The open-source agent harness - the runtime layer that turns an LLM into a working agent.
parent changes: rename trueforge to trueforge-server (#752)
Raman Tehlan2ec47e419c57
diff --git a/charts/trueforge/README.md b/charts/trueforge/README.mdindex c9721851e..64c6e9e0d 100644--- a/charts/trueforge/README.md+++ b/charts/trueforge/README.md@@ -151,17 +151,18 @@ The chart's own value wins on conflict; `tolerations` append to## Resource tiers`resourceTier` (`small` / `medium` / `large`) selects sizing presets for the-server and the controller. When set, it **replaces** the `resources` tables-(replica counts stay as set: server `replicaCount`, controller always 1); an-unknown tier fails the render. Empty (the default) keeps the explicit-`resources` / `controller.resources`. A parent chart may set-`global.resourceTier` instead; the chart's own `resourceTier` wins.--| Preset | Server requests | Controller requests |-| --- | --- | --- |-| `small` | 50m / 128Mi | 50m / 128Mi |-| `medium` | 100m / 256Mi | 100m / 256Mi |-| `large` | 500m / 512Mi | 500m / 512Mi |+server and the controller, and sets the server replica count (an explicit+`server.replicaCount` overrides it; the controller is always 1 replica). When+set, it **replaces** the `resources` tables; an unknown tier fails the render.+Empty (the default) keeps the explicit `resources` / `controller.resources`+and 1 server replica. A parent chart may set `global.resourceTier` instead;+the chart's own `resourceTier` wins.++| Preset | Server replicas | Server requests | Controller requests |+| --- | --- | --- | --- |+| `small` | 1 | 50m / 128Mi | 50m / 128Mi |+| `medium` | 2 | 100m / 256Mi | 100m / 256Mi |+| `large` | 3 | 500m / 512Mi | 500m / 512Mi |## Postgres@@ -322,8 +323,8 @@ extraObjects:| Value | Default | Description || --------------------- | ----------------------------------- | ------------------------------------- |-| `resourceTier` | `""` | Optional `small` / `medium` / `large` sizing preset; empty uses `resources`. |-| `server.replicaCount` | `1` | Number of server replicas. |+| `resourceTier` | `""` | Optional `small` / `medium` / `large` preset for resources and server replicas; empty uses `resources`. |+| `server.replicaCount` | `""` | Number of server replicas. Empty derives from the resource tier (small=1, medium=2, large=3; 1 with no tier). || `server.deploymentAnnotations` | `{}` | Annotations on the server Deployment, such as an Argo CD sync wave. || `controller.deploymentAnnotations` | `{}` | Annotations on the controller Deployment, such as an Argo CD sync wave. || `image.repository` | `tfy.jfrog.io/tfy-images/trueforge` | Image repository. |@@ -337,7 +338,7 @@ extraObjects:| `service.port` | `8790` | Service port. || `server.port` | `8790` | Container port (`PORT`). || `autoscaling.enabled` | `false` | Enable a HorizontalPodAutoscaler. |-| `podDisruptionBudget.enabled` | `false` | Enable a PodDisruptionBudget (`minAvailable` defaults to `1`). |+| `podDisruptionBudget.enabled` | `true` | PodDisruptionBudget (`minAvailable` defaults to `1`); rendered only when the server runs more than one replica. || `podSecurityContext` | non-root UID/GID `10001` | Pod-level restricted security defaults. || `securityContext` | read-only root FS + drop all capabilities | Container-level restricted security defaults. || `resources` | 100m/256Mi requests, 200m/512Mi limits | Server CPU, memory, and ephemeral-storage. Replaced when a resourceTier is set. |@@ -348,10 +349,13 @@ extraObjects:The server uses a RollingUpdate strategy by default (`server.strategy`); thecontroller is fixed to a single replica with `Recreate` and exposes neither.-Also available (defaults inert): `priorityClassName`,-`topologySpreadConstraints`, `initContainers`, `extraContainers`,-`extraVolumes`, `extraVolumeMounts`, `service.annotations`, `service.labels`,-`startupProbe`.+Server pods spread across nodes by default (`maxSkew: 1`,+`whenUnsatisfiable: ScheduleAnyway`); set `topologySpreadConstraints` to+override (entries apply verbatim to both server and controller pods).++Also available (defaults inert): `priorityClassName`, `initContainers`,+`extraContainers`, `extraVolumes`, `extraVolumeMounts`, `service.annotations`,+`service.labels`, `startupProbe`.The server container mounts an `emptyDir` at `/tmp` by default so the image canrun with `readOnlyRootFilesystem: true`. When set, `resources.limits.ephemeral-storage`@@ -369,4 +373,4 @@ also sets the `/tmp` `emptyDir.sizeLimit`.- If enabling `mtls`, set `mtls.secretName` and ensure any reverse proxy dials HTTPS with a trusted client cert (see Caddy `internal_mtls`).- Tune container `resources` (especially CPU requests) before enabling HPA.- Default `tfy.jfrog.io` images and the Helm chart are anonymously pullable — set `imagePullSecrets` only if you override to a private registry.-- Enable `podDisruptionBudget` when running multiple replicas (defaults to `minAvailable: 1`; set exactly one of `minAvailable` or `maxUnavailable`).+- Run multiple replicas (`server.replicaCount` or a `resourceTier` of `medium`/`large`); the PodDisruptionBudget (`minAvailable: 1`) then applies automatically.diff --git a/charts/trueforge/templates/_helpers.tpl b/charts/trueforge/templates/_helpers.tplindex 643029cd5..9c733e1c6 100644--- a/charts/trueforge/templates/_helpers.tpl+++ b/charts/trueforge/templates/_helpers.tpl@@ -232,8 +232,24 @@ chart's global.resourceTier.{{- end -}}{{- end }}+{{/*+Server replica count. An explicit server.replicaCount wins; otherwise the+resource tier decides (small=1, medium=2, large=3), falling back to 1 when no+tier is set.+*/}}{{- define "trueforge.replicas" -}}+{{- $tier := include "trueforge.resourceTier" . | trim -}}+{{- if .Values.server.replicaCount -}}{{- .Values.server.replicaCount -}}+{{- else if eq $tier "small" -}}+1+{{- else if eq $tier "medium" -}}+2+{{- else if eq $tier "large" -}}+3+{{- else -}}+1+{{- end -}}{{- end }}{{- define "trueforge.defaultResources.small" -}}diff --git a/charts/trueforge/templates/deployment.yaml b/charts/trueforge/templates/deployment.yamlindex 91090a839..13597322c 100644--- a/charts/trueforge/templates/deployment.yaml+++ b/charts/trueforge/templates/deployment.yaml@@ -1,7 +1,7 @@apiVersion: apps/v1kind: Deploymentmetadata:- name: {{ include "trueforge.fullname" . }}+ name: {{ include "trueforge.fullname" . }}-serverlabels:{{- include "trueforge.labels" . | nindent 4 }}{{- with (mergeOverwrite (include "trueforge.annotations" . | fromYaml) (deepCopy .Values.server.deploymentAnnotations)) }}@@ -57,7 +57,7 @@ spec:{{- end }}{{- end }}containers:- - name: server+ - name: trueforge-serverimage: {{ include "trueforge.image" . | quote }}imagePullPolicy: {{ .Values.image.pullPolicy }}{{- with .Values.securityContext }}@@ -120,7 +120,14 @@ spec:tolerations:{{- . | nindent 8 }}{{- end }}- {{- with .Values.topologySpreadConstraints }}topologySpreadConstraints:+ {{- with .Values.topologySpreadConstraints }}{{- toYaml . | nindent 8 }}- {{- end }}+ {{- else }}+ - maxSkew: 1+ topologyKey: kubernetes.io/hostname+ whenUnsatisfiable: ScheduleAnyway+ labelSelector:+ matchLabels:+ {{- include "trueforge.selectorLabels" . | nindent 14 }}+ {{- end }}diff --git a/charts/trueforge/templates/hpa.yaml b/charts/trueforge/templates/hpa.yamlindex a2b0cfa9f..43c4590ea 100644--- a/charts/trueforge/templates/hpa.yaml+++ b/charts/trueforge/templates/hpa.yaml@@ -13,7 +13,7 @@ spec:scaleTargetRef:apiVersion: apps/v1kind: Deployment- name: {{ include "trueforge.fullname" . }}+ name: {{ include "trueforge.fullname" . }}-serverminReplicas: {{ .Values.autoscaling.minReplicas }}maxReplicas: {{ .Values.autoscaling.maxReplicas }}metrics:diff --git a/charts/trueforge/templates/pdb.yaml b/charts/trueforge/templates/pdb.yamlindex b3095b6da..74d770a85 100644--- a/charts/trueforge/templates/pdb.yaml+++ b/charts/trueforge/templates/pdb.yaml@@ -1,4 +1,7 @@-{{- if .Values.podDisruptionBudget.enabled }}+{{- /* Guard on the guaranteed floor: minReplicas under autoscaling, else the+fixed count. A PDB over a single replica would block node drains. */}}+{{- $replicas := ternary .Values.autoscaling.minReplicas (include "trueforge.replicas" . | int) .Values.autoscaling.enabled | int }}+{{- if and .Values.podDisruptionBudget.enabled (gt $replicas 1) }}{{- /* toString so integer 0 is distinct from unset ("" / nil); if/with treat 0 as falsy. */}}{{- $min := .Values.podDisruptionBudget.minAvailable | toString }}{{- $max := .Values.podDisruptionBudget.maxUnavailable | toString }}diff --git a/charts/trueforge/values.yaml b/charts/trueforge/values.yamlindex bcc7b1cb7..0462d5f84 100644--- a/charts/trueforge/values.yaml+++ b/charts/trueforge/values.yaml@@ -1,4 +1,41 @@# Default values for the trueforge chart.+# Bitnami relocated its versioned container images to docker.io/bitnamilegacy and+# left only rolling `latest` tags under docker.io/bitnami, so the subchart image+# defaults no longer pull. We source pinned legacy images from the TrueFoundry+# JFrog mirror instead. Overriding the registry trips Bitnami's image guardrail,+# so allowInsecureImages must be set for the subcharts to render.+global:+ resourceTier: ""+ security:+ allowInsecureImages: true+ # Custom CA trust, set here or inherited from a parent chart. Give an inline+ # `certificate` or an `existingConfigMap.name` (key `ca-certificates.crt`);+ # mounted over /etc/ssl/certs (overrideCAList) or merged with the system+ # bundle by an initContainer. Sets NODE_EXTRA_CA_CERTS either way.+ customCA:+ enabled: false+ certificate: ""+ existingConfigMap:+ name: ""+ overrideCAList: false+ image:+ registry: tfy.jfrog.io+ repository: tfy-mirror/alpine+ tag: "3.21"+ securityContext:+ runAsNonRoot: true+ runAsUser: 1001+ runAsGroup: 1001+ readOnlyRootFilesystem: true+ allowPrivilegeEscalation: false+ capabilities:+ drop:+ - ALL+ # Extra env for the initContainer only, e.g. SSL_CERT_FILE, REQUESTS_CA_BUNDLE.+ env: {}+ emptyDir:+ sslCerts:+ sizeLimit: 10Miimage:repository: tfy.jfrog.io/tfy-images/trueforge@@ -48,7 +85,9 @@ service:# packages/trueforge/src/config.ts.server:# Number of server replicas (ignored when autoscaling.enabled is true).- replicaCount: 1+ # null = derived from the resource tier (small=1, medium=2, large=3),+ # or 1 when no tier is set. Set a number to override the tier.+ replicaCount: null# Annotations on the server Deployment object (for example, Argo CD sync waves).deploymentAnnotations: {}# Rolling update strategy for the server Deployment.@@ -132,42 +171,6 @@ configs:# Optional comma-separated exact emails and/or * globs (e.g. "*@company.com").# Empty = unrestricted. Matched against the ID token `email` claim.allowedEmails: ""-# Bitnami relocated its versioned container images to docker.io/bitnamilegacy and-# left only rolling `latest` tags under docker.io/bitnami, so the subchart image-# defaults no longer pull. We source pinned legacy images from the TrueFoundry-# JFrog mirror instead. Overriding the registry trips Bitnami's image guardrail,-# so allowInsecureImages must be set for the subcharts to render.-global:- security:- allowInsecureImages: true- # Custom CA trust, set here or inherited from a parent chart. Give an inline- # `certificate` or an `existingConfigMap.name` (key `ca-certificates.crt`);- # mounted over /etc/ssl/certs (overrideCAList) or merged with the system- # bundle by an initContainer. Sets NODE_EXTRA_CA_CERTS either way.- customCA:- enabled: false- certificate: ""- existingConfigMap:- name: ""- overrideCAList: false- image:- registry: tfy.jfrog.io- repository: tfy-mirror/alpine- tag: "3.21"- securityContext:- runAsNonRoot: true- runAsUser: 1001- runAsGroup: 1001- readOnlyRootFilesystem: true- allowPrivilegeEscalation: false- capabilities:- drop:- - ALL- # Extra env for the initContainer only, e.g. SSL_CERT_FILE, REQUESTS_CA_BUNDLE.- env: {}- emptyDir:- sslCerts:- sizeLimit: 10Mi# Resource tier: small | medium | large presets for server and controller.# Empty = use `resources` / `controller.resources`. Wins over a parent chart's# global.resourceTier.@@ -261,6 +264,7 @@ externalRedis:masterName: ""# string or valueFrom -> REDIS_SENTINEL_PASSWORD.password: ""+# Extra raw manifests deployed alongside the server, e.g. an Istio# VirtualService, a Gateway, an Ingress, or a NetworkPolicy. Each entry is a# full Kubernetes object rendered through `tpl`, so Helm templating works and@@ -335,7 +339,9 @@ autoscaling:maxReplicas: 5targetCPUUtilizationPercentage: 80podDisruptionBudget:- enabled: false+ # Rendered only when the server runs more than one replica, so single-replica+ # installs never block node drains.+ enabled: true# Exactly one of minAvailable or maxUnavailable is required when enabled.# Default keeps at least one pod available during voluntary disruptions.minAvailable: 1@@ -343,4 +349,6 @@ podDisruptionBudget:nodeSelector: {}tolerations: []affinity: {}+# Empty = server pods spread across nodes (maxSkew 1, ScheduleAnyway). Set to+# override; entries apply verbatim to both server and controller pods.topologySpreadConstraints: []