add podmonitor to coredb spec (#809)

charts/tembo-operator/Chart.yaml

@@ -3,7 +3,7 @@ name: tembo-operator
 description: "Helm chart to deploy the tembo-operator"
 type: application
 icon: https://cloud.tembo.io/images/TemboElephant.png
-version: 0.5.3
+version: 0.6.0
 home: https://tembo.io
 sources:
   - https://github.com/tembo-io/tembo

charts/tembo-operator/templates/crd.yaml

@@ -653,6 +653,22 @@ spec:
                     image:
                       description: Defines the container image to use for the appService.
                       type: string
+                    metrics:
+                      description: Defines the metrics endpoints to be scraped by Prometheus. This implements a subset of features available by PodMonitorPodMetricsEndpoints.
+                      nullable: true
+                      properties:
+                        path:
+                          description: path to scrape metrics
+                          type: string
+                        port:
+                          description: port must be also exposed in one of AppService.routing[]
+                          format: uint16
+                          minimum: 0.0
+                          type: integer
+                      required:
+                      - path
+                      - port
+                      type: object
                     middlewares:
                       description: Defines the ingress middeware configuration for the appService. This is specifically configured for the ingress controller Traefik.
                       items:

tembo-operator/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
 name = "controller"
 description = "Tembo Operator for Postgres"
-version = "0.47.3"
+version = "0.48.0"
 edition = "2021"
 default-run = "controller"
 license = "Apache-2.0"

tembo-operator/src/app_service/manager.rs

@@ -38,6 +38,8 @@ use super::{
 
 use crate::{app_service::types::IngressType, secret::fetch_all_decoded_data_from_secret};
 
+const APP_CONTAINER_PORT_PREFIX: &str = "app-";
+
 // private wrapper to hold the AppService Resources
 #[derive(Clone, Debug)]
 struct AppServiceResources {
@@ -48,6 +50,7 @@ struct AppServiceResources {
     ingress_tcp_routes: Option<Vec<IngressRouteTCPRoutes>>,
     entry_points: Option<Vec<String>>,
     entry_points_tcp: Option<Vec<String>>,
+    podmonitor: Option<podmon::PodMonitor>,
 }
 
 // generates Kubernetes Deployment and Service templates for a AppService
@@ -76,11 +79,13 @@ fn generate_resource(
         coredb_name,
         &resource_name,
         namespace,
-        oref,
+        oref.clone(),
         annotations,
-        placement,
+        placement.clone(),
     );
 
+    let maybe_podmonitor = generate_podmonitor(appsvc, &resource_name, namespace, annotations);
+
     // If DATA_PLANE_BASEDOMAIN is not set, don't generate IngressRoutes, IngressRouteTCPs, or EntryPoints
     if domain.is_none() {
         return AppServiceResources {
@@ -91,6 +96,7 @@ fn generate_resource(
             ingress_tcp_routes: None,
             entry_points: None,
             entry_points_tcp: None,
+            podmonitor: maybe_podmonitor,
         };
     }
     // It's safe to unwrap domain here because we've already checked if it's None
@@ -158,6 +164,7 @@ fn generate_resource(
         ingress_tcp_routes,
         entry_points,
         entry_points_tcp,
+        podmonitor: maybe_podmonitor,
     }
 }
 
@@ -194,7 +201,7 @@ fn generate_service(
                     port: p as i32,
                     // there can be more than one ServicePort per Service
                     // these must be unique, so we'll use the port number
-                    name: Some(format!("http-{}", p)),
+                    name: Some(format!("{APP_CONTAINER_PORT_PREFIX}{p}")),
                     target_port: None,
                     ..ServicePort::default()
                 })
@@ -279,6 +286,7 @@ fn generate_deployment(
         let container_ports: Vec<ContainerPort> = distinct_ports
             .into_iter()
             .map(|p| ContainerPort {
+                name: Some(format!("{APP_CONTAINER_PORT_PREFIX}{p}")),
                 container_port: p as i32,
                 protocol: Some("TCP".to_string()),
                 ..ContainerPort::default()
@@ -585,7 +593,6 @@ pub fn to_delete(desired: Vec<String>, actual: Vec<String>) -> Option<Vec<String
 
 async fn apply_resources(resources: Vec<AppServiceResources>, client: &Client, ns: &str) -> bool {
     let deployment_api: Api<Deployment> = Api::namespaced(client.clone(), ns);
-    let service_api: Api<Service> = Api::namespaced(client.clone(), ns);
     let ps = PatchParams::apply("cntrlr").force();
 
     let mut has_errors: bool = false;
@@ -612,6 +619,8 @@ async fn apply_resources(resources: Vec<AppServiceResources>, client: &Client, n
         if res.service.is_none() {
             continue;
         }
+
+        let service_api: Api<Service> = Api::namespaced(client.clone(), ns);
         match service_api
             .patch(&res.name, &ps, &Patch::Apply(&res.service))
             .await
@@ -629,6 +638,50 @@ async fn apply_resources(resources: Vec<AppServiceResources>, client: &Client, n
                 );
             }
         }
+
+        let podmon_api: Api<podmon::PodMonitor> = Api::namespaced(client.clone(), ns);
+        if let Some(mut pmon) = res.podmonitor {
+            // assign ownership of the PodMonitor to the Service
+            // if Service is deleted, so is the PodMonitor
+            let meta = service_api.get(&res.name).await;
+            if let Ok(svc) = meta {
+                let uid = svc.metadata.uid.unwrap_or_default();
+                let oref = OwnerReference {
+                    api_version: "v1".to_string(),
+                    kind: "Service".to_string(),
+                    name: res.name.clone(),
+                    uid,
+                    controller: Some(true),
+                    block_owner_deletion: Some(true),
+                };
+                pmon.metadata.owner_references = Some(vec![oref]);
+            }
+            match podmon_api
+                .patch(&res.name, &ps, &Patch::Apply(&pmon))
+                .await
+                .map_err(Error::KubeError)
+            {
+                Ok(_) => {
+                    debug!("ns: {}, applied PodMonitor: {}", ns, res.name);
+                }
+                Err(e) => {
+                    has_errors = true;
+                    error!(
+                        "ns: {}, failed to apply PodMonitor for AppService: {}, error: {}",
+                        ns, res.name, e
+                    );
+                }
+            }
+        } else {
+            match podmon_api.delete(&res.name, &Default::default()).await.ok() {
+                Some(_) => {
+                    debug!("ns: {}, deleted PodMonitor: {}", ns, res.name);
+                }
+                None => {
+                    debug!("ns: {}, PodMonitor does not exist: {}", ns, res.name);
+                }
+            }
+        }
     }
     has_errors
 }
@@ -947,6 +1000,51 @@ pub async fn prepare_apps_connection_secret(client: Client, cdb: &CoreDB) -> Res
     Ok(())
 }
 
+use crate::prometheus::podmonitor_crd as podmon;
+
+fn generate_podmonitor(
+    appsvc: &AppService,
+    resource_name: &str,
+    namespace: &str,
+    annotations: &BTreeMap<String, String>,
+) -> Option<podmon::PodMonitor> {
+    let metrics = appsvc.metrics.clone()?;
+
+    let mut selector_labels: BTreeMap<String, String> = BTreeMap::new();
+    selector_labels.insert("app".to_owned(), resource_name.to_string());
+
+    let mut labels = selector_labels.clone();
+    labels.insert("component".to_owned(), COMPONENT_NAME.to_owned());
+    labels.insert("coredb.io/name".to_owned(), namespace.to_owned());
+
+    let podmon_metadata = ObjectMeta {
+        name: Some(resource_name.to_string()),
+        namespace: Some(namespace.to_owned()),
+        labels: Some(labels.clone()),
+        annotations: Some(annotations.clone()),
+        ..ObjectMeta::default()
+    };
+
+    let metrics_endpoint = podmon::PodMonitorPodMetricsEndpoints {
+        path: Some(metrics.path),
+        port: Some(format!("{APP_CONTAINER_PORT_PREFIX}{}", metrics.port)),
+        ..podmon::PodMonitorPodMetricsEndpoints::default()
+    };
+
+    let pmonspec = podmon::PodMonitorSpec {
+        pod_metrics_endpoints: Some(vec![metrics_endpoint]),
+        selector: podmon::PodMonitorSelector {
+            match_labels: Some(selector_labels.clone()),
+            ..podmon::PodMonitorSelector::default()
+        },
+        ..podmon::PodMonitorSpec::default()
+    };
+    Some(podmon::PodMonitor {
+        metadata: podmon_metadata,
+        spec: pmonspec,
+    })
+}
+
 #[cfg(test)]
 mod tests {
     use crate::{apis::coredb_types::CoreDB, app_service::manager::generate_appsvc_annotations};

tembo-operator/src/app_service/types.rs

@@ -120,6 +120,10 @@ pub struct AppService {
     /// See the [Kubernetes docs](https://kubernetes.io/docs/tasks/configure-pod-container/configure-liveness-readiness-startup-probes/).
     pub probes: Option<Probes>,
 
+    /// Defines the metrics endpoints to be scraped by Prometheus.
+    /// This implements a subset of features available by PodMonitorPodMetricsEndpoints.
+    pub metrics: Option<AppMetrics>,
+
     /// Defines the ingress middeware configuration for the appService.
     /// This is specifically configured for the ingress controller Traefik.
     pub middlewares: Option<Vec<Middleware>>,
@@ -146,6 +150,14 @@ pub fn default_resources() -> ResourceRequirements {
     }
 }
 
+#[derive(Clone, Debug, Serialize, Deserialize, ToSchema, JsonSchema, PartialEq)]
+pub struct AppMetrics {
+    /// port must be also exposed in one of AppService.routing[]
+    pub port: u16,
+    /// path to scrape metrics
+    pub path: String,
+}
+
 // Secrets are injected into the container as environment variables
 // ths allows users to map these secrets to environment variable of their choice
 #[derive(Clone, Debug, Serialize, Deserialize, ToSchema, JsonSchema, PartialEq)]

tembo-operator/src/lib.rs

@@ -17,6 +17,7 @@ pub use metrics::Metrics;
 mod config;
 pub mod defaults;
 pub mod errors;
+pub mod prometheus;
 
 pub mod cloudnativepg;
 mod deployment_postgres_exporter;

tembo-operator/src/prometheus/mod.rs

@@ -0,0 +1 @@
+pub mod podmonitor_crd;