feat(persistence+k8s): root_workflow_id schema, PVC provisioning, name fallback, host name-or-UUID

wfe-kubernetes/src/config.rs

@@ -20,6 +20,18 @@ pub struct ClusterConfig {
     /// Node selector labels for Job pods.
     #[serde(default)]
     pub node_selector: HashMap<String, String>,
+    /// Default size (e.g. "10Gi") used when a workflow declares a
+    /// `shared_volume` without specifying its own `size`. The K8s executor
+    /// creates one PVC per top-level workflow instance and mounts it on
+    /// every step container so sub-workflows can share a cloned checkout,
+    /// sccache directory, etc. Cluster operators tune this based on the
+    /// typical working-set size of their pipelines.
+    #[serde(default = "default_shared_volume_size")]
+    pub default_shared_volume_size: String,
+    /// Optional StorageClass to use when provisioning shared-volume PVCs.
+    /// Falls back to the cluster's default StorageClass when unset.
+    #[serde(default)]
+    pub shared_volume_storage_class: Option<String>,
 }
 
 impl Default for ClusterConfig {
@@ -30,6 +42,8 @@ impl Default for ClusterConfig {
             service_account: None,
             image_pull_secrets: Vec::new(),
             node_selector: HashMap::new(),
+            default_shared_volume_size: default_shared_volume_size(),
+            shared_volume_storage_class: None,
         }
     }
 }
@@ -38,6 +52,10 @@ fn default_namespace_prefix() -> String {
     "wfe-".to_string()
 }
 
+fn default_shared_volume_size() -> String {
+    "10Gi".to_string()
+}
+
 /// Per-step configuration for a Kubernetes Job execution.
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct KubernetesStepConfig {
@@ -46,9 +64,19 @@ pub struct KubernetesStepConfig {
     /// Override entrypoint.
     #[serde(default)]
     pub command: Option<Vec<String>>,
-    /// Shorthand: runs via `/bin/sh -c "..."`. Mutually exclusive with `command`.
+    /// Shorthand: runs the given script via the configured `shell`
+    /// (default `/bin/sh`). Mutually exclusive with `command`. For scripts
+    /// that rely on bashisms like `set -o pipefail`, process substitution,
+    /// or arrays, set `shell: /bin/bash` explicitly — the default /bin/sh
+    /// keeps alpine/busybox containers working out of the box.
     #[serde(default)]
     pub run: Option<String>,
+    /// Shell used to execute a `run:` script. Defaults to `/bin/sh` so
+    /// minimal containers (alpine, distroless) work unchanged. Override
+    /// to `/bin/bash` or any other interpreter when the script needs
+    /// features dash doesn't support.
+    #[serde(default)]
+    pub shell: Option<String>,
     /// Environment variables injected into the container.
     #[serde(default)]
     pub env: HashMap<String, String>,
@@ -95,6 +123,8 @@ mod tests {
             service_account: Some("wfe-runner".into()),
             image_pull_secrets: vec!["ghcr-secret".into()],
             node_selector: [("tier".into(), "compute".into())].into(),
+            default_shared_volume_size: "20Gi".into(),
+            shared_volume_storage_class: Some("fast-ssd".into()),
         };
         let json = serde_json::to_string(&config).unwrap();
         let parsed: ClusterConfig = serde_json::from_str(&json).unwrap();
@@ -119,6 +149,7 @@ mod tests {
             image: "node:20-alpine".into(),
             command: None,
             run: Some("npm test".into()),
+            shell: None,
             env: [("NODE_ENV".into(), "test".into())].into(),
             working_dir: Some("/app".into()),
             memory: Some("512Mi".into()),

wfe-kubernetes/src/lib.rs

@@ -5,6 +5,7 @@ pub mod logs;
 pub mod manifests;
 pub mod namespace;
 pub mod output;
+pub mod pvc;
 pub mod service_manifests;
 pub mod service_provider;
 pub mod step;

wfe-kubernetes/src/manifests.rs

@@ -2,7 +2,8 @@ use std::collections::{BTreeMap, HashMap};
 
 use k8s_openapi::api::batch::v1::{Job, JobSpec};
 use k8s_openapi::api::core::v1::{
-    Container, EnvVar, LocalObjectReference, PodSpec, PodTemplateSpec, ResourceRequirements,
+    Container, EnvVar, LocalObjectReference, PersistentVolumeClaimVolumeSource, PodSpec,
+    PodTemplateSpec, ResourceRequirements, Volume, VolumeMount,
 };
 use k8s_openapi::apimachinery::pkg::api::resource::Quantity;
 use kube::api::ObjectMeta;
@@ -11,6 +12,18 @@ use crate::config::{ClusterConfig, KubernetesStepConfig};
 
 const LABEL_STEP_NAME: &str = "wfe.sunbeam.pt/step-name";
 const LABEL_MANAGED_BY: &str = "wfe.sunbeam.pt/managed-by";
+/// Name of the pod volume referencing the shared PVC. Stable so VolumeMount
+/// name and Volume name stay in sync inside a single pod spec.
+const SHARED_VOLUME_NAME: &str = "wfe-workspace";
+
+/// Request that the generated Job mount a pre-existing PVC into every
+/// step container at `mount_path`. Passed in by the step executor, which
+/// is responsible for creating the PVC before calling `build_job`.
+#[derive(Debug, Clone)]
+pub struct SharedVolumeMount {
+    pub claim_name: String,
+    pub mount_path: String,
+}
 
 /// Build a Kubernetes Job manifest from step configuration.
 pub fn build_job(
@@ -19,6 +32,7 @@ pub fn build_job(
     namespace: &str,
     env_overrides: &HashMap<String, String>,
     cluster: &ClusterConfig,
+    shared_volume: Option<&SharedVolumeMount>,
 ) -> Job {
     let job_name = sanitize_name(step_name);
 
@@ -41,6 +55,14 @@ pub fn build_job(
     labels.insert(LABEL_STEP_NAME.into(), step_name.to_string());
     labels.insert(LABEL_MANAGED_BY.into(), "wfe-kubernetes".into());
 
+    let volume_mounts = shared_volume.map(|sv| {
+        vec![VolumeMount {
+            name: SHARED_VOLUME_NAME.into(),
+            mount_path: sv.mount_path.clone(),
+            ..Default::default()
+        }]
+    });
+
     let container = Container {
         name: "step".into(),
         image: Some(config.image.clone()),
@@ -50,6 +72,7 @@ pub fn build_job(
         working_dir: config.working_dir.clone(),
         resources: Some(resources),
         image_pull_policy: Some(pull_policy),
+        volume_mounts,
         ..Default::default()
     };
 
@@ -77,6 +100,17 @@ pub fn build_job(
         )
     };
 
+    let volumes = shared_volume.map(|sv| {
+        vec![Volume {
+            name: SHARED_VOLUME_NAME.into(),
+            persistent_volume_claim: Some(PersistentVolumeClaimVolumeSource {
+                claim_name: sv.claim_name.clone(),
+                read_only: Some(false),
+            }),
+            ..Default::default()
+        }]
+    });
+
     Job {
         metadata: ObjectMeta {
             name: Some(job_name),
@@ -98,6 +132,7 @@ pub fn build_job(
                     service_account_name: cluster.service_account.clone(),
                     image_pull_secrets,
                     node_selector,
+                    volumes,
                     ..Default::default()
                 }),
             },
@@ -112,13 +147,15 @@ fn resolve_command(config: &KubernetesStepConfig) -> (Option<Vec<String>>, Optio
     if let Some(ref cmd) = config.command {
         (Some(cmd.clone()), None)
     } else if let Some(ref run) = config.run {
-        // Use bash so that scripts can rely on `set -o pipefail`, process
-        // substitution, arrays, and other bashisms that dash (/bin/sh on
-        // debian-family images) does not support.
-        (
-            Some(vec!["/bin/bash".into(), "-c".into()]),
-            Some(vec![run.clone()]),
-        )
+        // Default to /bin/sh so minimal container images (alpine/busybox,
+        // distroless) work unchanged. Scripts that need bash features
+        // (pipefail, arrays, process substitution) should set
+        // `shell: /bin/bash` explicitly on the step config.
+        let shell = config
+            .shell
+            .clone()
+            .unwrap_or_else(|| "/bin/sh".to_string());
+        (Some(vec![shell, "-c".into()]), Some(vec![run.clone()]))
     } else {
         (None, None)
     }
@@ -208,6 +245,7 @@ mod tests {
             image: "alpine:3.18".into(),
             command: None,
             run: None,
+            shell: None,
             env: HashMap::new(),
             working_dir: None,
             memory: None,
@@ -222,6 +260,7 @@ mod tests {
             "wfe-abc",
             &HashMap::new(),
             &default_cluster(),
+            None,
         );
 
         assert_eq!(job.metadata.name, Some("test-step".into()));
@@ -248,6 +287,7 @@ mod tests {
             image: "node:20".into(),
             command: None,
             run: Some("npm test".into()),
+            shell: None,
             env: HashMap::new(),
             working_dir: Some("/app".into()),
             memory: None,
@@ -256,13 +296,17 @@ mod tests {
             pull_policy: None,
             namespace: None,
         };
-        let job = build_job(&config, "test", "ns", &HashMap::new(), &default_cluster());
+        let job = build_job(
+            &config,
+            "test",
+            "ns",
+            &HashMap::new(),
+            &default_cluster(),
+            None,
+        );
         let container = &job.spec.unwrap().template.spec.unwrap().containers[0];
 
-        assert_eq!(
-            container.command,
-            Some(vec!["/bin/bash".into(), "-c".into()])
-        );
+        assert_eq!(container.command, Some(vec!["/bin/sh".into(), "-c".into()]));
         assert_eq!(container.args, Some(vec!["npm test".into()]));
         assert_eq!(container.working_dir, Some("/app".into()));
     }
@@ -273,6 +317,7 @@ mod tests {
             image: "gcc:latest".into(),
             command: Some(vec!["make".into(), "build".into()]),
             run: None,
+            shell: None,
             env: HashMap::new(),
             working_dir: None,
             memory: None,
@@ -281,7 +326,14 @@ mod tests {
             pull_policy: None,
             namespace: None,
         };
-        let job = build_job(&config, "build", "ns", &HashMap::new(), &default_cluster());
+        let job = build_job(
+            &config,
+            "build",
+            "ns",
+            &HashMap::new(),
+            &default_cluster(),
+            None,
+        );
         let container = &job.spec.unwrap().template.spec.unwrap().containers[0];
 
         assert_eq!(container.command, Some(vec!["make".into(), "build".into()]));
@@ -294,6 +346,7 @@ mod tests {
             image: "alpine".into(),
             command: None,
             run: None,
+            shell: None,
             env: [("APP_ENV".into(), "production".into())].into(),
             working_dir: None,
             memory: None,
@@ -308,7 +361,7 @@ mod tests {
         ]
         .into();
 
-        let job = build_job(&config, "step", "ns", &overrides, &default_cluster());
+        let job = build_job(&config, "step", "ns", &overrides, &default_cluster(), None);
         let container = &job.spec.unwrap().template.spec.unwrap().containers[0];
         let env = container.env.as_ref().unwrap();
 
@@ -327,6 +380,7 @@ mod tests {
             image: "alpine".into(),
             command: None,
             run: None,
+            shell: None,
             env: HashMap::new(),
             working_dir: None,
             memory: Some("512Mi".into()),
@@ -335,7 +389,14 @@ mod tests {
             pull_policy: None,
             namespace: None,
         };
-        let job = build_job(&config, "step", "ns", &HashMap::new(), &default_cluster());
+        let job = build_job(
+            &config,
+            "step",
+            "ns",
+            &HashMap::new(),
+            &default_cluster(),
+            None,
+        );
         let container = &job.spec.unwrap().template.spec.unwrap().containers[0];
         let resources = container.resources.as_ref().unwrap();
 
@@ -359,6 +420,7 @@ mod tests {
             image: "alpine".into(),
             command: None,
             run: None,
+            shell: None,
             env: HashMap::new(),
             working_dir: None,
             memory: None,
@@ -367,7 +429,7 @@ mod tests {
             pull_policy: Some("Always".into()),
             namespace: None,
         };
-        let job = build_job(&config, "step", "ns", &HashMap::new(), &cluster);
+        let job = build_job(&config, "step", "ns", &HashMap::new(), &cluster, None);
         let pod_spec = job.spec.unwrap().template.spec.unwrap();
 
         assert_eq!(pod_spec.service_account_name, Some("wfe-runner".into()));
@@ -387,6 +449,7 @@ mod tests {
             image: "alpine".into(),
             command: None,
             run: None,
+            shell: None,
             env: HashMap::new(),
             working_dir: None,
             memory: None,
@@ -401,6 +464,7 @@ mod tests {
             "ns",
             &HashMap::new(),
             &default_cluster(),
+            None,
         );
         let labels = job.metadata.labels.as_ref().unwrap();
         assert_eq!(labels.get(LABEL_STEP_NAME), Some(&"my-step".to_string()));

wfe-kubernetes/src/pvc.rs

@@ -0,0 +1,96 @@
+//! PersistentVolumeClaim provisioning for cross-step shared workspaces.
+//!
+//! When a workflow definition declares a `shared_volume`, the K8s executor
+//! materializes it as a single PVC in the workflow's namespace. Every step
+//! container mounts that PVC at the declared `mount_path`, so sub-workflows
+//! of a top-level run see the same filesystem — a clone in `checkout` is
+//! still visible to `cargo fmt --check` in `lint`.
+//!
+//! The PVC name is derived from the namespace (one PVC per namespace) so
+//! multiple steps racing to create it are idempotent: the first `ensure_pvc`
+//! wins, and every subsequent call sees the existing claim.
+
+use std::collections::BTreeMap;
+
+use k8s_openapi::api::core::v1::{
+    PersistentVolumeClaim, PersistentVolumeClaimSpec, VolumeResourceRequirements,
+};
+use k8s_openapi::apimachinery::pkg::api::resource::Quantity;
+use kube::api::{ObjectMeta, PostParams};
+use kube::{Api, Client};
+use wfe_core::WfeError;
+
+const LABEL_MANAGED_BY: &str = "wfe.sunbeam.pt/managed-by";
+const MANAGED_BY_VALUE: &str = "wfe-kubernetes";
+
+/// Canonical name for the shared volume PVC within a given namespace. Using
+/// a stable name (rather than e.g. a UUID) means every step in the same
+/// namespace references the same claim without needing to pass the name
+/// through workflow metadata.
+pub fn shared_volume_pvc_name() -> &'static str {
+    "wfe-workspace"
+}
+
+/// Create the shared-volume PVC in `namespace` if it does not already exist.
+/// `size` is a K8s resource quantity string (e.g. `"10Gi"`). `storage_class`
+/// is optional — when `None` the PVC uses the cluster's default StorageClass.
+pub async fn ensure_shared_volume_pvc(
+    client: &Client,
+    namespace: &str,
+    size: &str,
+    storage_class: Option<&str>,
+) -> Result<(), WfeError> {
+    let api: Api<PersistentVolumeClaim> = Api::namespaced(client.clone(), namespace);
+    let name = shared_volume_pvc_name();
+
+    // Idempotent: if it already exists we're done. This races cleanly with
+    // concurrent steps because `get` + `create(AlreadyExists)` is tolerated.
+    if api.get(name).await.is_ok() {
+        return Ok(());
+    }
+
+    let mut labels = BTreeMap::new();
+    labels.insert(LABEL_MANAGED_BY.into(), MANAGED_BY_VALUE.into());
+
+    let pvc = PersistentVolumeClaim {
+        metadata: ObjectMeta {
+            name: Some(name.to_string()),
+            namespace: Some(namespace.to_string()),
+            labels: Some(labels),
+            ..Default::default()
+        },
+        spec: Some(PersistentVolumeClaimSpec {
+            access_modes: Some(vec!["ReadWriteOnce".to_string()]),
+            resources: Some(VolumeResourceRequirements {
+                requests: Some(
+                    [("storage".to_string(), Quantity(size.to_string()))]
+                        .into_iter()
+                        .collect(),
+                ),
+                ..Default::default()
+            }),
+            storage_class_name: storage_class.map(|s| s.to_string()),
+            ..Default::default()
+        }),
+        ..Default::default()
+    };
+
+    match api.create(&PostParams::default(), &pvc).await {
+        Ok(_) => Ok(()),
+        // Another step created it between our get and create — also fine.
+        Err(kube::Error::Api(err)) if err.code == 409 => Ok(()),
+        Err(e) => Err(WfeError::StepExecution(format!(
+            "failed to create shared-volume PVC '{name}' in '{namespace}': {e}"
+        ))),
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn shared_volume_pvc_name_is_stable() {
+        assert_eq!(shared_volume_pvc_name(), "wfe-workspace");
+    }
+}

wfe-kubernetes/src/step.rs

@@ -13,9 +13,10 @@ use wfe_core::traits::step::{StepBody, StepExecutionContext};
 use crate::cleanup::delete_job;
 use crate::config::{ClusterConfig, KubernetesStepConfig};
 use crate::logs::{stream_logs, wait_for_pod_running};
-use crate::manifests::build_job;
+use crate::manifests::{SharedVolumeMount, build_job};
 use crate::namespace::{ensure_namespace, namespace_name};
 use crate::output::{build_output_data, parse_outputs};
+use crate::pvc::{ensure_shared_volume_pvc, shared_volume_pvc_name};
 
 /// A workflow step that runs as a Kubernetes Job.
 pub struct KubernetesStep {
@@ -65,6 +66,15 @@ impl StepBody for KubernetesStep {
             .unwrap_or("unknown")
             .to_string();
 
+        // Isolation domain is keyed on the *root* workflow when set so
+        // sub-workflows started via `type: workflow` share their parent's
+        // namespace + PVC. Top-level (user-started) workflows fall back
+        // to their own id.
+        let isolation_id = context
+            .workflow
+            .root_workflow_id
+            .as_deref()
+            .unwrap_or(&context.workflow.id);
         let workflow_id = &context.workflow.id;
         let definition_id = &context.workflow.workflow_definition_id;
 
@@ -72,21 +82,59 @@ impl StepBody for KubernetesStep {
         let _ = self.get_client().await?;
         let client = self.client.as_ref().unwrap().clone();
 
-        // 1. Determine namespace.
+        // 1. Determine namespace. Honors explicit step-level override,
+        // otherwise derives from the isolation domain.
         let ns = self
             .config
             .namespace
             .clone()
-            .unwrap_or_else(|| namespace_name(&self.cluster.namespace_prefix, workflow_id));
+            .unwrap_or_else(|| namespace_name(&self.cluster.namespace_prefix, isolation_id));
 
         // 2. Ensure namespace exists.
         ensure_namespace(&client, &ns, workflow_id).await?;
 
+        // 2b. If the definition declares a shared volume, ensure the PVC
+        // exists in the namespace and compute the mount spec we'll inject
+        // into the Job. The PVC is created once per namespace (one per
+        // top-level workflow run) and reused by every step and
+        // sub-workflow. Backends with no definition in the step context
+        // (test fixtures) just skip the shared volume.
+        let shared_mount = if let Some(def) = context.definition {
+            if let Some(sv) = &def.shared_volume {
+                let size = sv
+                    .size
+                    .as_deref()
+                    .unwrap_or(&self.cluster.default_shared_volume_size);
+                ensure_shared_volume_pvc(
+                    &client,
+                    &ns,
+                    size,
+                    self.cluster.shared_volume_storage_class.as_deref(),
+                )
+                .await?;
+                Some(SharedVolumeMount {
+                    claim_name: shared_volume_pvc_name().to_string(),
+                    mount_path: sv.mount_path.clone(),
+                })
+            } else {
+                None
+            }
+        } else {
+            None
+        };
+
         // 3. Merge env vars: workflow.data (uppercased) + config.env.
         let env_overrides = extract_workflow_env(&context.workflow.data);
 
         // 4. Build Job manifest.
-        let job_manifest = build_job(&self.config, &step_name, &ns, &env_overrides, &self.cluster);
+        let job_manifest = build_job(
+            &self.config,
+            &step_name,
+            &ns,
+            &env_overrides,
+            &self.cluster,
+            shared_mount.as_ref(),
+        );
         let job_name = job_manifest
             .metadata
             .name

wfe-postgres/src/lib.rs

@@ -227,17 +227,27 @@ impl WorkflowRepository for PostgresPersistenceProvider {
         } else {
             instance.id.clone()
         };
+        // Fall back to the UUID when the caller didn't assign a human name.
+        // In production `WorkflowHost::start_workflow` always fills this in
+        // via `next_definition_sequence`, but test fixtures and any external
+        // caller that forgets shouldn't trip the UNIQUE constraint.
+        let name = if instance.name.is_empty() {
+            id.clone()
+        } else {
+            instance.name.clone()
+        };
 
         let mut tx = self.pool.begin().await.map_err(Self::map_sqlx_err)?;
 
         sqlx::query(
             r#"INSERT INTO wfc.workflows
-               (id, name, definition_id, version, description, reference, status, data,
-                next_execution, create_time, complete_time)
-               VALUES ($1,$2,$3,$4,$5,$6,$7,$8,$9,$10,$11)"#,
+               (id, name, root_workflow_id, definition_id, version, description, reference,
+                status, data, next_execution, create_time, complete_time)
+               VALUES ($1,$2,$3,$4,$5,$6,$7,$8,$9,$10,$11,$12)"#,
         )
         .bind(&id)
-        .bind(&instance.name)
+        .bind(&name)
+        .bind(&instance.root_workflow_id)
         .bind(&instance.workflow_definition_id)
         .bind(instance.version as i32)
         .bind(&instance.description)
@@ -264,12 +274,14 @@ impl WorkflowRepository for PostgresPersistenceProvider {
 
         sqlx::query(
             r#"UPDATE wfc.workflows SET
-               name=$2, definition_id=$3, version=$4, description=$5, reference=$6,
-               status=$7, data=$8, next_execution=$9, create_time=$10, complete_time=$11
+               name=$2, root_workflow_id=$3, definition_id=$4, version=$5,
+               description=$6, reference=$7, status=$8, data=$9, next_execution=$10,
+               create_time=$11, complete_time=$12
                WHERE id=$1"#,
         )
         .bind(&instance.id)
         .bind(&instance.name)
+        .bind(&instance.root_workflow_id)
         .bind(&instance.workflow_definition_id)
         .bind(instance.version as i32)
         .bind(&instance.description)
@@ -306,12 +318,14 @@ impl WorkflowRepository for PostgresPersistenceProvider {
 
         sqlx::query(
             r#"UPDATE wfc.workflows SET
-               name=$2, definition_id=$3, version=$4, description=$5, reference=$6,
-               status=$7, data=$8, next_execution=$9, create_time=$10, complete_time=$11
+               name=$2, root_workflow_id=$3, definition_id=$4, version=$5,
+               description=$6, reference=$7, status=$8, data=$9, next_execution=$10,
+               create_time=$11, complete_time=$12
                WHERE id=$1"#,
         )
         .bind(&instance.id)
         .bind(&instance.name)
+        .bind(&instance.root_workflow_id)
         .bind(&instance.workflow_definition_id)
         .bind(instance.version as i32)
         .bind(&instance.description)
@@ -396,6 +410,7 @@ impl WorkflowRepository for PostgresPersistenceProvider {
         Ok(WorkflowInstance {
             id: row.get("id"),
             name: row.get("name"),
+            root_workflow_id: row.get("root_workflow_id"),
             workflow_definition_id: row.get("definition_id"),
             version: row.get::<i32, _>("version") as u32,
             description: row.get("description"),
@@ -832,6 +847,7 @@ impl PersistenceProvider for PostgresPersistenceProvider {
             r#"CREATE TABLE IF NOT EXISTS wfc.workflows (
                 id TEXT PRIMARY KEY,
                 name TEXT NOT NULL UNIQUE,
+                root_workflow_id TEXT,
                 definition_id TEXT NOT NULL,
                 version INT NOT NULL,
                 description TEXT,
@@ -864,6 +880,13 @@ impl PersistenceProvider for PostgresPersistenceProvider {
                     CREATE UNIQUE INDEX IF NOT EXISTS idx_workflows_name
                         ON wfc.workflows (name);
                 END IF;
+                IF NOT EXISTS (
+                    SELECT 1 FROM information_schema.columns
+                    WHERE table_schema = 'wfc' AND table_name = 'workflows'
+                      AND column_name = 'root_workflow_id'
+                ) THEN
+                    ALTER TABLE wfc.workflows ADD COLUMN root_workflow_id TEXT;
+                END IF;
             END$$;"#,
         )
         .execute(&self.pool)

wfe-sqlite/src/lib.rs

@@ -58,6 +58,7 @@ impl SqlitePersistenceProvider {
             "CREATE TABLE IF NOT EXISTS workflows (
                 id TEXT PRIMARY KEY,
                 name TEXT NOT NULL UNIQUE,
+                root_workflow_id TEXT,
                 definition_id TEXT NOT NULL,
                 version INTEGER NOT NULL,
                 description TEXT,
@@ -250,6 +251,9 @@ fn row_to_workflow(
     Ok(WorkflowInstance {
         id: row.try_get("id").map_err(to_persistence_err)?,
         name: row.try_get("name").map_err(to_persistence_err)?,
+        root_workflow_id: row
+            .try_get("root_workflow_id")
+            .map_err(to_persistence_err)?,
         workflow_definition_id: row.try_get("definition_id").map_err(to_persistence_err)?,
         version: row
             .try_get::<i64, _>("version")
@@ -429,6 +433,15 @@ impl WorkflowRepository for SqlitePersistenceProvider {
         } else {
             instance.id.clone()
         };
+        // Fall back to the UUID when the caller didn't assign a human name.
+        // Production callers go through `WorkflowHost::start_workflow` which
+        // always fills this in, but test fixtures and external callers
+        // shouldn't trip the UNIQUE constraint.
+        let name = if instance.name.is_empty() {
+            id.clone()
+        } else {
+            instance.name.clone()
+        };
 
         let status_str = serde_json::to_value(instance.status)
             .map_err(|e| WfeError::Persistence(e.to_string()))?
@@ -443,11 +456,12 @@ impl WorkflowRepository for SqlitePersistenceProvider {
         let mut tx = self.pool.begin().await.map_err(to_persistence_err)?;
 
         sqlx::query(
-            "INSERT INTO workflows (id, name, definition_id, version, description, reference, status, data, next_execution, create_time, complete_time)
-             VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10, ?11)",
+            "INSERT INTO workflows (id, name, root_workflow_id, definition_id, version, description, reference, status, data, next_execution, create_time, complete_time)
+             VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10, ?11, ?12)",
         )
         .bind(&id)
-        .bind(&instance.name)
+        .bind(&name)
+        .bind(&instance.root_workflow_id)
         .bind(&instance.workflow_definition_id)
         .bind(instance.version as i64)
         .bind(&instance.description)
@@ -482,11 +496,13 @@ impl WorkflowRepository for SqlitePersistenceProvider {
         let mut tx = self.pool.begin().await.map_err(to_persistence_err)?;
 
         sqlx::query(
-            "UPDATE workflows SET name = ?1, definition_id = ?2, version = ?3, description = ?4, reference = ?5,
-             status = ?6, data = ?7, next_execution = ?8, complete_time = ?9
-             WHERE id = ?10",
+            "UPDATE workflows SET name = ?1, root_workflow_id = ?2, definition_id = ?3,
+             version = ?4, description = ?5, reference = ?6, status = ?7, data = ?8,
+             next_execution = ?9, complete_time = ?10
+             WHERE id = ?11",
         )
         .bind(&instance.name)
+        .bind(&instance.root_workflow_id)
         .bind(&instance.workflow_definition_id)
         .bind(instance.version as i64)
         .bind(&instance.description)
@@ -532,11 +548,13 @@ impl WorkflowRepository for SqlitePersistenceProvider {
         let mut tx = self.pool.begin().await.map_err(to_persistence_err)?;
 
         sqlx::query(
-            "UPDATE workflows SET name = ?1, definition_id = ?2, version = ?3, description = ?4, reference = ?5,
-             status = ?6, data = ?7, next_execution = ?8, complete_time = ?9
-             WHERE id = ?10",
+            "UPDATE workflows SET name = ?1, root_workflow_id = ?2, definition_id = ?3,
+             version = ?4, description = ?5, reference = ?6, status = ?7, data = ?8,
+             next_execution = ?9, complete_time = ?10
+             WHERE id = ?11",
         )
         .bind(&instance.name)
+        .bind(&instance.root_workflow_id)
         .bind(&instance.workflow_definition_id)
         .bind(instance.version as i64)
         .bind(&instance.description)

wfe/src/host.rs

@@ -36,8 +36,10 @@ impl HostContext for HostContextImpl {
         definition_id: &str,
         version: u32,
         data: serde_json::Value,
+        parent_root_workflow_id: Option<String>,
     ) -> Pin<Box<dyn Future<Output = Result<String>> + Send + '_>> {
         let def_id = definition_id.to_string();
+        let parent_root = parent_root_workflow_id;
         Box::pin(async move {
             // Look up the definition.
             let reg = self.registry.read().await;
@@ -54,6 +56,11 @@ impl HostContext for HostContextImpl {
                 instance.execution_pointers.push(ExecutionPointer::new(0));
             }
 
+            // Inherit the parent's root so every descendant of a given
+            // top-level workflow lands in the same Kubernetes namespace
+            // and can share a provisioned volume.
+            instance.root_workflow_id = parent_root;
+
             // Auto-assign a human-friendly name before persisting so the
             // child shows up as `{definition_id}-{N}` in lookups and logs.
             // Sub-workflows always use the default; callers wanting a custom