Conflicts resolved with rel-v7r2

integration_tests.py

@@ -139,22 +139,24 @@ def create(
     prepare_environment(flags, editable, extra_module, release_var)
     install_server()
     install_client()
-    error = 0
+    exit_code = 0
     if run_server_tests:
         try:
             test_server()
         except TestExit as e:
-            error += e.exit_code
+            exit_code += e.exit_code
         else:
             raise NotImplementedError()
     if run_client_tests:
         try:
             test_client()
         except TestExit as e:
-            error += e.exit_code
+            exit_code += e.exit_code
         else:
             raise NotImplementedError()
-    raise typer.Exit(0)
+    if exit_code != 0:
+        typer.secho("One or more tests failed", err=True, fg=c.RED)
+    raise typer.Exit(exit_code)
 
 
 @app.command()

release.notes

@@ -28,6 +28,13 @@ CHANGE: (#4937) removed StatesMonitoringAgent (use StatesAccountingAgent agent i
 *tests
 CHANGE: (#5046) don't use mail in the self generated certificates
 
+[v7r2p5]
+
+FIX: fixes from v7r0p56, v7r1p39
+
+*WMS
+CHANGE: (#5102) JobCleaningAgent will first DELETE and only then REMOVE jobs
+
 [v7r2p4]
 
 *Core
@@ -199,6 +206,11 @@ NEW: (#4910) --runslow option on unit tests to allow faster local tests
 NEW: (#4938) added a helloworld test for the (yet to be implemented) cloud testing in certification
 CHANGE: (#4968) Change the defaults for tests (to MySQL 8 and ES 7)
 
+[v7r1p39]
+
+*WMS
+CHANGE: (#5121) for HTCondor, the SiteDirectory write the executable in the globally defined working directory
+
 [v7r1p38]
 
 FIX: fixes from v7r0p55
@@ -828,6 +840,14 @@ FIX: (#4551) align ProxyDB test to current changes
 NEW: (#4289) Document how to run integration tests in docker
 NEW: (#4551) add DNProperties description to Registry/Users subsection
 
+[v7r0p56]
+
+*Resources
+FIX: (#5119) HTCondorCE: Limit calls to actual cleanup (find and delete files on disk) to
+     once per minute per SiteDirector, fixes #5118
+CHANGE: (#5119) HTCondorCE cleanup: Run the DIRAC_ executable purge with -O3 and -maxdepth
+        1 to speed up the find
+
 [v7r0p55]
 
 *TS

src/DIRAC/Resources/Computing/HTCondorCEComputingElement.py

@@ -62,7 +62,9 @@
 except ImportError:
   # Python 3's subprocess module contains a compatibility layer
   import subprocess as commands
+import datetime
 import errno
+import threading
 
 from DIRAC import S_OK, S_ERROR, gConfig
 from DIRAC.Resources.Computing.ComputingElement import ComputingElement
@@ -164,6 +166,10 @@ class HTCondorCEComputingElement(ComputingElement):
       implementing the functions jobSubmit, getJobOutput
   """
 
+  # static variables to ensure single cleanup every minute
+  _lastCleanupTime = datetime.datetime.utcnow()
+  _cleanupLock = threading.Lock()
+
   #############################################################################
   def __init__(self, ceUniqueID):
     """ Standard constructor.
@@ -529,21 +535,33 @@ def __cleanup(self):
     # FIXME: again some issue with the working directory...
     # workingDirectory = self.ceParameters.get( 'WorkingDirectory', DEFAULT_WORKINGDIRECTORY )
 
+    if not HTCondorCEComputingElement._cleanupLock.acquire(False):
+      return
+
+    now = datetime.datetime.utcnow()
+    if (now - HTCondorCEComputingElement._lastCleanupTime).total_seconds() < 60:
+      HTCondorCEComputingElement._cleanupLock.release()
+      return
+
+    HTCondorCEComputingElement._lastCleanupTime = now
+
     self.log.debug("Cleaning working directory: %s" % self.workingDirectory)
 
     # remove all files older than 120 minutes starting with DIRAC_ Condor will
     # push files on submission, but it takes at least a few seconds until this
     # happens so we can't directly unlink after condor_submit
-    status, stdout = commands.getstatusoutput('find %s -mmin +120 -name "DIRAC_*" -delete ' % self.workingDirectory)
+    status, stdout = commands.getstatusoutput('find -O3 %s -maxdepth 1 -mmin +120 -name "DIRAC_*" -delete ' %
+                                              self.workingDirectory)
     if status:
       self.log.error("Failure during HTCondorCE __cleanup", stdout)
 
-    # remove all out/err/log files older than "DaysToKeepLogs" days in the CE part of the working Directory
-    workDir = os.path.join(self.workingDirectory, self.ceName)
-    findPars = dict(workDir=workDir, days=self.daysToKeepLogs)
+    # remove all out/err/log files older than "DaysToKeepLogs" days in the working directory
+    # not running this for each CE so we do global cleanup
+    findPars = dict(workDir=self.workingDirectory, days=self.daysToKeepLogs)
     # remove all out/err/log files older than "DaysToKeepLogs" days
     status, stdout = commands.getstatusoutput(
         r'find %(workDir)s -mtime +%(days)s -type f \( -name "*.out" -o -name "*.err" -o -name "*.log" \) -delete ' %
         findPars)
     if status:
       self.log.error("Failure during HTCondorCE __cleanup", stdout)
+    self._cleanupLock.release()