[Merged by Bors] - Add proper startup, liveness and readiness probes

CHANGELOG.md

@@ -9,9 +9,14 @@ All notable changes to this project will be documented in this file.
 - BREAKING: HDFS deep storage now configurable via HDFS discovery config map instead of an url to a HDFS name node ([#262]).
 - Include chart name when installing with a custom release name ([#263], [#264]).
 
+### Fixed
+
+- Add proper startup, liveness and readiness probes ([#273])
+
 [#262]: https://github.com/stackabletech/druid-operator/pull/262
 [#263]: https://github.com/stackabletech/druid-operator/pull/263
 [#264]: https://github.com/stackabletech/druid-operator/pull/264
+[#273]: https://github.com/stackabletech/druid-operator/pull/273
 
 ## [0.6.0] - 2022-06-30
 

rust/operator-binary/src/druid_controller.rs

@@ -2,6 +2,7 @@
 use crate::{
     config::{get_jvm_config, get_log4j_config},
     discovery::{self, build_discovery_configmaps},
+    probes::add_probes,
 };
 
 use snafu::{OptionExt, ResultExt, Snafu};
@@ -25,9 +26,7 @@ use stackable_operator::{
     k8s_openapi::{
         api::{
             apps::v1::{StatefulSet, StatefulSetSpec},
-            core::v1::{
-                ConfigMap, EnvVar, Probe, Service, ServicePort, ServiceSpec, TCPSocketAction,
-            },
+            core::v1::{ConfigMap, EnvVar, Service, ServicePort, ServiceSpec},
         },
         apimachinery::pkg::{apis::meta::v1::LabelSelector, util::intstr::IntOrString},
     },
@@ -596,17 +595,7 @@ fn build_rolegroup_statefulset(
             .build(),
     );
 
-    // readiness probe
-    let probe = Probe {
-        tcp_socket: Some(TCPSocketAction {
-            port: IntOrString::Int(role.get_http_port().into()),
-            ..Default::default()
-        }),
-        initial_delay_seconds: Some(30),
-        period_seconds: Some(5),
-        ..Default::default()
-    };
-    cb.readiness_probe(probe);
+    add_probes(&mut cb, &role);
 
     let mut container = cb.build();
     container.image_pull_policy = Some("IfNotPresent".to_string());

rust/operator-binary/src/main.rs

@@ -1,6 +1,7 @@
 mod config;
 mod discovery;
 mod druid_controller;
+mod probes;
 
 use std::sync::Arc;
 

rust/operator-binary/src/probes.rs

@@ -0,0 +1,71 @@
+use stackable_druid_crd::DruidRole;
+use stackable_operator::{
+    builder::ContainerBuilder,
+    k8s_openapi::{
+        api::core::v1::{HTTPGetAction, Probe},
+        apimachinery::pkg::util::intstr::IntOrString,
+    },
+};
+
+pub fn add_probes(container_builder: &mut ContainerBuilder, role: &DruidRole) {
+    // /status/selfDiscovered:
+    // Indicating whether the node has received a confirmation from the central node discovery mechanism (currently ZooKeeper) of the Druid cluster that the node has been added to the cluster.
+    // Returns 200 OK response with empty body if the node has discovered itself and 503 SERVICE UNAVAILABLE if the node hasn't discovered itself yet.
+    // It is recommended to not consider a Druid node "healthy" or "ready" in automated deployment/container management systems until it returns 200 OK response.
+    // see https://druid.apache.org/docs/latest/operations/api-reference.html#process-information
+    let startup_probe_path = "/status/selfDiscovered";
+    let liveness_probe_path = "/status/selfDiscovered";
+
+    let readiness_probe_path = match role {
+        // /druid/broker/v1/readiness:
+        // Returns if the Broker knows about all segments in the cluster.
+        // This can be used to know when a Broker process is ready to be queried after a restart.
+        // see https://druid.apache.org/docs/latest/operations/api-reference.html#broker
+        DruidRole::Broker => "/druid/broker/v1/readiness",
+
+        // /druid/historical/v1/readiness:
+        // Returns if all segments in the local cache have been loaded.
+        // This can be used to know when a Historical process is ready to be queried after a restart.
+        // see https://druid.apache.org/docs/latest/operations/api-reference.html#historical
+        DruidRole::Historical => "/druid/historical/v1/readiness",
+
+        // For the other roles we use the normal discovery-check
+        DruidRole::Coordinator | DruidRole::MiddleManager | DruidRole::Router => {
+            "/status/selfDiscovered"
+        }
+    };
+
+    container_builder.startup_probe(Probe {
+        failure_threshold: Some(60), // 60 * 10s = 10min time to start up and register itself
+        period_seconds: Some(10),
+        timeout_seconds: Some(3),
+        http_get: Some(HTTPGetAction {
+            port: IntOrString::Int(role.get_http_port() as i32),
+            path: Some(startup_probe_path.to_string()),
+            ..HTTPGetAction::default()
+        }),
+        ..Probe::default()
+    });
+    container_builder.liveness_probe(Probe {
+        failure_threshold: Some(6), // After not being healthy for 6 * 5s = 30s => restart
+        period_seconds: Some(5),
+        timeout_seconds: Some(3),
+        http_get: Some(HTTPGetAction {
+            port: IntOrString::Int(role.get_http_port() as i32),
+            path: Some(liveness_probe_path.to_string()),
+            ..HTTPGetAction::default()
+        }),
+        ..Probe::default()
+    });
+    container_builder.readiness_probe(Probe {
+        failure_threshold: Some(1), // After not being healthy for 1 * 5s = 5s => take it out of the service
+        period_seconds: Some(5),
+        timeout_seconds: Some(3),
+        http_get: Some(HTTPGetAction {
+            port: IntOrString::Int(role.get_http_port() as i32),
+            path: Some(readiness_probe_path.to_string()),
+            ..HTTPGetAction::default()
+        }),
+        ..Probe::default()
+    });
+}