Make drain delay configurable via drainDelaySeconds field

crd/RestateDeployment.pkl

@@ -64,6 +64,10 @@ class Restate {
 
   /// Force the use of HTTP/1.1 when registering with Restate
   useHttp11: Boolean?
+
+  /// Seconds to wait before removing old versions after they are drained. Defaults to 300 (5
+  /// minutes).
+  drainDelaySeconds: Int?
 }
 
 /// The location of the Restate Admin API to register this deployment against

crd/restatedeployments.yaml

@@ -138,6 +138,14 @@ spec:
               restate:
                 description: Restate specific configuration
                 properties:
+                  drainDelaySeconds:
+                    description: |-
+                      Seconds to wait before removing old versions after they are drained.
+                      Defaults to 300 (5 minutes).
+                    format: int64
+                    minimum: 0.0
+                    nullable: true
+                    type: integer
                   register:
                     description: The location of the Restate Admin API to register this deployment against
                     oneOf:

release-notes/unreleased/96-configurable-drain-delay.md

@@ -0,0 +1,30 @@
+# Release Notes for PR #96: Configurable drain delay
+
+## New Feature
+
+### What Changed
+Added a new optional `drainDelaySeconds` field to the RestateDeployment CRD's
+`spec.restate` section. This controls how long the operator waits after a
+deployment is drained before removing the old version. Previously hardcoded to
+5 minutes (300 seconds).
+
+### Why This Matters
+The default 5-minute safety buffer isn't always appropriate. Some environments
+may want a longer window before old versions are cleaned up, while others may
+want a shorter one.
+
+### Impact on Users
+- **Existing deployments**: No impact. The default remains 300 seconds (5 minutes).
+- **New deployments**: Can now configure the drain delay per RestateDeployment.
+
+### Migration Guidance
+No migration needed. To configure a custom drain delay:
+
+```yaml
+spec:
+  restate:
+    drainDelaySeconds: 600  # 10 minutes
+```
+
+### Related Issues
+- PR #96: Make drain delay configurable via drainDelaySeconds field

src/controllers/restatedeployment/reconcilers/knative.rs

@@ -929,15 +929,16 @@ pub async fn cleanup_old_configurations(
             }
             (None, _, true) => {
                 // endpoint exists and there's no valid remove_version_at annotation, create one
+                let drain_delay_seconds = rsd.spec.restate.drain_delay_seconds();
                 info!(
                     configuration = %config_name,
                     namespace = %namespace,
-                    drain_delay = "5 minutes",
+                    drain_delay_seconds,
                     "Scheduling removal of old Configuration (after drain delay)"
                 );
 
                 let remove_at = chrono::Utc::now()
-                    .checked_add_signed(chrono::TimeDelta::minutes(5)) // Same as ReplicaSet cleanup
+                    .checked_add_signed(chrono::TimeDelta::seconds(drain_delay_seconds))
                     .expect("remove_version_at in bounds");
 
                 let config_api: Api<Configuration> = Api::namespaced(ctx.client.clone(), namespace);

src/controllers/restatedeployment/reconcilers/replicaset.rs

@@ -359,13 +359,15 @@ pub async fn cleanup_old_replicasets(
             }
             (None, _, true) => {
                 // endpoint exists and there's no valid remove_version_at annotation, create one
-                debug!(
-                    "Scheduling removal (after drain delay) of old ReplicaSet {} in namespace {namespace}",
-                    rs_name,
+                let drain_delay_seconds = rsd.spec.restate.drain_delay_seconds();
+                info!(
+                    replicaset = %rs_name,
+                    namespace = %namespace,
+                    drain_delay_seconds,
+                    "Scheduling removal of old ReplicaSet (after drain delay)"
                 );
-
                 let remove_at = chrono::Utc::now()
-                    .checked_add_signed(chrono::TimeDelta::minutes(5)) // todo configurable?
+                    .checked_add_signed(chrono::TimeDelta::seconds(drain_delay_seconds))
                     .expect("remove_version_at in bounds");
 
                 let params = PatchParams::apply("restate-operator/remove-version-at").force();

src/resources/restatedeployments.rs

@@ -247,6 +247,18 @@ pub struct RestateSpec {
     /// Force the use of HTTP/1.1 when registering with Restate
     #[serde(skip_serializing_if = "Option::is_none")]
     pub use_http11: Option<bool>,
+
+    /// Seconds to wait before removing old versions after they are drained.
+    /// Defaults to 300 (5 minutes).
+    #[serde(skip_serializing_if = "Option::is_none")]
+    #[schemars(range(min = 0))]
+    pub drain_delay_seconds: Option<i64>,
+}
+
+impl RestateSpec {
+    pub fn drain_delay_seconds(&self) -> i64 {
+        self.drain_delay_seconds.unwrap_or(300).max(0)
+    }
 }
 
 /// The location of the Restate Admin API to register this deployment against