chore: Follow-ups to the containerdebug log collection

CHANGELOG.md

@@ -13,7 +13,7 @@
 - Add `/ready` endpoint to the operator Deployment, which reports the CRD installation status ([#868]).
 - Webservers now have a default affinity to the OPA Pods when OPA authorization is configured ([#872]).
 - The logs of the `containerdebug` process are now collected by the Vector agent and forwarded to
-  the Vector aggregator ([#888]).
+  the Vector aggregator ([#888], [#892]).
 
 ### Changed
 
@@ -112,6 +112,7 @@
 [#886]: https://github.com/stackabletech/airflow-operator/pull/886
 [#887]: https://github.com/stackabletech/airflow-operator/pull/887
 [#888]: https://github.com/stackabletech/airflow-operator/pull/888
+[#892]: https://github.com/stackabletech/airflow-operator/pull/892
 
 ## [26.7.0] - 2026-07-21
 

rust/operator-binary/src/controller/build/properties/env_vars.rs

@@ -1,5 +1,6 @@
 use std::{collections::BTreeSet, str::FromStr};
 
+use const_format::concatcp;
 use stackable_operator::{
     constant,
     crd::{authentication::oidc, git_sync},
@@ -188,9 +189,11 @@ pub fn build_airflow_statefulset_envs(
     // other telemetry-based process in it would log into the same directory and the Vector
     // agent would label its events with `container: containerdebug`.
     env_vars = env_vars
+        // The directory must be exactly one level below the log directory, because the Vector
+        // agent derives the container name from that path segment.
         .with_value(
             &FILE_LOG_DIRECTORY,
-            format!("{STACKABLE_LOG_DIR}/containerdebug"),
+            concatcp!(STACKABLE_LOG_DIR, "/containerdebug"),
         )
         .with_value(&FILE_LOG_ROTATION_PERIOD, "hourly")
         .with_value(&FILE_LOG_MAX_FILES, "2")

rust/operator-binary/src/controller/build/volumes.rs

@@ -32,11 +32,9 @@ pub fn create_volumes(
         name: LOG_VOLUME_NAME.to_string(),
         empty_dir: Some(EmptyDirVolumeSource {
             medium: None,
-            // Only the Airflow log files are budgeted here. The `containerdebug` files in this
-            // volume are rotated by time, which bounds their number but not their size, so they
-            // are an unbudgeted writer. They stay negligible at the default INFO level and a 30
-            // minute loop, but raising `FILE_LOG_LEVEL` via `envOverrides` can exceed the limit,
-            // which makes the kubelet evict the Pod.
+            // Only the Airflow log files are budgeted. The `containerdebug` files in this
+            // volume are rotated by time, not by size, so they have no budget; raising
+            // `FILE_LOG_LEVEL` via `envOverrides` can therefore evict the Pod.
             size_limit: Some(product_logging::framework::calculate_log_volume_size_limit(
                 &[MAX_LOG_FILES_SIZE],
             )),

rust/operator-binary/src/crd/mod.rs

@@ -776,7 +776,13 @@ impl Deref for AirflowRole {
 }
 
 fn container_debug_command() -> String {
-    format!("containerdebug --output={STACKABLE_LOG_DIR}/containerdebug-state.json --loop &")
+    // The log directory is created here, because `FILE_LOG_MAX_FILES` makes the telemetry library
+    // prune old log files before it creates the first one, which complains about the directory not
+    // existing yet.
+    format!(
+        "mkdir --parents \"$FILE_LOG_DIRECTORY\"\n\
+         containerdebug --output={STACKABLE_LOG_DIR}/containerdebug-state.json --loop &"
+    )
 }
 
 #[derive(Clone, Debug, Deserialize, JsonSchema, PartialEq, Serialize)]