Remove DO_KILL_NODE _FAILURE and use DO_KILL state instead if needed

xjules · Aug 25, 2023 · 5a43497 · 5a43497
1 parent a0e1259
commit 5a43497
Show file tree

Hide file tree

Showing 7 changed files with 3 additions and 9 deletions.
diff --git a/src/_ert_com_protocol/status_type_enum.py b/src/_ert_com_protocol/status_type_enum.py
@@ -15,7 +15,6 @@
     "RUNNING_EXIT_CALLBACK": "STEP_RUNNING",
     "STATUS_FAILURE": "STEP_UNKNOWN",
     "FAILED": "STEP_FAILED",
-    "DO_KILL_NODE_FAILURE": "STEP_FAILED",
     "UNKNOWN": "STEP_UNKNOWN",
 }
 

diff --git a/src/clib/lib/include/ert/job_queue/job_queue_status.hpp b/src/clib/lib/include/ert/job_queue/job_queue_status.hpp
@@ -26,7 +26,6 @@ struct job_queue_status_struct {
         JOB_QUEUE_RUNNING_EXIT_CALLBACK, // Temporary state, while running requested callbacks after an ended job - controlled by job_queue
         JOB_QUEUE_STATUS_FAILURE, // The driver call to get status has failed, job status remains unchanged
         JOB_QUEUE_FAILED, // Job has failed, no more retries, FINAL STATE
-        JOB_QUEUE_DO_KILL_NODE_FAILURE, // Job has failed, node should be blacklisted
         JOB_QUEUE_UNKNOWN}; // Unable to get status from submitted job
 };
 

diff --git a/src/clib/lib/include/ert/job_queue/job_status.hpp b/src/clib/lib/include/ert/job_queue/job_status.hpp
@@ -90,7 +90,6 @@ typedef enum {
     JOB_QUEUE_RUNNING_EXIT_CALLBACK = 2048,
     JOB_QUEUE_STATUS_FAILURE = 4096,
     JOB_QUEUE_FAILED = 8192,
-    JOB_QUEUE_DO_KILL_NODE_FAILURE = 16384,
     JOB_QUEUE_UNKNOWN = 32768
 } job_status_type;
 
@@ -109,7 +108,6 @@ const std::map<const job_status_type, const std::string> job_status_names = {
     {JOB_QUEUE_RUNNING_EXIT_CALLBACK, "JOB_QUEUE_RUNNING_EXIT_CALLBACK"},
     {JOB_QUEUE_STATUS_FAILURE, "JOB_QUEUE_STATUS_FAILURE"},
     {JOB_QUEUE_FAILED, "JOB_QUEUE_FAILED"},
-    {JOB_QUEUE_DO_KILL_NODE_FAILURE, "JOB_QUEUE_DO_KILL_NODE_FAILURE"},
     {JOB_QUEUE_UNKNOWN, "JOB_QUEUE_UNKNOWN"},
 };
 
@@ -123,7 +121,7 @@ const std::map<const job_status_type, const std::string> job_status_names = {
    */
 #define JOB_QUEUE_CAN_KILL                                                     \
     (JOB_QUEUE_WAITING + JOB_QUEUE_RUNNING + JOB_QUEUE_PENDING +               \
-     JOB_QUEUE_SUBMITTED + JOB_QUEUE_DO_KILL + JOB_QUEUE_DO_KILL_NODE_FAILURE)
+     JOB_QUEUE_SUBMITTED + JOB_QUEUE_DO_KILL)
 
 #define JOB_QUEUE_CAN_UPDATE_STATUS                                            \
     (JOB_QUEUE_RUNNING + JOB_QUEUE_PENDING + JOB_QUEUE_SUBMITTED +             \

diff --git a/src/clib/lib/job_queue/job_node.cpp b/src/clib/lib/job_queue/job_node.cpp
@@ -244,7 +244,7 @@ ERT_CLIB_SUBMODULE("queue", m) {
                     MAX_CONFIRMED_WAIT, node->job_name, node->submit_attempt);
                 logger->info(error_msg);
                 msg = error_msg;
-                job_status_type new_status = JOB_QUEUE_DO_KILL_NODE_FAILURE;
+                job_status_type new_status = JOB_QUEUE_DO_KILL;
                 job_queue_node_set_status(node, new_status);
             }
         }

diff --git a/src/clib/old_tests/job_queue/test_job_status.cpp b/src/clib/old_tests/job_queue/test_job_status.cpp
@@ -157,7 +157,6 @@ void test_index() {
     add_and_count_job_status(status, JOB_QUEUE_RUNNING_EXIT_CALLBACK, count++);
     add_and_count_job_status(status, JOB_QUEUE_STATUS_FAILURE, count++);
     add_and_count_job_status(status, JOB_QUEUE_FAILED, count++);
-    add_and_count_job_status(status, JOB_QUEUE_DO_KILL_NODE_FAILURE, count++);
     job_queue_status_free(status);
 }
 

diff --git a/src/ert/job_queue/job_queue_node.py b/src/ert/job_queue/job_queue_node.py
@@ -358,7 +358,7 @@ def _log_kill_thread_stopping_status(self) -> None:
     DONE_STATES = [
         JobStatus.SUCCESS,
         JobStatus.IS_KILLED,
-        JobStatus.DO_KILL_NODE_FAILURE,
+        JobStatus.DO_KILL,
     ]
     FAILURE_STATES = [JobStatus.FAILED]
 

diff --git a/src/ert/job_queue/queue.py b/src/ert/job_queue/queue.py
@@ -67,7 +67,6 @@
     "RUNNING_EXIT_CALLBACK": _FM_STEP_RUNNING,
     "STATUS_FAILURE": _FM_STEP_UNKNOWN,
     "FAILED": _FM_STEP_FAILURE,
-    "DO_KILL_NODE_FAILURE": _FM_STEP_FAILURE,
     "UNKNOWN": _FM_STEP_UNKNOWN,
 }