From ff1535d4ea15b403f26240e2b045b7d86fca0ab2 Mon Sep 17 00:00:00 2001 From: Abhimanyu Sharma Date: Tue, 10 Sep 2024 18:46:17 +0530 Subject: [PATCH] feat(chart): exposed retry constants as helm chart --- .../templates/seldon-v2-components.yaml | 32 +++++++++++++++++++ .../seldon-core-v2-setup/values.yaml | 7 ++++ .../seldon-core-v2-setup/values.yaml.template | 7 ++++ .../helm-components-sc/patch_mlserver.yaml | 14 ++++++++ .../helm-components-sc/patch_triton.yaml | 14 ++++++++ k8s/yaml/components.yaml | 14 ++++++++ 6 files changed, 88 insertions(+) diff --git a/k8s/helm-charts/seldon-core-v2-setup/templates/seldon-v2-components.yaml b/k8s/helm-charts/seldon-core-v2-setup/templates/seldon-v2-components.yaml index 7fdfb7fdbe..6b1e08a4fc 100644 --- a/k8s/helm-charts/seldon-core-v2-setup/templates/seldon-v2-components.yaml +++ b/k8s/helm-charts/seldon-core-v2-setup/templates/seldon-v2-components.yaml @@ -1539,6 +1539,22 @@ spec: value: '{{ .Values.serverConfig.agent.modelInactiveSecondsThreshold }}' - name: SELDON_SCALING_STATS_PERIOD_SECONDS value: '{{ .Values.serverConfig.agent.scalingStatsPeriodSeconds }}' + - name: SELDON_MAX_TIME_READY_SUB_SERVICE_AFTER_START_SECONDS + value: '{{ .Values.serverConfig.agent.maxElapsedTimeReadySubServiceAfterStartSeconds + }}' + - name: SELDON_MAX_ELAPSED_TIME_READY_SUB_SERVICE_BEFORE_START_MINUTES + value: '{{ .Values.serverConfig.agent.maxElapsedTimeReadySubServiceBeforeStartMinutes + }}' + - name: SELDON_PERIOD_READY_SUB_SERVICE_SECONDS + value: '{{ .Values.serverConfig.agent.periodReadySubServiceSeconds }}' + - name: SELDON_MAX_LOAD_ELAPSED_TIME_MINUTES + value: '{{ .Values.serverConfig.agent.maxLoadElapsedTimeMinutes }}' + - name: SELDON_MAX_UNLOAD_ELAPSED_TIME_MINUTES + value: '{{ .Values.serverConfig.agent.maxUnloadElapsedTimeMinutes }}' + - name: SELDON_MAX_LOAD_RETRY_COUNT + value: '{{ .Values.serverConfig.agent.maxLoadRetryCount }}' + - name: SELDON_MAX_UNLOAD_RETRY_COUNT + value: '{{ .Values.serverConfig.agent.maxUnloadRetryCount }}' - name: SELDON_OVERCOMMIT_PERCENTAGE value: '{{ .Values.serverConfig.agent.overcommitPercentage }}' - name: CONTROL_PLANE_SECURITY_PROTOCOL @@ -1787,6 +1803,22 @@ spec: value: '{{ .Values.serverConfig.agent.modelInactiveSecondsThreshold }}' - name: SELDON_SCALING_STATS_PERIOD_SECONDS value: '{{ .Values.serverConfig.agent.scalingStatsPeriodSeconds }}' + - name: SELDON_MAX_TIME_READY_SUB_SERVICE_AFTER_START_SECONDS + value: '{{ .Values.serverConfig.agent.maxElapsedTimeReadySubServiceAfterStartSeconds + }}' + - name: SELDON_MAX_ELAPSED_TIME_READY_SUB_SERVICE_BEFORE_START_MINUTES + value: '{{ .Values.serverConfig.agent.maxElapsedTimeReadySubServiceBeforeStartMinutes + }}' + - name: SELDON_PERIOD_READY_SUB_SERVICE_SECONDS + value: '{{ .Values.serverConfig.agent.periodReadySubServiceSeconds }}' + - name: SELDON_MAX_LOAD_ELAPSED_TIME_MINUTES + value: '{{ .Values.serverConfig.agent.maxLoadElapsedTimeMinutes }}' + - name: SELDON_MAX_UNLOAD_ELAPSED_TIME_MINUTES + value: '{{ .Values.serverConfig.agent.maxUnloadElapsedTimeMinutes }}' + - name: SELDON_MAX_LOAD_RETRY_COUNT + value: '{{ .Values.serverConfig.agent.maxLoadRetryCount }}' + - name: SELDON_MAX_UNLOAD_RETRY_COUNT + value: '{{ .Values.serverConfig.agent.maxUnloadRetryCount }}' - name: SELDON_OVERCOMMIT_PERCENTAGE value: '{{ .Values.serverConfig.agent.overcommitPercentage }}' - name: CONTROL_PLANE_SECURITY_PROTOCOL diff --git a/k8s/helm-charts/seldon-core-v2-setup/values.yaml b/k8s/helm-charts/seldon-core-v2-setup/values.yaml index 0e861042d5..1452e49fc7 100644 --- a/k8s/helm-charts/seldon-core-v2-setup/values.yaml +++ b/k8s/helm-charts/seldon-core-v2-setup/values.yaml @@ -243,6 +243,13 @@ serverConfig: scalingStatsPeriodSeconds: "20" modelInferenceLagThreshold: "30" modelInactiveSecondsThreshold: "600" + maxElapsedTimeReadySubServiceAfterStartSeconds: "30" + maxElapsedTimeReadySubServiceBeforeStartMinutes: "15" + periodReadySubServiceSeconds: "60" + maxLoadElapsedTimeMinutes: "120" + maxUnloadElapsedTimeMinutes: "15" + maxLoadRetryCount: "5" + maxUnloadRetryCount: "1" resources: cpu: 200m memory: 1Gi diff --git a/k8s/helm-charts/seldon-core-v2-setup/values.yaml.template b/k8s/helm-charts/seldon-core-v2-setup/values.yaml.template index 655c3cb4f0..c4cb9ab798 100644 --- a/k8s/helm-charts/seldon-core-v2-setup/values.yaml.template +++ b/k8s/helm-charts/seldon-core-v2-setup/values.yaml.template @@ -243,6 +243,13 @@ serverConfig: scalingStatsPeriodSeconds: "20" modelInferenceLagThreshold: "30" modelInactiveSecondsThreshold: "600" + maxElapsedTimeReadySubServiceAfterStartSeconds: "30" + maxElapsedTimeReadySubServiceBeforeStartMinutes: "15" + periodReadySubServiceSeconds: "60" + maxLoadElapsedTimeMinutes: "120" + maxUnloadElapsedTimeMinutes: "15" + maxLoadRetryCount: "5" + maxUnloadRetryCount: "1" resources: cpu: 200m memory: 1Gi diff --git a/k8s/kustomize/helm-components-sc/patch_mlserver.yaml b/k8s/kustomize/helm-components-sc/patch_mlserver.yaml index 5ffdb45340..a9e6f8c500 100644 --- a/k8s/kustomize/helm-components-sc/patch_mlserver.yaml +++ b/k8s/kustomize/helm-components-sc/patch_mlserver.yaml @@ -24,6 +24,20 @@ spec: value: '{{ .Values.serverConfig.agent.modelInactiveSecondsThreshold }}' - name: SELDON_SCALING_STATS_PERIOD_SECONDS value: '{{ .Values.serverConfig.agent.scalingStatsPeriodSeconds }}' + - name: SELDON_MAX_TIME_READY_SUB_SERVICE_AFTER_START_SECONDS + value: '{{ .Values.serverConfig.agent.maxElapsedTimeReadySubServiceAfterStartSeconds }}' + - name: SELDON_MAX_ELAPSED_TIME_READY_SUB_SERVICE_BEFORE_START_MINUTES + value: '{{ .Values.serverConfig.agent.maxElapsedTimeReadySubServiceBeforeStartMinutes }}' + - name: SELDON_PERIOD_READY_SUB_SERVICE_SECONDS + value: '{{ .Values.serverConfig.agent.periodReadySubServiceSeconds }}' + - name: SELDON_MAX_LOAD_ELAPSED_TIME_MINUTES + value: '{{ .Values.serverConfig.agent.maxLoadElapsedTimeMinutes }}' + - name: SELDON_MAX_UNLOAD_ELAPSED_TIME_MINUTES + value: '{{ .Values.serverConfig.agent.maxUnloadElapsedTimeMinutes }}' + - name: SELDON_MAX_LOAD_RETRY_COUNT + value: '{{ .Values.serverConfig.agent.maxLoadRetryCount }}' + - name: SELDON_MAX_UNLOAD_RETRY_COUNT + value: '{{ .Values.serverConfig.agent.maxUnloadRetryCount }}' - name: SELDON_OVERCOMMIT_PERCENTAGE value: '{{ .Values.serverConfig.agent.overcommitPercentage }}' - name: CONTROL_PLANE_SECURITY_PROTOCOL diff --git a/k8s/kustomize/helm-components-sc/patch_triton.yaml b/k8s/kustomize/helm-components-sc/patch_triton.yaml index ffd06955ef..4d1d668434 100644 --- a/k8s/kustomize/helm-components-sc/patch_triton.yaml +++ b/k8s/kustomize/helm-components-sc/patch_triton.yaml @@ -24,6 +24,20 @@ spec: value: '{{ .Values.serverConfig.agent.modelInactiveSecondsThreshold }}' - name: SELDON_SCALING_STATS_PERIOD_SECONDS value: '{{ .Values.serverConfig.agent.scalingStatsPeriodSeconds }}' + - name: SELDON_MAX_TIME_READY_SUB_SERVICE_AFTER_START_SECONDS + value: '{{ .Values.serverConfig.agent.maxElapsedTimeReadySubServiceAfterStartSeconds }}' + - name: SELDON_MAX_ELAPSED_TIME_READY_SUB_SERVICE_BEFORE_START_MINUTES + value: '{{ .Values.serverConfig.agent.maxElapsedTimeReadySubServiceBeforeStartMinutes }}' + - name: SELDON_PERIOD_READY_SUB_SERVICE_SECONDS + value: '{{ .Values.serverConfig.agent.periodReadySubServiceSeconds }}' + - name: SELDON_MAX_LOAD_ELAPSED_TIME_MINUTES + value: '{{ .Values.serverConfig.agent.maxLoadElapsedTimeMinutes }}' + - name: SELDON_MAX_UNLOAD_ELAPSED_TIME_MINUTES + value: '{{ .Values.serverConfig.agent.maxUnloadElapsedTimeMinutes }}' + - name: SELDON_MAX_LOAD_RETRY_COUNT + value: '{{ .Values.serverConfig.agent.maxLoadRetryCount }}' + - name: SELDON_MAX_UNLOAD_RETRY_COUNT + value: '{{ .Values.serverConfig.agent.maxUnloadRetryCount }}' - name: SELDON_OVERCOMMIT_PERCENTAGE value: '{{ .Values.serverConfig.agent.overcommitPercentage }}' - name: CONTROL_PLANE_SECURITY_PROTOCOL diff --git a/k8s/yaml/components.yaml b/k8s/yaml/components.yaml index 14d4755a81..867791454b 100644 --- a/k8s/yaml/components.yaml +++ b/k8s/yaml/components.yaml @@ -1157,6 +1157,20 @@ spec: value: '20' - name: SELDON_OVERCOMMIT_PERCENTAGE value: '10' + - name: SELDON_MAX_TIME_READY_SUB_SERVICE_AFTER_START_SECONDS + value: '30' + - name: SELDON_MAX_ELAPSED_TIME_READY_SUB_SERVICE_BEFORE_START_MINUTES + value: '15' + - name: SELDON_PERIOD_READY_SUB_SERVICE_SECONDS + value: '60' + - name: SELDON_MAX_LOAD_ELAPSED_TIME_MINUTES + value: '120' + - name: SELDON_MAX_UNLOAD_ELAPSED_TIME_MINUTES + value: '15' + - name: SELDON_MAX_LOAD_RETRY_COUNT + value: '5' + - name: SELDON_MAX_UNLOAD_RETRY_COUNT + value: '1' - name: CONTROL_PLANE_SECURITY_PROTOCOL value: 'PLAINTEXT' - name: CONTROL_PLANE_CLIENT_TLS_SECRET_NAME