feat: log debug pod information before deleting pods by agent

Author: mb-ii

clearml_agent/glue/definitions.py

@@ -30,3 +30,8 @@
 ENV_DEFAULT_SCHEDULER_QUEUE_TAGS = EnvEntry(
     "K8S_GLUE_DEFAULT_SCHEDULER_QUEUE_TAGS", default=["k8s-glue"], converter=shlex.split
 )
+
+ENV_LOG_POD_STATUS_BEFORE_DELETING = EnvEntry("K8S_GLUE_LOG_POD_STATUS_BEFORE_DELETING", default=False, converter=bool)
+"""
+If True, log pod status and exit code before deleting it.
+"""

clearml_agent/glue/k8s.py

@@ -42,6 +42,7 @@
     ENV_POD_USE_IMAGE_ENTRYPOINT,
     ENV_KUBECTL_IGNORE_ERROR,
     ENV_DEFAULT_SCHEDULER_QUEUE_TAGS,
+    ENV_LOG_POD_STATUS_BEFORE_DELETING,
 )
 from .._vendor import pyyaml as yaml
 
@@ -220,6 +221,7 @@ def __init__(
         self.ignore_kubectl_errors_re = (
             re.compile(ENV_KUBECTL_IGNORE_ERROR.get()) if ENV_KUBECTL_IGNORE_ERROR.get() else None
         )
+        self.log_pod_status_before_deleting = ENV_LOG_POD_STATUS_BEFORE_DELETING.get()
 
     @property
     def agent_label(self):
@@ -431,7 +433,7 @@ def get_pods(self, filters: List[str] = None, debug_msg: str = None):
             output_config = json.loads(output)
         except Exception as ex:
             self.log.warning('Failed parsing kubectl output:\n{}\nEx: {}'.format(output, ex))
-            return
+            return []
         return output_config.get('items', [])
 
     def _get_pod_count(self, extra_labels: List[str] = None, msg: str = None):
@@ -490,7 +492,6 @@ def run_one_task(self, queue: Text, task_id: Text, worker_args=None, task_sessio
                             "task": task_id,
                             "queue": self.k8s_pending_queue_id,
                             "status_reason": "k8s pending scheduler",
-                            "update_execution_queue": False,
                         }
                     )
                     if res.ok:
@@ -1037,7 +1038,74 @@ def _process_bash_lines_response(self, bash_cmd: str, raise_error=True):
         ]
         return lines
 
+    def _log_pod_statuses(self, pods: List[Dict]):
+        """
+        Log pod status and exit codes to the task's console log.
+        :param pods: A list of pod dictionaries to log.
+        :param msg: A message to log.
+        """
+        if not pods:
+            return
+        for pod in pods:
+            pod_name = get_path(pod, "metadata", "name")
+            task_id = pod_name[len(self.pod_name_prefix) :]
+
+            # Log pod status and exit codes
+            pod_status = get_path(pod, "status", "phase")
+            ctr_statuses = get_path(pod, "status", "containerStatuses") or []
+
+            if pod_status.lower() == "succeeded":
+                continue
+
+            log_lines = []
+            log_lines.append(
+                "Pod '{pod_name}' resulted in status '{pod_status}'".format(
+                    pod_name=pod_name, pod_status=pod_status
+                )
+            )
+
+            for ctr_status in ctr_statuses:
+                ctr_name = ctr_status.get("name")
+                ctr_state = get_path(ctr_status, "state")
+                ctr_reason = get_path(ctr_state, "terminated", "reason")
+                ctr_exit_code = get_path(ctr_state, "terminated", "exitCode")
+                log_lines.append(
+                    "Container '{name}' resulted with reason '{reason}' and exit code '{exit_code}'".format(
+                        name=ctr_name, reason=ctr_reason, exit_code=ctr_exit_code
+                    )
+                )
+
+            try:
+                self.send_logs(task_id, [os.linesep.join(log_lines)])
+            except Exception as ex:
+                self.log.warning(f"Failed sending pod status logs for task {task_id}: {ex}")
+
+    def _delete_pods_by_names(self, names: List[str], namespace: str, msg: str = None) -> List[str]:
+        if not names:
+            return []
+        kubectl_cmd = "kubectl delete pod --namespace={ns} {names} --output=name".format(
+            ns=namespace, names=" ".join(names)
+        )
+        self.log.debug("Deleting pods by name {}: {}".format(
+            msg or "", kubectl_cmd
+        ))
+        lines = self._process_bash_lines_response(kubectl_cmd)
+        self.log.debug(" - deleted pods by name %s", ", ".join(lines))
+        return lines
+
     def _delete_pods(self, selectors: List[str], namespace: str, msg: str = None) -> List[str]:
+        if self.log_pod_status_before_deleting:
+            pods_to_delete = self.get_pods(
+                filters=selectors,
+                debug_msg="Getting pods to delete: {cmd}",
+            )
+            self._log_pod_statuses(pods_to_delete)
+            return self._delete_pods_by_names(
+                    [get_path(p, "metadata", "name") for p in pods_to_delete if get_path(p, "metadata", "name")],
+                    namespace,
+                    msg=msg
+                )
+
         kubectl_cmd = \
             "kubectl delete pod -l={agent_label} " \
             "--namespace={namespace} --field-selector={selector} --output name".format(
@@ -1089,6 +1157,8 @@ def _delete_completed_or_failed_pods(self, namespace, msg: str = None):
             debug_msg="Deleting failed pods: {cmd}"
         )
         if failed_pods:
+            if self.log_pod_status_before_deleting:
+                self._log_pod_statuses(failed_pods)
             job_names_to_delete = {
                 get_path(pod, "metadata", "labels", "job-name"): get_path(pod, "metadata", "namespace")
                 for pod in failed_pods

clearml_agent/helper/docker_args.py

@@ -294,7 +294,6 @@ class CustomTemplate(Template):
     ${TASK.hyperparams.properties.user_key.value}
     """
 
-    idpattern = r'(?a:[_a-z][_a-z0-9|.|:]*)'
     prefix = "CLEARML_"
     queue_id_to_name_map = {}