Have BKILL signal SIGKILL after SIGTERM

src/clib/lib/job_queue/lsf_driver.cpp

@@ -735,24 +735,42 @@ void lsf_driver_kill_job(void *_driver, void *_job) {
         char **argv = (char **)calloc(2, sizeof *argv);
         CHECK_ALLOC(argv);
         argv[0] = driver->remote_lsf_server;
-        argv[1] = saprintf("%s %s %s", driver->bkill_cmd, "-s SIGKILL",
+        argv[1] = saprintf("%s %s %s", driver->bkill_cmd, "-s SIGTERM",
                            job->lsf_jobnr_char);
-
         spawn_blocking(driver->rsh_cmd, 2, (const char **)argv, NULL, NULL);
-
         free(argv[1]);
         free(argv);
+
+        char **argv2 = (char **)calloc(2, sizeof *argv2);
+        argv2[0] = driver->remote_lsf_server;
+        argv2[1] = saprintf("%s %s %s %s %s", "sleep 30;", driver->bkill_cmd,
+                            "-s", "SIGKILL", job->lsf_jobnr_char);
+        spawn(driver->rsh_cmd, 2, (const char **)argv2, NULL, NULL);
+        free(argv2[0]);
+        free(argv2);
+
     } else if (driver->submit_method == LSF_SUBMIT_LOCAL_SHELL) {
         char **argv = (char **)calloc(3, sizeof *argv);
         CHECK_ALLOC(argv);
         argv[0] = saprintf("%s", "-s");
-        argv[1] = saprintf("%s", "SIGKILL");
+        argv[1] = saprintf("%s", "SIGTERM");
         argv[2] = saprintf("%s", job->lsf_jobnr_char);
         spawn_blocking(driver->bkill_cmd, 3, (const char **)argv, NULL, NULL);
         free(argv[0]);
         free(argv[1]);
         free(argv[2]);
         free(argv);
+
+        char **argv2 = (char **)calloc(2, sizeof *argv2);
+        CHECK_ALLOC(argv2);
+        argv2[0] = saprintf("%s", "-c");
+        argv2[1] = saprintf("%s %s %s %s %s", "sleep 30;", driver->bkill_cmd,
+                            "-s", "SIGKILL", job->lsf_jobnr_char);
+        spawn((const char *)saprintf("%s", "/bin/sh"), 2, (const char **)argv2,
+              "/dev/null", "/dev/null");
+        free(argv2[0]);
+        free(argv2[1]);
+        free(argv2);
     }
 }
 

src/ert/scheduler/lsf_driver.py

@@ -7,6 +7,7 @@
 import shlex
 import shutil
 import stat
+import subprocess
 import tempfile
 from pathlib import Path
 from typing import (
@@ -115,6 +116,7 @@ def __init__(
         self._jobs: MutableMapping[str, Tuple[int, AnyJob]] = {}
         self._iens2jobid: MutableMapping[int, str] = {}
         self._max_attempt: int = 100
+        self._sleep_time_between_bkills = 30
         self._sleep_time_between_cmd_retries = 3
         self._bsub_retries = 10
 
@@ -191,12 +193,17 @@ async def kill(self, iens: int) -> None:
         process = await asyncio.create_subprocess_exec(
             self._bkill_cmd,
             "-s",
-            "SIGKILL",
+            "SIGTERM",
             job_id,
             stdout=asyncio.subprocess.PIPE,
             stderr=asyncio.subprocess.PIPE,
         )
         stdout, stderr = await process.communicate()
+        _sigkill_process = subprocess.Popen(
+            f"sleep {self._sleep_time_between_bkills}; {self._bkill_cmd} -s SIGKILL {job_id}",
+            shell=True,
+            start_new_session=True,
+        )
         if process.returncode:
             logger.error(
                 f"LSF kill failed with returncode {process.returncode} "

tests/unit_tests/scheduler/test_lsf_driver.py

@@ -267,20 +267,33 @@ async def test_kill(
     bkill_path.write_text(
         f"#!/bin/sh\necho '{bkill_stdout}'\n"
         f"echo '{bkill_stderr}' >&2\n"
-        f"echo $@ > 'bkill_args'\n"
+        f"echo $@ >> 'bkill_args'\n"
         f"exit {bkill_returncode}",
         encoding="utf-8",
     )
     bkill_path.chmod(bkill_path.stat().st_mode | stat.S_IEXEC)
 
     driver = LsfDriver()
     driver._iens2jobid = mocked_iens2jobid
+    driver._sleep_time_between_bkills = 0
+
     await driver.kill(iens_to_kill)
+
+    async def wait_for_sigkill_in_file():
+        while True:
+            bkill_args_file_content = Path("bkill_args").read_text(encoding="utf-8")
+
+            if "-s SIGKILL" in bkill_args_file_content:
+                break
+            await asyncio.sleep(0.1)
+
     if expected_logged_error:
         assert expected_logged_error in caplog.text
     else:
-        bkill_args = Path("bkill_args").read_text(encoding="utf-8").strip()
-        assert f"-s SIGKILL {mocked_iens2jobid[iens_to_kill]}" in bkill_args
+        bkill_args = Path("bkill_args").read_text(encoding="utf-8").strip().split("\n")
+        assert f"-s SIGTERM {mocked_iens2jobid[iens_to_kill]}" in bkill_args
+
+        await asyncio.wait_for(wait_for_sigkill_in_file(), timeout=5)
 
 
 @given(st.text())