Merge branch 'CMSSW_14_0_X' into scouting_nano_test_14_0_15

CondCore/Utilities/python/tier0.py

@@ -11,7 +11,7 @@
 
 import pycurl
 
-tier0Url = 'https://cmsweb.cern.ch/t0wmadatasvc/prod/'
+tier0Url = os.getenv('TIER0_API_URL', 'https://cmsweb.cern.ch/t0wmadatasvc/prod/')
 
 class Tier0Error(Exception):
     '''Tier0 exception.
@@ -23,7 +23,7 @@ def __init__(self, message):
 
 def unique(seq, keepstr=True):
     t = type(seq)
-    if t in (unicode, str):
+    if t is str:
         t = (list, t('').join)[bool(keepstr)]
     try:
         remaining = set(seq)
@@ -38,44 +38,52 @@ def unique(seq, keepstr=True):
             seen = []
             return t(c for c in seq if not (c in seen or seen.append(c)))
 
+#note: this exception seems unused
 class ResponseError( Tier0Error ):
 
-    def __init__( self, curl, response, proxy, timeout ):
+    def __init__( self, curl, response, proxy, timeout, maxTime ):
         super( ResponseError, self ).__init__( response )
         self.args += ( curl, proxy )
         self.timeout = timeout
-
-    def __str__( self ):
-        errStr = """Wrong response for curl connection to Tier0DataSvc from URL \"%s\"""" %( self.args[1].getinfo( self.args[1].EFFECTIVE_URL ), )
-        if self.args[ -1 ]:
-            errStr += """ using proxy \"%s\"""" %( str( self.args[ -1 ] ), )
-        errStr += """ with timeout \"%d\" with error code \"%d\".""" %( self.timeout, self.args[1].getinfo( self.args[1].RESPONSE_CODE) )
-        if self.args[0].find( '<p>' ) != -1:
-            errStr += """\nFull response: \"%s\".""" %( self.args[0].partition('<p>')[-1].rpartition('</p>')[0], )
+        self.maxTime = maxTime
+
+    def __str__(self):
+        errStr = f'Wrong response for curl connection to Tier0DataSvc'\
+                 f' from URL "{self.args[1].getinfo(self.args[1].EFFECTIVE_URL)}"'
+        if self.args[-1]:
+            errStr += f' using proxy "{str(self.args[-1])}"'
+        errStr += f' with connection-timeout "{self.timeout}", max-time "{self.maxtime}"'\
+                  f' with error code "{self.args[1].getinfo(self.args[1].RESPONSE_CODE)}".'
+        if '<p>' in self.args[0]:
+            full_response = self.args[0].partition('<p>')[-1].rpartition('</p>')[0]
+            errStr += f'\nFull response: "{full_response}".'
         else:
-            errStr += """\nFull response: \"%s\".""" %( self.args[0], )
+            errStr += f'\nFull response: "{self.args[0]}".'
+
         return errStr
 
 #TODO: Add exceptions for each category of HTTP error codes
 #TODO: check response code and raise corresponding exceptions
-
-def _raise_http_error( curl, response, proxy, timeout ):
-    raise ResponseError( curl, response, proxy, timeout )
+#note: this function seems to be unused
+def _raise_http_error( curl, response, proxy, timeout, maxTime ):
+    raise ResponseError( curl, response, proxy, timeout, maxTime )
 
 class Tier0Handler( object ):
 
-    def __init__( self, uri, timeOut, retries, retryPeriod, proxy, debug ):
+    def __init__( self, uri, timeOut, maxTime, retries, retryPeriod, proxy, debug ):
         """
         Parameters:
         uri: Tier0DataSvc URI;
-        timeOut: time out for Tier0DataSvc HTTPS calls;
+        timeOut: time out for connection of Tier0DataSvc HTTPS calls [seconds];
+        maxTime: maximum time for Tier0DataSvc HTTPS calls (including data transfer) [seconds];
         retries: maximum retries for Tier0DataSvc HTTPS calls;
-        retryPeriod: sleep time between two Tier0DataSvc HTTPS calls;
+        retryPeriod: sleep time between two Tier0DataSvc HTTPS calls [seconds];
         proxy: HTTP proxy for accessing Tier0DataSvc HTTPS calls;
         debug: if set to True, enables debug information.
         """
         self._uri = uri
         self._timeOut = timeOut
+        self._maxTime = maxTime
         self._retries = retries
         self._retryPeriod = retryPeriod
         self._proxy = proxy
@@ -90,51 +98,69 @@ def unsetDebug( self ):
     def setProxy( self, proxy ):
         self._proxy = proxy
 
-    def _queryTier0DataSvc( self, url ):
-        """
-        Queries Tier0DataSvc.
-        url: Tier0DataSvc URL.
-        @returns: dictionary, from whence the required information must be retrieved according to the API call.
-        Raises if connection error, bad response, or timeout after retries occur.
-        """
+    def _getCerts( self ) -> str:
+        cert_path = os.getenv('X509_USER_CERT', '')
+        key_path = os.getenv('X509_USER_KEY', '')
 
-        userAgent = "User-Agent: ConditionWebServices/1.0 python/%d.%d.%d PycURL/%s" % ( sys.version_info[ :3 ] + ( pycurl.version_info()[ 1 ], ) )
-
-        proxy = ""
-        if self._proxy: proxy = ' --proxy=%s ' % self._proxy
-
-        debug = " -s -S "
-        if self._debug: debug = " -v "
+        certs = ""
+        if cert_path:
+            certs += f' --cert {cert_path}'
+        else:
+            logging.warning("No certificate provided for Tier0 access, use X509_USER_CERT and"
+                            " optionally X509_USER_KEY env variables to specify the path to the cert"
+                            " (and the key unless included in the cert file)")
+        if key_path:
+            certs += f' --key {key_path}'
+        return certs
+
+    def _curlQueryTier0( self, url:str, force_debug:bool = False, force_cert:bool = False):
+        userAgent = "User-Agent: ConditionWebServices/1.0 python/%d.%d.%d PycURL/%s" \
+            % ( sys.version_info[ :3 ] + ( pycurl.version_info()[ 1 ], ) )
+        debug = "-v" if self._debug or force_debug else "-s -S"
+
+        proxy = f"--proxy {self._proxy}" if self._proxy else ""
+        certs = self._getCerts() if not self._proxy or force_cert else ""
 
-        cmd = '/usr/bin/curl -k -L --user-agent "%s" %s --connect-timeout %i --retry %i %s %s ' % (userAgent, proxy, self._timeOut, self._retries, debug, url)
+        cmd = f'/usr/bin/curl -k -L --user-agent "{userAgent}" {proxy}'\
+              f' --connect-timeout {self._timeOut} --max-time {self._maxTime} --retry {self._retries}'\
+              f' {debug} {url} {certs}'
 
         # time the curl to understand if re-tries have been carried out
         start = time.time()
         process = subprocess.Popen(cmd, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
         (stdoutdata, stderrdata) =  process.communicate()
-        retcode = process.returncode
         end = time.time()
+        return process.returncode, stdoutdata, stderrdata, end-start
+
+    def _queryTier0DataSvc( self, url ):
+        """
+        Queries Tier0DataSvc.
+        url: Tier0DataSvc URL.
+        @returns: dictionary, from whence the required information must be retrieved according to the API call.
+        Raises if connection error, bad response, or timeout after retries occur.
+        """
+
+        retcode, stdoutdata, stderrdata, query_time = self._curlQueryTier0(url)
 
         if retcode != 0 or stderrdata:
-
-           # if the first curl has failed, logg its stderror and prepare and independent retry
-           msg = "looks like curl returned an error: retcode=%s and took %s seconds" % (retcode,(end-start),)
-           msg += ' msg = "'+str(stderrdata)+'"'
-           logging.error(msg)
-
-           time.sleep(10)
-           cmd = '/usr/bin/curl -k -L --user-agent "%s" %s --connect-timeout %i --retry %i %s %s ' % (userAgent, proxy, self._timeOut, self._retries, "-v", url)
-           process = subprocess.Popen(cmd, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
-           (stdoutdata, stderrdata) =  process.communicate()
-           retcode = process.returncode
-           if retcode != 0:
-              msg = "looks like curl returned an error for the second time: retcode=%s" % (retcode,)
-              msg += ' msg = "'+str(stderrdata)+'"'
-              logging.error(msg)
-              raise Tier0Error(msg)
-           else :
-              msg = "curl returned ok upon the second try"
-              logging.info(msg)
+
+            # if the first curl has failed, logg its stderror and prepare and independent retry
+            msg = "looks like curl returned an error: retcode=%s and took %s seconds" % (retcode, query_time,)
+            msg += ' msg = "'+str(stderrdata)+'"'
+            logging.error(msg)
+            if self._proxy:
+                logging.info("before assumed proxy provides authentication, now trying with both proxy and certificate")
+
+            time.sleep(self._retryPeriod)
+            retcode, stdoutdata, stderrdata, query_time = self._curlQueryTier0(url, force_debug=True, force_cert=True)
+            if retcode != 0:
+                msg = "looks like curl returned an error for the second time: retcode=%s" % (retcode,)
+                msg += ' msg = "'+str(stderrdata)+'"'
+                logging.error(msg)
+                raise Tier0Error(msg)
+            else:
+                msg = "curl returned ok upon the second try"
+                logging.info(msg)
         resp = json.loads( ''.join(stdoutdata.decode()).replace( "'", '"').replace(' None', ' "None"') )
         return resp
 
@@ -149,7 +175,8 @@ def getFirstSafeRun( self ):
         firstConditionSafeRunAPI = "firstconditionsaferun"
         safeRunDict = self._queryTier0DataSvc( os.path.join( self._uri, firstConditionSafeRunAPI ) )
         if safeRunDict is None:
-            errStr = """First condition safe run is not available in Tier0DataSvc from URL \"%s\"""" %( os.path.join( self._uri, firstConditionSafeRunAPI ), )
+            errStr = """First condition safe run is not available in Tier0DataSvc from URL \"%s\"""" \
+                %( os.path.join( self._uri, firstConditionSafeRunAPI ), )
             if self._proxy:
                 errStr += """ using proxy \"%s\".""" %( str( self._proxy ), )
             raise Tier0Error( errStr )
@@ -164,19 +191,20 @@ def getGlobalTag( self, config ):
         Raises if connection error, bad response, timeout after retries occur, or if no Global Tags are available.
         """
         data = self._queryTier0DataSvc( os.path.join( self._uri, config ) )
-        gtnames = sorted(unique( [ str( di[ 'global_tag' ] ) for di in data['result'] if di[ 'global_tag' ] is not None ] ))
+        gtnames = sorted(unique( [ str( di['global_tag'] ) for di in data['result'] if di['global_tag'] is not None ] ))
         try:
             recentGT = gtnames[-1]
             return recentGT
         except IndexError:
-            errStr = """No Global Tags for \"%s\" are available in Tier0DataSvc from URL \"%s\"""" %( config, os.path.join( self._uri, config ) )
+            errStr = """No Global Tags for \"%s\" are available in Tier0DataSvc from URL \"%s\"""" \
+                %( config, os.path.join( self._uri, config ) )
             if self._proxy:
                 errStr += """ using proxy \"%s\".""" %( str( self._proxy ), )
         raise Tier0Error( errStr )
 
 
 def test( url ):
-    t0 = Tier0Handler( url, 1, 1, 1, None, debug=False)
+    t0 = Tier0Handler( url, 1, 5, 1, 10, None, debug=False)
 
     print('   fcsr = %s (%s)' % (t0.getFirstSafeRun(), type(t0.getFirstSafeRun()) ))
     print('   reco_config = %s' % t0.getGlobalTag('reco_config'))
@@ -186,4 +214,3 @@ def test( url ):
 
 if __name__ == '__main__':
     test( tier0Url )
-

CondCore/Utilities/scripts/conddb

@@ -706,12 +706,13 @@ def _get_hlt_fcsr( session, timeType ):
 
 def _get_prompt_fcsr( session, timeType ):
     tier0timeout = 5
+    tier0maxtime = 60
     tier0retries = 3 
     tier0retryPeriod = 5
     tier0proxy = None
     try:
         t0DataSvc = Tier0Handler( tier0Url,
-                                  tier0timeout, tier0retries, tier0retryPeriod,
+                                  tier0timeout, tier0maxtime, tier0retries, tier0retryPeriod,
                                   tier0proxy, False )
         try:
             fcsr = t0DataSvc.getFirstSafeRun()

Configuration/AlCa/python/autoCond.py

@@ -37,9 +37,9 @@
     'run3_data_express'            :    '140X_dataRun3_Express_frozen_v1',
     # GlobalTag for Run3 data relvals (prompt GT) - 140X_dataRun3_Prompt_v3 but snapshot at 2024-05-31 09:09:12 (UTC)
     'run3_data_prompt'             :    '140X_dataRun3_Prompt_frozen_v3',
-    # GlobalTag for Run3 offline data reprocessing - snapshot at 2024-02-07 16:38:59 (UTC)
-    'run3_data'                    :    '140X_dataRun3_v4',
-    # GlobalTag for Run3 offline data reprocessing with Prompt GT, currenlty for 2022FG - snapshot at 2024-02-12 12:00:00 (UTC)
+    # GlobalTag for Run3 offline data reprocessing - snapshot at 2024-09-04 16:25:09 (UTC)
+    'run3_data'                    :    '140X_dataRun3_v9',
+    # GlobalTag for Run3 offline data reprocessing with Prompt GT, currently for 2022FG - snapshot at 2024-02-12 12:00:00 (UTC)
     'run3_data_PromptAnalysis'     :    '140X_dataRun3_PromptAnalysis_v2',
     # GlobalTag for MC production with perfectly aligned and calibrated detector for Phase1 2017 (and 0,0,~0-centred beamspot)
     'phase1_2017_design'           :    '131X_mc2017_design_v3',

Configuration/Eras/python/Era_Run3_2024_cff.py

@@ -1,7 +1,7 @@
 import FWCore.ParameterSet.Config as cms
 
 from Configuration.Eras.Era_Run3_cff import Run3
-from Configuration.Eras.Modifier_run3_2024_L1T_cff import run3_2024_L1T
+from Configuration.Eras.Modifier_stage2L1Trigger_2024_cff import stage2L1Trigger_2024
 from Configuration.Eras.Modifier_run3_scouting_nanoAOD_post2023_cff import run3_scouting_nanoAOD_post2023
 
-Run3_2024 = cms.ModifierChain(Run3, run3_2024_L1T, run3_scouting_nanoAOD_post2023)
+Run3_2024 = cms.ModifierChain(Run3, stage2L1Trigger_2024, run3_scouting_nanoAOD_post2023)

Configuration/Eras/python/Modifier_run3_2024_L1T_cff.py → Configuration/Eras/python/Modifier_stage2L1Trigger_2024_cff.py

@@ -1,3 +1,3 @@
 import FWCore.ParameterSet.Config as cms
 
-run3_2024_L1T = cms.Modifier()
+stage2L1Trigger_2024 = cms.Modifier()

Configuration/PyReleaseValidation/python/upgradeWorkflowComponents.py

@@ -1639,7 +1639,8 @@ def setup_(self, step, stepName, stepDict, k, properties):
         '--customise' : 'HeterogeneousCore/AlpakaServices/customiseAlpakaServiceMemoryFilling.customiseAlpakaServiceMemoryFilling',
     },
     harvest = {
-        '-s': 'HARVESTING:@trackingOnlyValidation+@pixelTrackingOnlyDQM'
+        '-s': 'HARVESTING:@trackingOnlyValidation+@pixelTrackingOnlyDQM',
+        '--procModifiers': 'alpakaValidation',
     },
     suffix = 'Patatrack_PixelOnlyAlpaka_Validation',
     offset = 0.403,

DQM/EcalMonitorClient/python/IntegrityClient_cfi.py

@@ -27,7 +27,7 @@
             kind = cms.untracked.string('TH2F'),
             otype = cms.untracked.string('Ecal3P'),
             btype = cms.untracked.string('Crystal'),
-            description = cms.untracked.string('Summary of the data integrity. A channel is red if more than ' + str(errFractionThreshold) + ' of its entries have integrity errors.')
+            description = cms.untracked.string('Summary of the data integrity. A channel is red if more than ' + str(errFractionThreshold) + ' of its entries have integrity errors. Also, an entire SuperModule can show red if more than 0.01 of its entries have DCC-SRP or DCC-TCC Desync errors.')
         ),
         Quality = cms.untracked.PSet(
             path = cms.untracked.string('%(subdet)s/%(prefix)sIntegrityClient/%(prefix)sIT data integrity quality %(sm)s'),

DQM/EcalMonitorClient/src/IntegrityClient.cc

@@ -123,15 +123,20 @@ namespace ecaldqm {
       }
     }
 
-    // Quality check: set an entire FED to BAD if "any" DCC-SRP or DCC-TCC mismatch errors are detected
+    // Quality check: set an entire FED to BAD if "any" DCC-SRP or DCC-TCC mismatch errors are detected AND the number of events affected by the DCC-SRP or DCC-TCC mismatch errors is more than 1% of the events analyzed in the run
     // Fill mismatch statistics
     MESet const& sBXSRP(sources_.at("BXSRP"));
     MESet const& sBXTCC(sources_.at("BXTCC"));
     std::vector<bool> hasMismatchDCC(nDCC, false);
     for (unsigned iDCC(0); iDCC < nDCC; ++iDCC) {
-      if (sBXSRP.getBinContent(getEcalDQMSetupObjects(), iDCC + 1) > 50. ||
-          sBXTCC.getBinContent(getEcalDQMSetupObjects(), iDCC + 1) > 50.)  // "any" => 50
-        hasMismatchDCC[iDCC] = true;
+      int nBXSRPdesync = sBXSRP.getBinContent(getEcalDQMSetupObjects(), iDCC + 1);
+      int nBXTCCdesync = sBXTCC.getBinContent(getEcalDQMSetupObjects(), iDCC + 1);
+
+      if (nBXSRPdesync > 50. || nBXTCCdesync > 50.)  { // "any" => 50
+        if (nBXSRPdesync > int(0.01 * processedEvents) || nBXTCCdesync > int(0.01 * processedEvents)) { // check if the events with DCC-SRP or DCC-TCC desyncs for the given DCC is more than 1% of the events analyzed
+          hasMismatchDCC[iDCC] = true;
+        }
+      }
     }
     // Analyze mismatch statistics
     for (MESet::iterator qsItr(meQualitySummary.beginChannel(GetElectronicsMap()));

DQM/EcalMonitorTasks/python/RawDataTask_cfi.py

@@ -49,6 +49,34 @@
 
 ecalRawDataTask = cms.untracked.PSet(
     MEs = cms.untracked.PSet(
+        TrendBXTCC = cms.untracked.PSet(
+            path = cms.untracked.string('Ecal/Trends/RawDataTask number of %(prefix)sRDT bunch crossing TCC errors'),
+            kind = cms.untracked.string('TH1F'),
+            otype = cms.untracked.string('Ecal2P'),
+            btype = cms.untracked.string('Trend'),
+            description = cms.untracked.string('Trend of the number of bunch crossing value mismatches between DCC and TCC.')
+        ),
+        TrendL1ATCC = cms.untracked.PSet(
+            path = cms.untracked.string('Ecal/Trends/RawDataTask number of %(prefix)sRDT L1A TCC errors'),
+            kind = cms.untracked.string('TH1F'),
+            otype = cms.untracked.string('Ecal2P'),
+            btype = cms.untracked.string('Trend'),
+            description = cms.untracked.string('Trend of the number of L1A value mismatches between DCC and TCC.')
+        ),
+        TrendBXSRP = cms.untracked.PSet(
+            path = cms.untracked.string('Ecal/Trends/RawDataTask number of %(prefix)sRDT bunch crossing SRP errors'),
+            kind = cms.untracked.string('TH1F'),
+            otype = cms.untracked.string('Ecal2P'),
+            btype = cms.untracked.string('Trend'),
+            description = cms.untracked.string('Trend of the number of bunch crossing value mismatches between DCC and SRP.')
+        ),
+        TrendL1ASRP = cms.untracked.PSet(
+            path = cms.untracked.string('Ecal/Trends/RawDataTask number of %(prefix)sRDT L1A SRP errors'),
+            kind = cms.untracked.string('TH1F'),
+            otype = cms.untracked.string('Ecal2P'),
+            btype = cms.untracked.string('Trend'),
+            description = cms.untracked.string('Trend of the number of L1A value mismatches between DCC and SRP.')
+        ),
         BXSRP = cms.untracked.PSet(
             path = cms.untracked.string('%(subdet)s/%(prefix)sRawDataTask/%(prefix)sRDT bunch crossing SRP errors'),
             kind = cms.untracked.string('TH1F'),