Merge pull request #195 from PanDAWMS/next

3.12.2.2

Author: PalNilsson

PILOTVERSION

@@ -1 +1 @@
-3.12.1.99
+3.12.2.2

pilot/common/errorcodes.py

@@ -17,7 +17,7 @@
 # under the License.
 #
 # Authors:
-# - Paul Nilsson, paul.nilsson@cern.ch, 2017-25
+# - Paul Nilsson, paul.nilsson@cern.ch, 2017-26
 # - Wen Guan, wen.guan@cern.ch, 2018
 
 """Error codes set by the pilot."""
@@ -397,6 +397,7 @@ def add_error_code(
         pilot_error_diags = ErrorCodes.pilot_error_diags
         if errorcode not in pilot_error_codes:
             error_msg = msg if msg else self.get_error_message(errorcode)
+            errorcode = int(errorcode)  # make sure errorcode is an int (in case it was passed as a string)
             if priority:
                 pilot_error_codes.insert(0, errorcode)
                 pilot_error_diags.insert(0, error_msg)

pilot/scripts/stagein.py

@@ -379,7 +379,7 @@ def extract_error_info(errc: str) -> (int, str):
 
     _code = re.search(r'error code: (\d+)', errc)
     if _code:
-        error_code = _code.group(1)
+        error_code = int(_code.group(1))  # int, e.g.  1099
 
     _msg = re.search('details: (.+)', errc)
     if _msg:

pilot/scripts/stageout.py

@@ -277,7 +277,7 @@ def extract_error_info(_err: str) -> tuple:
 
     _code = re.search(r'error code: (\d+)', _err)
     if _code:
-        error_code = _code.group(1)
+        error_code = int(_code.group(1))
 
     _msg = re.search('details: (.+)', _err)
     if _msg:

pilot/user/atlas/common.py

@@ -2009,6 +2009,7 @@ def get_redundants() -> list:
                 "/panda_pilot*",
                 "/work",
                 "README*",
+                "CLAUDE.md",
                 "MANIFEST*",
                 "*.part*",
                 "docs",

pilot/user/atlas/setup.py

@@ -417,7 +417,7 @@ def get_payload_environment_variables(cmd: str, job_id: int, task_id: str, attem
     if analysis_job:
         variables.append('export ROOT_TTREECACHE_SIZE=1;')
         try:
-            core_count = int(os.environ.get('ATHENA_PROC_NUMBER'))
+            core_count = int(os.environ.get('ATHENA_PROC_NUMBER') or os.environ.get('ATHENA_PROC_NUMBER_JOB'))
         except Exception:
             _core_count = 'export ROOTCORE_NCPUS=1;'
         else:

pilot/util/constants.py

@@ -27,8 +27,8 @@
 # Pilot version
 RELEASE = '3'   # released number should be fixed at 3 for Pilot 3
 VERSION = '12'  # version number is '1' for first release, '0' until then, increased for bigger updates
-REVISION = '1'  # revision number should be reset to '0' for every new version release, increased for small updates
-BUILD = '99'        # build number should be reset to '1' for every new development cycle
+REVISION = '2'  # revision number should be reset to '0' for every new version release, increased for small updates
+BUILD = '2'        # build number should be reset to '1' for every new development cycle
 
 SUCCESS = 0
 FAILURE = 1

pilot/util/https.py

@@ -1446,9 +1446,9 @@ def add_error_codes(data: dict, job: JobData) -> None:
     pilot_error_codes = job.piloterrorcodes
     if pilot_error_codes != []:
         logger.warning(f'pilot_error_code(s) = {pilot_error_codes} (will report primary/first error code)')
-        data['pilot_error_code'] = pilot_error_codes[0]
+        data['pilot_error_code'] = int(pilot_error_codes[0])
     else:
-        data['pilot_error_code'] = pilot_error_code
+        data['pilot_error_code'] = int(pilot_error_code)
 
     def remove_timestamp(log_entry: str) -> str:
         """Strip a ``YYYY-MM-DD HH:MM:SS[,mmm]`` timestamp from a log entry.

pilot/util/middleware.py

@@ -373,7 +373,7 @@ def handle_updated_job_object(job: JobData, xdata: list, label: str = "stage-in"
         error_code = file_dictionary["error"][1]
         if error_code:
             job.piloterrorcodes, job.piloterrordiags = errors.add_error_code(
-                error_code, msg=error_diag
+                int(error_code), msg=error_diag
             )
     else:
         msg = f"{label} file dictionary not found"