2020-06-27 17:26:46 +09:00
package controllers
import (
"context"
2020-07-03 09:05:46 +09:00
"errors"
2020-06-27 17:26:46 +09:00
"fmt"
2020-12-14 21:38:01 -08:00
"math"
2020-12-12 15:48:19 -08:00
"strconv"
2020-06-27 17:26:46 +09:00
"strings"
2021-02-07 17:37:27 +09:00
"time"
2020-10-07 17:00:44 -07:00
"github.com/summerwind/actions-runner-controller/api/v1alpha1"
2021-03-05 10:15:39 +09:00
kerrors "k8s.io/apimachinery/pkg/api/errors"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
2020-12-12 15:48:19 -08:00
"sigs.k8s.io/controller-runtime/pkg/client"
)
const (
defaultScaleUpThreshold = 0.8
defaultScaleDownThreshold = 0.3
defaultScaleUpFactor = 1.3
defaultScaleDownFactor = 0.7
2020-06-27 17:26:46 +09:00
)
2021-02-07 17:37:27 +09:00
func getValueAvailableAt ( now time . Time , from , to * time . Time , reservedValue int ) * int {
if to != nil && now . After ( * to ) {
return nil
}
if from != nil && now . Before ( * from ) {
return nil
}
return & reservedValue
}
Do not delay min/maxReplicas propagation from HRA to RD due to caching (#406)
As part of #282, I have introduced some caching mechanism to avoid excessive GitHub API calls due to the autoscaling calculation involving GitHub API calls is executed on each Webhook event.
Apparently, it was saving the wrong value in the cache- The value was one after applying `HRA.Spec.{Max,Min}Replicas` so manual changes to {Max,Min}Replicas doesn't affect RunnerDeployment.Spec.Replicas until the cache expires. This isn't what I had wanted.
This patch fixes that, by changing the value being cached to one before applying {Min,Max}Replicas.
Additionally, I've also updated logging so that you observe which number was fetched from cache, and what number was suggested by either TotalNumberOfQueuedAndInProgressWorkflowRuns or PercentageRunnersBusy, and what was the final number used as the desired-replicas(after applying {Min,Max}Replicas).
Follow-up for #282
2021-03-19 12:58:02 +09:00
func ( r * HorizontalRunnerAutoscalerReconciler ) fetchSuggestedReplicasFromCache ( hra v1alpha1 . HorizontalRunnerAutoscaler ) * int {
2021-02-07 17:37:27 +09:00
var entry * v1alpha1 . CacheEntry
for i := range hra . Status . CacheEntries {
ent := hra . Status . CacheEntries [ i ]
if ent . Key != v1alpha1 . CacheEntryKeyDesiredReplicas {
continue
}
if ! time . Now ( ) . Before ( ent . ExpirationTime . Time ) {
continue
}
entry = & ent
break
}
if entry != nil {
v := getValueAvailableAt ( time . Now ( ) , nil , & entry . ExpirationTime . Time , entry . Value )
if v != nil {
return v
}
}
return nil
}
Do not delay min/maxReplicas propagation from HRA to RD due to caching (#406)
As part of #282, I have introduced some caching mechanism to avoid excessive GitHub API calls due to the autoscaling calculation involving GitHub API calls is executed on each Webhook event.
Apparently, it was saving the wrong value in the cache- The value was one after applying `HRA.Spec.{Max,Min}Replicas` so manual changes to {Max,Min}Replicas doesn't affect RunnerDeployment.Spec.Replicas until the cache expires. This isn't what I had wanted.
This patch fixes that, by changing the value being cached to one before applying {Min,Max}Replicas.
Additionally, I've also updated logging so that you observe which number was fetched from cache, and what number was suggested by either TotalNumberOfQueuedAndInProgressWorkflowRuns or PercentageRunnersBusy, and what was the final number used as the desired-replicas(after applying {Min,Max}Replicas).
Follow-up for #282
2021-03-19 12:58:02 +09:00
func ( r * HorizontalRunnerAutoscalerReconciler ) suggestDesiredReplicas ( rd v1alpha1 . RunnerDeployment , hra v1alpha1 . HorizontalRunnerAutoscaler ) ( * int , error ) {
2020-07-19 18:42:06 +09:00
if hra . Spec . MinReplicas == nil {
return nil , fmt . Errorf ( "horizontalrunnerautoscaler %s/%s is missing minReplicas" , hra . Namespace , hra . Name )
} else if hra . Spec . MaxReplicas == nil {
return nil , fmt . Errorf ( "horizontalrunnerautoscaler %s/%s is missing maxReplicas" , hra . Namespace , hra . Name )
2020-06-27 17:26:46 +09:00
}
2020-12-12 15:48:19 -08:00
metrics := hra . Spec . Metrics
2021-05-05 14:27:17 +09:00
numMetrics := len ( metrics )
if numMetrics == 0 {
2021-03-14 11:03:00 +09:00
if len ( hra . Spec . ScaleUpTriggers ) == 0 {
2021-05-05 14:27:17 +09:00
return r . suggestReplicasByQueuedAndInProgressWorkflowRuns ( rd , hra , nil )
2021-03-14 11:03:00 +09:00
}
Do not delay min/maxReplicas propagation from HRA to RD due to caching (#406)
As part of #282, I have introduced some caching mechanism to avoid excessive GitHub API calls due to the autoscaling calculation involving GitHub API calls is executed on each Webhook event.
Apparently, it was saving the wrong value in the cache- The value was one after applying `HRA.Spec.{Max,Min}Replicas` so manual changes to {Max,Min}Replicas doesn't affect RunnerDeployment.Spec.Replicas until the cache expires. This isn't what I had wanted.
This patch fixes that, by changing the value being cached to one before applying {Min,Max}Replicas.
Additionally, I've also updated logging so that you observe which number was fetched from cache, and what number was suggested by either TotalNumberOfQueuedAndInProgressWorkflowRuns or PercentageRunnersBusy, and what was the final number used as the desired-replicas(after applying {Min,Max}Replicas).
Follow-up for #282
2021-03-19 12:58:02 +09:00
return nil , nil
2021-05-05 14:27:17 +09:00
} else if numMetrics > 2 {
return nil , fmt . Errorf ( "Too many autoscaling metrics configured: It must be 0 to 2, but got %d" , numMetrics )
}
primaryMetric := metrics [ 0 ]
primaryMetricType := primaryMetric . Type
var (
suggested * int
err error
)
switch primaryMetricType {
case v1alpha1 . AutoscalingMetricTypeTotalNumberOfQueuedAndInProgressWorkflowRuns :
suggested , err = r . suggestReplicasByQueuedAndInProgressWorkflowRuns ( rd , hra , & primaryMetric )
case v1alpha1 . AutoscalingMetricTypePercentageRunnersBusy :
suggested , err = r . suggestReplicasByPercentageRunnersBusy ( rd , hra , primaryMetric )
default :
return nil , fmt . Errorf ( "validting autoscaling metrics: unsupported metric type %q" , primaryMetric )
}
if err != nil {
return nil , err
}
if suggested != nil && * suggested > 0 {
return suggested , nil
2020-12-12 15:48:19 -08:00
}
2021-05-05 14:27:17 +09:00
if len ( metrics ) == 1 {
// This is never supposed to happen but anyway-
// Fall-back to `minReplicas + capacityReservedThroughWebhook`.
return nil , nil
}
// At this point, we are sure that there are exactly 2 Metrics entries.
fallbackMetric := metrics [ 1 ]
fallbackMetricType := fallbackMetric . Type
if primaryMetricType != v1alpha1 . AutoscalingMetricTypePercentageRunnersBusy ||
fallbackMetricType != v1alpha1 . AutoscalingMetricTypeTotalNumberOfQueuedAndInProgressWorkflowRuns {
return nil , fmt . Errorf (
"invalid HRA Spec: Metrics[0] of %s cannot be combined with Metrics[1] of %s: The only allowed combination is 0=PercentageRunnersBusy and 1=TotalNumberOfQueuedAndInProgressWorkflowRuns" ,
primaryMetricType , fallbackMetricType ,
)
}
return r . suggestReplicasByQueuedAndInProgressWorkflowRuns ( rd , hra , & fallbackMetric )
2020-12-12 15:48:19 -08:00
}
2021-05-05 14:27:17 +09:00
func ( r * HorizontalRunnerAutoscalerReconciler ) suggestReplicasByQueuedAndInProgressWorkflowRuns ( rd v1alpha1 . RunnerDeployment , hra v1alpha1 . HorizontalRunnerAutoscaler , metrics * v1alpha1 . MetricSpec ) ( * int , error ) {
2020-06-27 17:26:46 +09:00
2020-12-12 15:48:19 -08:00
var repos [ ] [ ] string
2020-06-27 17:26:46 +09:00
repoID := rd . Spec . Template . Spec . Repository
if repoID == "" {
2020-07-03 09:05:46 +09:00
orgName := rd . Spec . Template . Spec . Organization
if orgName == "" {
return nil , fmt . Errorf ( "asserting runner deployment spec to detect bug: spec.template.organization should not be empty on this code path" )
}
2021-03-09 15:03:47 +09:00
// In case it's an organizational runners deployment without any scaling metrics defined,
// we assume that the desired replicas should always be `minReplicas + capacityReservedThroughWebhook`.
// See https://github.com/summerwind/actions-runner-controller/issues/377#issuecomment-793372693
2021-05-05 14:27:17 +09:00
if metrics == nil {
Do not delay min/maxReplicas propagation from HRA to RD due to caching (#406)
As part of #282, I have introduced some caching mechanism to avoid excessive GitHub API calls due to the autoscaling calculation involving GitHub API calls is executed on each Webhook event.
Apparently, it was saving the wrong value in the cache- The value was one after applying `HRA.Spec.{Max,Min}Replicas` so manual changes to {Max,Min}Replicas doesn't affect RunnerDeployment.Spec.Replicas until the cache expires. This isn't what I had wanted.
This patch fixes that, by changing the value being cached to one before applying {Min,Max}Replicas.
Additionally, I've also updated logging so that you observe which number was fetched from cache, and what number was suggested by either TotalNumberOfQueuedAndInProgressWorkflowRuns or PercentageRunnersBusy, and what was the final number used as the desired-replicas(after applying {Min,Max}Replicas).
Follow-up for #282
2021-03-19 12:58:02 +09:00
return nil , nil
2021-03-09 15:03:47 +09:00
}
2021-05-05 14:27:17 +09:00
if len ( metrics . RepositoryNames ) == 0 {
2020-07-03 09:05:46 +09:00
return nil , errors . New ( "validating autoscaling metrics: spec.autoscaling.metrics[].repositoryNames is required and must have one more more entries for organizational runner deployment" )
}
2021-05-05 14:27:17 +09:00
for _ , repoName := range metrics . RepositoryNames {
2020-07-03 09:05:46 +09:00
repos = append ( repos , [ ] string { orgName , repoName } )
}
} else {
repo := strings . Split ( repoID , "/" )
repos = append ( repos , repo )
2020-06-27 17:26:46 +09:00
}
var total , inProgress , queued , completed , unknown int
2020-10-07 17:00:44 -07:00
type callback func ( )
listWorkflowJobs := func ( user string , repoName string , runID int64 , fallback_cb callback ) {
if runID == 0 {
fallback_cb ( )
return
}
jobs , _ , err := r . GitHubClient . Actions . ListWorkflowJobs ( context . TODO ( ) , user , repoName , runID , nil )
if err != nil {
r . Log . Error ( err , "Error listing workflow jobs" )
fallback_cb ( )
} else if len ( jobs . Jobs ) == 0 {
fallback_cb ( )
} else {
for _ , job := range jobs . Jobs {
switch job . GetStatus ( ) {
case "completed" :
// We add a case for `completed` so it is not counted in `unknown`.
// And we do not increment the counter for completed because
// that counter only refers to workflows. The reason for
// this is because we do not get a list of jobs for
// completed workflows in order to keep the number of API
// calls to a minimum.
case "in_progress" :
inProgress ++
case "queued" :
queued ++
default :
unknown ++
}
}
}
}
2020-06-27 17:26:46 +09:00
2020-07-03 09:05:46 +09:00
for _ , repo := range repos {
user , repoName := repo [ 0 ] , repo [ 1 ]
2021-02-09 10:13:53 +09:00
workflowRuns , err := r . GitHubClient . ListRepositoryWorkflowRuns ( context . TODO ( ) , user , repoName )
2020-07-03 09:05:46 +09:00
if err != nil {
return nil , err
}
2021-02-09 10:13:53 +09:00
for _ , run := range workflowRuns {
2020-07-03 09:05:46 +09:00
total ++
// In May 2020, there are only 3 statuses.
// Follow the below links for more details:
// - https://developer.github.com/v3/actions/workflow-runs/#list-repository-workflow-runs
// - https://developer.github.com/v3/checks/runs/#create-a-check-run
2020-10-07 17:00:44 -07:00
switch run . GetStatus ( ) {
2020-07-03 09:05:46 +09:00
case "completed" :
completed ++
case "in_progress" :
2020-10-07 17:00:44 -07:00
listWorkflowJobs ( user , repoName , run . GetID ( ) , func ( ) { inProgress ++ } )
2020-07-03 09:05:46 +09:00
case "queued" :
2020-10-07 17:00:44 -07:00
listWorkflowJobs ( user , repoName , run . GetID ( ) , func ( ) { queued ++ } )
2020-07-03 09:05:46 +09:00
default :
unknown ++
}
2020-06-27 17:26:46 +09:00
}
}
necessaryReplicas := queued + inProgress
r . Log . V ( 1 ) . Info (
Do not delay min/maxReplicas propagation from HRA to RD due to caching (#406)
As part of #282, I have introduced some caching mechanism to avoid excessive GitHub API calls due to the autoscaling calculation involving GitHub API calls is executed on each Webhook event.
Apparently, it was saving the wrong value in the cache- The value was one after applying `HRA.Spec.{Max,Min}Replicas` so manual changes to {Max,Min}Replicas doesn't affect RunnerDeployment.Spec.Replicas until the cache expires. This isn't what I had wanted.
This patch fixes that, by changing the value being cached to one before applying {Min,Max}Replicas.
Additionally, I've also updated logging so that you observe which number was fetched from cache, and what number was suggested by either TotalNumberOfQueuedAndInProgressWorkflowRuns or PercentageRunnersBusy, and what was the final number used as the desired-replicas(after applying {Min,Max}Replicas).
Follow-up for #282
2021-03-19 12:58:02 +09:00
fmt . Sprintf ( "Suggested desired replicas of %d by TotalNumberOfQueuedAndInProgressWorkflowRuns" , necessaryReplicas ) ,
2020-06-27 17:26:46 +09:00
"workflow_runs_completed" , completed ,
"workflow_runs_in_progress" , inProgress ,
"workflow_runs_queued" , queued ,
"workflow_runs_unknown" , unknown ,
2021-02-16 12:44:51 +09:00
"namespace" , hra . Namespace ,
"runner_deployment" , rd . Name ,
"horizontal_runner_autoscaler" , hra . Name ,
2020-06-27 17:26:46 +09:00
)
Do not delay min/maxReplicas propagation from HRA to RD due to caching (#406)
As part of #282, I have introduced some caching mechanism to avoid excessive GitHub API calls due to the autoscaling calculation involving GitHub API calls is executed on each Webhook event.
Apparently, it was saving the wrong value in the cache- The value was one after applying `HRA.Spec.{Max,Min}Replicas` so manual changes to {Max,Min}Replicas doesn't affect RunnerDeployment.Spec.Replicas until the cache expires. This isn't what I had wanted.
This patch fixes that, by changing the value being cached to one before applying {Min,Max}Replicas.
Additionally, I've also updated logging so that you observe which number was fetched from cache, and what number was suggested by either TotalNumberOfQueuedAndInProgressWorkflowRuns or PercentageRunnersBusy, and what was the final number used as the desired-replicas(after applying {Min,Max}Replicas).
Follow-up for #282
2021-03-19 12:58:02 +09:00
return & necessaryReplicas , nil
2020-06-27 17:26:46 +09:00
}
2020-12-12 15:48:19 -08:00
2021-05-05 14:27:17 +09:00
func ( r * HorizontalRunnerAutoscalerReconciler ) suggestReplicasByPercentageRunnersBusy ( rd v1alpha1 . RunnerDeployment , hra v1alpha1 . HorizontalRunnerAutoscaler , metrics v1alpha1 . MetricSpec ) ( * int , error ) {
2020-12-12 15:48:19 -08:00
ctx := context . Background ( )
scaleUpThreshold := defaultScaleUpThreshold
scaleDownThreshold := defaultScaleDownThreshold
scaleUpFactor := defaultScaleUpFactor
scaleDownFactor := defaultScaleDownFactor
if metrics . ScaleUpThreshold != "" {
sut , err := strconv . ParseFloat ( metrics . ScaleUpThreshold , 64 )
if err != nil {
return nil , errors . New ( "validating autoscaling metrics: spec.autoscaling.metrics[].scaleUpThreshold cannot be parsed into a float64" )
}
scaleUpThreshold = sut
}
if metrics . ScaleDownThreshold != "" {
sdt , err := strconv . ParseFloat ( metrics . ScaleDownThreshold , 64 )
if err != nil {
return nil , errors . New ( "validating autoscaling metrics: spec.autoscaling.metrics[].scaleDownThreshold cannot be parsed into a float64" )
}
scaleDownThreshold = sdt
}
2021-02-16 17:16:26 +09:00
scaleUpAdjustment := metrics . ScaleUpAdjustment
if scaleUpAdjustment != 0 {
if metrics . ScaleUpAdjustment < 0 {
return nil , errors . New ( "validating autoscaling metrics: spec.autoscaling.metrics[].scaleUpAdjustment cannot be lower than 0" )
}
if metrics . ScaleUpFactor != "" {
return nil , errors . New ( "validating autoscaling metrics: spec.autoscaling.metrics[]: scaleUpAdjustment and scaleUpFactor cannot be specified together" )
}
} else if metrics . ScaleUpFactor != "" {
2020-12-12 15:48:19 -08:00
suf , err := strconv . ParseFloat ( metrics . ScaleUpFactor , 64 )
if err != nil {
return nil , errors . New ( "validating autoscaling metrics: spec.autoscaling.metrics[].scaleUpFactor cannot be parsed into a float64" )
}
scaleUpFactor = suf
}
2021-02-16 17:16:26 +09:00
scaleDownAdjustment := metrics . ScaleDownAdjustment
if scaleDownAdjustment != 0 {
if metrics . ScaleDownAdjustment < 0 {
return nil , errors . New ( "validating autoscaling metrics: spec.autoscaling.metrics[].scaleDownAdjustment cannot be lower than 0" )
}
if metrics . ScaleDownFactor != "" {
return nil , errors . New ( "validating autoscaling metrics: spec.autoscaling.metrics[]: scaleDownAdjustment and scaleDownFactor cannot be specified together" )
}
} else if metrics . ScaleDownFactor != "" {
2020-12-12 15:48:19 -08:00
sdf , err := strconv . ParseFloat ( metrics . ScaleDownFactor , 64 )
if err != nil {
return nil , errors . New ( "validating autoscaling metrics: spec.autoscaling.metrics[].scaleDownFactor cannot be parsed into a float64" )
}
scaleDownFactor = sdf
}
2021-03-11 20:16:36 +09:00
// return the list of runners in namespace. Horizontal Runner Autoscaler should only be responsible for scaling resources in its own ns.
var runnerList v1alpha1 . RunnerList
var opts [ ] client . ListOption
opts = append ( opts , client . InNamespace ( rd . Namespace ) )
selector , err := metav1 . LabelSelectorAsSelector ( getSelector ( & rd ) )
2021-03-05 10:15:39 +09:00
if err != nil {
return nil , err
}
2021-03-11 20:16:36 +09:00
opts = append ( opts , client . MatchingLabelsSelector { Selector : selector } )
r . Log . V ( 2 ) . Info ( "Finding runners with selector" , "ns" , rd . Namespace )
2021-03-05 10:15:39 +09:00
if err := r . List (
ctx ,
& runnerList ,
2021-03-11 20:16:36 +09:00
opts ... ,
2021-03-05 10:15:39 +09:00
) ; err != nil {
if ! kerrors . IsNotFound ( err ) {
return nil , err
}
2020-12-12 15:48:19 -08:00
}
2021-03-05 10:15:39 +09:00
2020-12-12 15:48:19 -08:00
runnerMap := make ( map [ string ] struct { } )
for _ , items := range runnerList . Items {
runnerMap [ items . Name ] = struct { } { }
}
2021-02-16 12:44:51 +09:00
var (
enterprise = rd . Spec . Template . Spec . Enterprise
organization = rd . Spec . Template . Spec . Organization
repository = rd . Spec . Template . Spec . Repository
)
2020-12-12 15:48:19 -08:00
// ListRunners will return all runners managed by GitHub - not restricted to ns
2021-02-16 12:44:51 +09:00
runners , err := r . GitHubClient . ListRunners (
ctx ,
enterprise ,
organization ,
repository )
2020-12-12 15:48:19 -08:00
if err != nil {
return nil , err
}
2021-03-05 10:21:20 +09:00
var desiredReplicasBefore int
if v := rd . Spec . Replicas ; v == nil {
desiredReplicasBefore = 1
} else {
desiredReplicasBefore = * v
}
var (
numRunners int
numRunnersRegistered int
numRunnersBusy int
)
numRunners = len ( runnerList . Items )
2020-12-12 15:48:19 -08:00
for _ , runner := range runners {
2021-03-05 10:21:20 +09:00
if _ , ok := runnerMap [ * runner . Name ] ; ok {
numRunnersRegistered ++
if runner . GetBusy ( ) {
numRunnersBusy ++
}
2020-12-12 15:48:19 -08:00
}
}
var desiredReplicas int
2021-03-05 10:21:20 +09:00
fractionBusy := float64 ( numRunnersBusy ) / float64 ( desiredReplicasBefore )
2020-12-12 15:48:19 -08:00
if fractionBusy >= scaleUpThreshold {
2021-02-16 17:16:26 +09:00
if scaleUpAdjustment > 0 {
2021-03-05 10:21:20 +09:00
desiredReplicas = desiredReplicasBefore + scaleUpAdjustment
2021-02-16 17:16:26 +09:00
} else {
2021-03-05 10:21:20 +09:00
desiredReplicas = int ( math . Ceil ( float64 ( desiredReplicasBefore ) * scaleUpFactor ) )
2021-02-16 17:16:26 +09:00
}
2020-12-12 15:48:19 -08:00
} else if fractionBusy < scaleDownThreshold {
2021-02-16 17:16:26 +09:00
if scaleDownAdjustment > 0 {
2021-03-05 10:21:20 +09:00
desiredReplicas = desiredReplicasBefore - scaleDownAdjustment
2021-02-16 17:16:26 +09:00
} else {
2021-03-05 10:21:20 +09:00
desiredReplicas = int ( float64 ( desiredReplicasBefore ) * scaleDownFactor )
2021-02-16 17:16:26 +09:00
}
2020-12-12 15:48:19 -08:00
} else {
desiredReplicas = * rd . Spec . Replicas
}
2021-01-24 10:58:35 +09:00
2021-03-05 10:21:20 +09:00
// NOTES for operators:
//
// - num_runners can be as twice as large as replicas_desired_before while
// the runnerdeployment controller is replacing RunnerReplicaSet for runner update.
2020-12-12 15:48:19 -08:00
r . Log . V ( 1 ) . Info (
Do not delay min/maxReplicas propagation from HRA to RD due to caching (#406)
As part of #282, I have introduced some caching mechanism to avoid excessive GitHub API calls due to the autoscaling calculation involving GitHub API calls is executed on each Webhook event.
Apparently, it was saving the wrong value in the cache- The value was one after applying `HRA.Spec.{Max,Min}Replicas` so manual changes to {Max,Min}Replicas doesn't affect RunnerDeployment.Spec.Replicas until the cache expires. This isn't what I had wanted.
This patch fixes that, by changing the value being cached to one before applying {Min,Max}Replicas.
Additionally, I've also updated logging so that you observe which number was fetched from cache, and what number was suggested by either TotalNumberOfQueuedAndInProgressWorkflowRuns or PercentageRunnersBusy, and what was the final number used as the desired-replicas(after applying {Min,Max}Replicas).
Follow-up for #282
2021-03-19 12:58:02 +09:00
fmt . Sprintf ( "Suggested desired replicas of %d by PercentageRunnersBusy" , desiredReplicas ) ,
2021-03-05 10:21:20 +09:00
"replicas_desired_before" , desiredReplicasBefore ,
"replicas_desired" , desiredReplicas ,
2020-12-12 15:48:19 -08:00
"num_runners" , numRunners ,
2021-03-05 10:21:20 +09:00
"num_runners_registered" , numRunnersRegistered ,
2020-12-12 15:48:19 -08:00
"num_runners_busy" , numRunnersBusy ,
2021-02-16 12:44:51 +09:00
"namespace" , hra . Namespace ,
"runner_deployment" , rd . Name ,
"horizontal_runner_autoscaler" , hra . Name ,
"enterprise" , enterprise ,
"organization" , organization ,
"repository" , repository ,
2020-12-12 15:48:19 -08:00
)
Do not delay min/maxReplicas propagation from HRA to RD due to caching (#406)
As part of #282, I have introduced some caching mechanism to avoid excessive GitHub API calls due to the autoscaling calculation involving GitHub API calls is executed on each Webhook event.
Apparently, it was saving the wrong value in the cache- The value was one after applying `HRA.Spec.{Max,Min}Replicas` so manual changes to {Max,Min}Replicas doesn't affect RunnerDeployment.Spec.Replicas until the cache expires. This isn't what I had wanted.
This patch fixes that, by changing the value being cached to one before applying {Min,Max}Replicas.
Additionally, I've also updated logging so that you observe which number was fetched from cache, and what number was suggested by either TotalNumberOfQueuedAndInProgressWorkflowRuns or PercentageRunnersBusy, and what was the final number used as the desired-replicas(after applying {Min,Max}Replicas).
Follow-up for #282
2021-03-19 12:58:02 +09:00
return & desiredReplicas , nil
2020-12-12 15:48:19 -08:00
}