-- vale on -->

diff --git a/docs/content/reference/blueprints/quota-scheduling/base.md b/docs/content/reference/blueprints/quota-scheduling/base.md
@@ -135,7 +135,7 @@ href={`https://github.com/fluxninja/aperture/tree/${aver}/blueprints/quota-sched
     description='Rate Limiter Parameters.'
     type='Object (aperture.spec.v1.RateLimiterParameters)'
     reference='../../configuration/spec#rate-limiter-parameters'
-    value='{"interval": "__REQUIRED_FIELD__", "limit_by_label_key": "limit_key"}'
+    value='{"interval": "__REQUIRED_FIELD__"}'
 />
 
 <!-- vale on -->
@@ -149,7 +149,7 @@ href={`https://github.com/fluxninja/aperture/tree/${aver}/blueprints/quota-sched
     description='Scheduler configuration.'
     type='Object (aperture.spec.v1.Scheduler)'
     reference='../../configuration/spec#scheduler'
-    value='{"priority_label_key": "priority", "tokens_label_key": "tokens", "workload_label_key": "workload"}'
+    value='{}'
 />
 
 <!-- vale on -->

diff --git a/docs/content/reference/blueprints/rate-limiting/base.md b/docs/content/reference/blueprints/rate-limiting/base.md
@@ -135,7 +135,7 @@ href={`https://github.com/fluxninja/aperture/tree/${aver}/blueprints/rate-limiti
     description='Parameters.'
     type='Object (aperture.spec.v1.RateLimiterParameters)'
     reference='../../configuration/spec#rate-limiter-parameters'
-    value='{"interval": "__REQUIRED_FIELD__", "limit_by_label_key": "limit_key"}'
+    value='{"interval": "__REQUIRED_FIELD__"}'
 />
 
 <!-- vale on -->

diff --git a/pkg/scheduler/wfq.go b/pkg/scheduler/wfq.go
@@ -528,11 +528,32 @@ func (pMetrics *preemptionMetrics) onQueueEntry(request *Request, qRequest *queu
 // Update metrics for preemption and delay
 // WARNING: Unsafe and should be called with scheduler lock.
 func (pMetrics *preemptionMetrics) onQueueExit(request *Request, qRequest *queuedRequest, allowed bool) {
+	initMetrics := func(labels prometheus.Labels) error {
+		var err error
+		_, err = pMetrics.workloadPreemptedTokensSummary.GetMetricWith(labels)
+		if err != nil {
+			return fmt.Errorf("%w: failed to get workload_preempted_tokens summary", err)
+		}
+		_, err = pMetrics.workloadDelayedTokensSummary.GetMetricWith(labels)
+		if err != nil {
+			return fmt.Errorf("%w: failed to get workload_delayed_tokens summary", err)
+		}
+		_, err = pMetrics.workloadOnTimeCounter.GetMetricWith(labels)
+		if err != nil {
+			return fmt.Errorf("%w: failed to get workload_on_time_total counter", err)
+		}
+		return nil
+	}
+
 	publishSummary := func(summary *prometheus.SummaryVec, value float64) {
 		if summary == nil {
 			return
 		}
 		metricsLabels := appendWorkloadLabel(pMetrics.metricsLabels, request.FairnessLabel)
+		err := initMetrics(metricsLabels)
+		if err != nil {
+			log.Error().Err(err).Msg("Failed to initialize metrics")
+		}
 		observer, err := summary.GetMetricWith(metricsLabels)
 		if err != nil {
 			log.Error().Err(err).Msg("Failed to get workload preempted tokens summary")
@@ -546,6 +567,10 @@ func (pMetrics *preemptionMetrics) onQueueExit(request *Request, qRequest *queue
 			return
 		}
 		metricsLabels := appendWorkloadLabel(pMetrics.metricsLabels, request.FairnessLabel)
+		err := initMetrics(metricsLabels)
+		if err != nil {
+			log.Error().Err(err).Msg("Failed to initialize metrics")
+		}
 		counter, err := counterVec.GetMetricWith(metricsLabels)
 		if err != nil {
 			log.Error().Err(err).Msg("Failed to get workload on time counter")

diff --git a/playground/scenarios/concurrency-scheduler/policies/concurrency-scheduler-cr.yaml b/playground/scenarios/concurrency-scheduler/policies/concurrency-scheduler-cr.yaml
@@ -20,9 +20,6 @@ spec:
             accept_percentage:
               signal_name: ACCEPT_PERCENTAGE
           scheduler:
-            priority_label_key: priority
-            tokens_label_key: tokens
-            workload_label_key: workload
             workloads:
             - label_matcher:
                 match_labels:

diff --git a/playground/scenarios/postgres-concurrency/policies/concurrency-scheduling-cr.yaml b/playground/scenarios/postgres-concurrency/policies/concurrency-scheduling-cr.yaml
@@ -10,20 +10,18 @@ spec:
     - flow_control:
         concurrency_scheduler:
           concurrency_limiter:
-            limit_by_label_key: limit_by_label_key
             max_idle_time: 7200s
             max_inflight_duration: 60s
           in_ports:
             max_concurrency:
               constant_signal:
-                value: 10
+                value: 2
           out_ports:
             accept_percentage:
               signal_name: ACCEPT_PERCENTAGE
           scheduler:
             denied_response_status_code: BadRequest
             priority_label_key: priority
-            tokens_label_key: tokens
             workload_label_key: userType
           selectors:
           - agent_group: default

diff --git a/playground/scenarios/postgres-concurrency/policies/concurrency-scheduling.yaml b/playground/scenarios/postgres-concurrency/policies/concurrency-scheduling.yaml
@@ -17,7 +17,7 @@ policy:
     concurrency_limiter:
       max_inflight_duration: 60s
       max_idle_time: "7200s"
-    max_concurrency: 10
+    max_concurrency: 2
     scheduler:
       workload_label_key: "userType"
       priority_label_key: "priority"

diff --git a/playground/scenarios/quota-scheduler/policies/quota-scheduler-cr.yaml b/playground/scenarios/quota-scheduler/policies/quota-scheduler-cr.yaml
@@ -26,9 +26,6 @@ spec:
               num_sync: 4
             limit_by_label_key: http.request.header.api_key
           scheduler:
-            priority_label_key: priority
-            tokens_label_key: tokens
-            workload_label_key: workload
             workloads:
             - label_matcher:
                 match_labels:

diff --git a/sdks/aperture-go/examples/manual/go.mod b/sdks/aperture-go/examples/manual/go.mod
@@ -12,7 +12,7 @@ require (
 require (
 	github.com/cenkalti/backoff/v4 v4.2.1 // indirect
 	github.com/envoyproxy/protoc-gen-validate v1.0.2 // indirect
-	github.com/fluxninja/aperture/api/v2 v2.0.0-20240101092959-235caef8dbad // indirect
+	github.com/fluxninja/aperture/api/v2 v2.0.0-20240104011105-0680993ed282 // indirect
 	github.com/go-logr/logr v1.3.0 // indirect
 	github.com/go-logr/stdr v1.2.2 // indirect
 	github.com/golang/protobuf v1.5.3 // indirect

diff --git a/sdks/aperture-go/examples/manual/go.sum b/sdks/aperture-go/examples/manual/go.sum
@@ -4,8 +4,8 @@ github.com/davecgh/go-spew v1.1.2-0.20180830191138-d8f796af33cc h1:U9qPSI2PIWSS1
 github.com/davecgh/go-spew v1.1.2-0.20180830191138-d8f796af33cc/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/envoyproxy/protoc-gen-validate v1.0.2 h1:QkIBuU5k+x7/QXPvPPnWXWlCdaBFApVqftFV6k087DA=
 github.com/envoyproxy/protoc-gen-validate v1.0.2/go.mod h1:GpiZQP3dDbg4JouG/NNS7QWXpgx6x8QiMKdmN72jogE=
-github.com/fluxninja/aperture/api/v2 v2.0.0-20240101092959-235caef8dbad h1:J4SgDerz6NPEARbKEXvR5nAjSN1DWJ2J3NnzJKM+JcU=
-github.com/fluxninja/aperture/api/v2 v2.0.0-20240101092959-235caef8dbad/go.mod h1:KSjIteqXmGJl1WOxQeBF9/K6/0sMHfKsRl5VOQkxyNg=
+github.com/fluxninja/aperture/api/v2 v2.0.0-20240104011105-0680993ed282 h1:3UK/8Z97YO2o6WutDRmKiWLeURT/x5+aFTBIIf/fh4Q=
+github.com/fluxninja/aperture/api/v2 v2.0.0-20240104011105-0680993ed282/go.mod h1:KSjIteqXmGJl1WOxQeBF9/K6/0sMHfKsRl5VOQkxyNg=
 github.com/go-logr/logr v1.2.2/go.mod h1:jdQByPbusPIv2/zmleS9BjJVeZ6kBagPoEUsqbVz/1A=
 github.com/go-logr/logr v1.3.0 h1:2y3SDp0ZXuc6/cjLSZ+Q3ir+QB9T/iG5yYRXqsagWSY=
 github.com/go-logr/logr v1.3.0/go.mod h1:9T104GzyrTigFIr8wt5mBrctHMim0Nb2HLGrmQ40KvY=