feat(agent): default worker idleTTL to 30m so stale workers get reaped

nullplatform/agent/README.md

@@ -169,7 +169,7 @@ resource "example_resource" "this" {
 | <a name="input_service_template"></a> [service\_template](#input\_service\_template) | Path, inside the worker image, of the Service template the k8s scope renders. Empty (default) uses the template worker\_ingress selects; set it only to point at a custom template. | `string` | `""` | no |
 | <a name="input_tags_selectors"></a> [tags\_selectors](#input\_tags\_selectors) | Map of tags used to select and filter channels and agents | `map(string)` | n/a | yes |
 | <a name="input_use_account_slug"></a> [use\_account\_slug](#input\_use\_account\_slug) | Flag to determine whether to use the account slug in resource naming | `string` | `""` | no |
-| <a name="input_worker"></a> [worker](#input\_worker) | Extra worker-orchestration config, merged on top of the module's own computed<br/>worker block: backend ("kubernetes" by default), allowedRegistries<br/>(["public.ecr.aws/nullplatform/*"] by default, so the platform's own scope<br/>images keep working), and a patch for the worker container (2Gi memory<br/>limit, the deploy/DNS env vars below, and a serviceAccountName that always<br/>mirrors service\_account\_name). allowedRegistries and patches set here are<br/>concatenated with (not replacing) the module defaults — add your own<br/>registries or an extra patch rather than having to repeat the defaults;<br/>set backend here to override it outright. Anything else — security, idleTTL<br/>(reap idle workers), the legacy defaults/rules/pins — passes through as-is.<br/>See the nullplatform-agent chart values (>= 2.37.0) for the full shape.<br/>null = nothing extra.<br/><br/>Example:<br/>  worker = {<br/>    allowedRegistries = ["123456789012.dkr.ecr.us-east-1.amazonaws.com/your-org/*"]<br/>    patches           = [{ target = { package = "my-pkg" }, merge = { spec = { serviceAccountName = "np-agent-sa" } } }]<br/>    idleTTL           = "30m"<br/>  } | `any` | `null` | no |
+| <a name="input_worker"></a> [worker](#input\_worker) | Extra worker-orchestration config, merged on top of the module's own computed<br/>worker block: backend ("kubernetes" by default), allowedRegistries<br/>(["public.ecr.aws/nullplatform/*"] by default, so the platform's own scope<br/>images keep working), idleTTL ("30m" by default, so worker Deployments<br/>left behind by an old/removed package revision get reaped instead of<br/>accumulating forever), and a patch for the worker container (2Gi memory<br/>limit, the deploy/DNS env vars below, and a serviceAccountName that always<br/>mirrors service\_account\_name). allowedRegistries and patches set here are<br/>concatenated with (not replacing) the module defaults — add your own<br/>registries or an extra patch rather than having to repeat the defaults;<br/>set backend or idleTTL here to override them outright (e.g. idleTTL = ""<br/>to disable the reaper, matching this module's pre-idleTTL-default<br/>behavior). Anything else — security, the legacy defaults/rules/pins —<br/>passes through as-is. See the nullplatform-agent chart values (>= 2.37.0)<br/>for the full shape. null = nothing extra beyond the defaults above.<br/><br/>Example:<br/>  worker = {<br/>    allowedRegistries = ["123456789012.dkr.ecr.us-east-1.amazonaws.com/your-org/*"]<br/>    patches           = [{ target = { package = "my-pkg" }, merge = { spec = { serviceAccountName = "np-agent-sa" } } }]<br/>    idleTTL           = "1h"<br/>  } | `any` | `null` | no |
 | <a name="input_worker_ingress"></a> [worker\_ingress](#input\_worker\_ingress) | Ingress stack the containers worker deploys scopes with: "alb" keeps the k8s scope's own templates (AWS Load Balancer Controller Ingress), "istio" points it at the Gateway API templates baked in the scopes/containers image. service\_template, initial\_ingress\_path and blue\_green\_ingress\_path override the derived paths when set. | `string` | `"alb"` | no |
 | <a name="input_worker_k8s_packages"></a> [worker\_k8s\_packages](#input\_worker\_k8s\_packages) | Package slugs whose worker runs the k8s scope and must receive its env vars (DNS\_TYPE, K8S\_NAMESPACE, template paths, CLUSTER\_NAME, extra\_envs). Add every package that runs the k8s scope code, e.g. ["containers", "scheduled-task"]. | `list(string)` | <pre>[<br/>  "containers"<br/>]</pre> | no |
 | <a name="input_worker_memory_limit"></a> [worker\_memory\_limit](#input\_worker\_memory\_limit) | Memory limit for a worker-orchestrated package's pod (packages in var.worker\_orchestrated\_packages). The chart's own default is small enough to OOM mid-tofu-apply for packages that run real IaC tooling. | `string` | `"2Gi"` | no |

nullplatform/agent/locals.tf

@@ -172,6 +172,13 @@ locals {
     backend           = "kubernetes"
     allowedRegistries = ["public.ecr.aws/nullplatform/*"]
     patches           = concat(local.worker_common_patches, local.worker_k8s_env_patches)
+    # Reap worker-orchestrated pods (and their Deployments) after 30m with no
+    # activity. Previously unset (NP_WORKER_IDLE_TTL empty), which disables
+    # the reaper entirely — stale workers from old package revisions or
+    # removed packages accumulate forever instead of being cleaned up.
+    # Override per-install via var.worker.idleTTL (see its docs for the
+    # shape) if a longer/shorter window is needed.
+    idleTTL = "30m"
   }
 
   worker_final = merge(

nullplatform/agent/tests/agent_values.tftest.hcl

@@ -234,6 +234,32 @@ run "worker_defaults" {
     condition     = strcontains(helm_release.agent.values[0], "\"serviceAccountName\": \"nullplatform-agent\"")
     error_message = "the worker's serviceAccountName must default to service_account_name's default (nullplatform-agent)"
   }
+
+  assert {
+    condition     = strcontains(helm_release.agent.values[0], "\"idleTTL\": \"30m\"")
+    error_message = "idleTTL must default to 30m so worker Deployments from an old/removed package revision get reaped instead of accumulating forever"
+  }
+}
+
+# idleTTL is a plain override (like backend), not additive (like
+# allowedRegistries/patches) — var.worker's value must win outright,
+# including the empty string, which disables the reaper.
+run "worker_idle_ttl_override_wins_outright" {
+  command = plan
+
+  variables {
+    worker = {
+      idleTTL = ""
+    }
+  }
+
+  assert {
+    condition = (
+      strcontains(helm_release.agent.values[0], "\"idleTTL\": \"\"") &&
+      !strcontains(helm_release.agent.values[0], "\"idleTTL\": \"30m\"")
+    )
+    error_message = "var.worker.idleTTL must override the module default outright, including disabling it with an empty string"
+  }
 }
 
 # var.worker stays available as an extra/override layer on top of the computed

nullplatform/agent/variables.tf

@@ -140,21 +140,24 @@ variable "worker" {
     Extra worker-orchestration config, merged on top of the module's own computed
     worker block: backend ("kubernetes" by default), allowedRegistries
     (["public.ecr.aws/nullplatform/*"] by default, so the platform's own scope
-    images keep working), and a patch for the worker container (2Gi memory
+    images keep working), idleTTL ("30m" by default, so worker Deployments
+    left behind by an old/removed package revision get reaped instead of
+    accumulating forever), and a patch for the worker container (2Gi memory
     limit, the deploy/DNS env vars below, and a serviceAccountName that always
     mirrors service_account_name). allowedRegistries and patches set here are
     concatenated with (not replacing) the module defaults — add your own
     registries or an extra patch rather than having to repeat the defaults;
-    set backend here to override it outright. Anything else — security, idleTTL
-    (reap idle workers), the legacy defaults/rules/pins — passes through as-is.
-    See the nullplatform-agent chart values (>= 2.37.0) for the full shape.
-    null = nothing extra.
+    set backend or idleTTL here to override them outright (e.g. idleTTL = ""
+    to disable the reaper, matching this module's pre-idleTTL-default
+    behavior). Anything else — security, the legacy defaults/rules/pins —
+    passes through as-is. See the nullplatform-agent chart values (>= 2.37.0)
+    for the full shape. null = nothing extra beyond the defaults above.
 
     Example:
       worker = {
         allowedRegistries = ["123456789012.dkr.ecr.us-east-1.amazonaws.com/your-org/*"]
         patches           = [{ target = { package = "my-pkg" }, merge = { spec = { serviceAccountName = "np-agent-sa" } } }]
-        idleTTL           = "30m"
+        idleTTL           = "1h"
       }
   EOT
   type        = any