Updates and some new multi-core monitoring.

CompareCollectors.sh

@@ -0,0 +1,19 @@
+#!/bin/sh
+COLLECTOR1=$1
+COLLECTOR2=$2
+
+list_all_glidein_names() {
+  condor_status -pool $COLLECTOR1 -format '%s\n' Name 
+  condor_status -pool $COLLECTOR2 -format '%s\n' Name 
+}
+
+
+unique=`list_all_glidein_names | sort | uniq | wc -l`
+regto2=`list_all_glidein_names | sort | uniq -c | awk '($1==2){print $0}' | wc -l`
+pct=$[$[$unique-$regto2]*100/$unique]
+
+echo Unique pilots on either collector:    $unique
+echo Pilots registered to both collectors: $regto2
+echo  ... troublesome pilots = $pct \%
+
+exit 0

condor_check

@@ -45,7 +45,7 @@ echo
 # Frontend ClassAds in the Collector:
 echo Frontend ClassAds in the Collector:
 condor_status -pool $COLLECTOR1 -any \ 
-  -const '(GlideinMyType=?=\"glideresource\")' \
+  -const '(GlideinMyType=?="glideresource")' \
   -format '%s\n' GlideClientName | sort | uniq -c
 echo
 

multi-core-t1.sh

@@ -0,0 +1,285 @@
+#!/bin/sh
+
+if [ -z $glideinWMSMonitor_RELEASE_DIR ] ; then
+  echo "ERROR: glideinWMSMonitor source code missing."
+  exit 1
+else
+  source $glideinWMSMonitor_RELEASE_DIR/bashrc
+fi
+
+
+writeoutjsonfile() {
+
+COLLECTOR=$1
+
+echo "{"
+echo "  \"Multi-core pilot monitoring\": {"
+echo "    \"Collector\": \"$COLLECTOR\","
+echo "    \"Time\": `/bin/date +%s`,"
+
+
+echo "    \"Partitionable glideins\": {"
+echo "      \"header\": [\"State\",\"Activity\",\"glideins\"],"
+echo "      \"data\": ["
+condor_status -pool $COLLECTOR        \
+-const '(regexp("T1_",GLIDEIN_CMSSite)=?=True)' \
+-const '(SlotType=?="Partitionable")' \
+-format '%s ' State -format '%s\n' Activity | sort | uniq -c | \
+awk '{printf("        [\"%s\",\"%s\",%i],\n",$2,$3,$1)}' 
+echo "        [null,null,0]"
+echo "      ]"
+echo "    },"
+
+
+echo "    \"Partitionable glidein Cpus\": {"
+echo "      \"header\": [\"State\",\"Activity\",\"Cpus\"],"
+echo "      \"data\": ["
+condor_status -pool $COLLECTOR        \
+-const '(regexp("T1_",GLIDEIN_CMSSite)=?=True)' \
+-const '(SlotType=?="Partitionable")' \
+-format '%s ' Cpus  -format '%s ' State -format '%s\n' Activity | \
+awk ' { for (i=$1; i>0; i--) { print $2 " " $3 } }' | sort | uniq -c | \
+awk '{printf("        [\"%s\",\"%s\",%i],\n",$2,$3,$1)}' 
+echo "        [null,null,0]"
+echo "      ]"
+echo "    },"
+
+
+echo "    \"Partitionable retiring glideins\": {"
+echo "      \"header\": [\"State\",\"Activity\",\"glideins\"],"
+echo "      \"data\": ["
+condor_status -pool $COLLECTOR        \
+-const '(regexp("T1_",GLIDEIN_CMSSite)=?=True)' \
+-const '(SlotType=?="Partitionable")' \
+-const '(GLIDEIN_ToRetire=!=UNDEFINED)&&(CurrentTime>GLIDEIN_ToRetire)' \
+-format '%s ' State -format '%s\n' Activity | sort | uniq -c | \
+awk '{printf("        [\"%s\",\"%s\",%i],\n",$2,$3,$1)}' 
+echo "        [null,null,0]"
+echo "      ]"
+echo "    },"
+
+
+echo "    \"Partitionable retiring glidein Cpus\": {"
+echo "      \"header\": [\"State\",\"Activity\",\"Cpus\"],"
+echo "      \"data\": ["
+condor_status -pool $COLLECTOR        \
+-const '(regexp("T1_",GLIDEIN_CMSSite)=?=True)' \
+-const '(SlotType=?="Partitionable")' \
+-const '(GLIDEIN_ToRetire=!=UNDEFINED)&&(CurrentTime>GLIDEIN_ToRetire)' \
+-format '%s ' Cpus  -format '%s ' State -format '%s\n' Activity | \
+awk ' { for (i=$1; i>0; i--) { print $2 " " $3 } }' | sort | uniq -c | \
+awk '{printf("        [\"%s\",\"%s\",%i],\n",$2,$3,$1)}' 
+echo "        [null,null,0]"
+echo "      ]"
+echo "    },"
+
+
+echo "    \"Dynamic glideins\": {"
+echo "      \"header\": [\"State\",\"Activity\",\"glideins\"],"
+echo "      \"data\": ["
+condor_status -pool $COLLECTOR        \
+-const '(regexp("T1_",GLIDEIN_CMSSite)=?=True)' \
+-const '(SlotType=?="Dynamic")' \
+-format '%s ' State -format '%s\n' Activity | sort | uniq -c | \
+awk '{printf("        [\"%s\",\"%s\",%i],\n",$2,$3,$1)}' 
+echo "        [null,null,0]"
+echo "      ]"
+echo "    },"
+
+
+echo "    \"Dynamic glidein Cpus\": {"
+echo "      \"header\": [\"State\",\"Activity\",\"Cpus\"],"
+echo "      \"data\": ["
+condor_status -pool $COLLECTOR        \
+-const '(regexp("T1_",GLIDEIN_CMSSite)=?=True)' \
+-const '(SlotType=?="Dynamic")' \
+-format '%s ' Cpus  -format '%s ' State -format '%s\n' Activity | \
+awk ' { for (i=$1; i>0; i--) { print $2 " " $3 } }' | sort | uniq -c | \
+awk '{printf("        [\"%s\",\"%s\",%i],\n",$2,$3,$1)}' 
+echo "        [null,null,0]"
+echo "      ]"
+echo "    },"
+
+
+echo "    \"Dynamic retiring glideins\": {"
+echo "      \"header\": [\"State\",\"Activity\",\"glideins\"],"
+echo "      \"data\": ["
+condor_status -pool $COLLECTOR        \
+-const '(regexp("T1_",GLIDEIN_CMSSite)=?=True)' \
+-const '(SlotType=?="Dynamic")' \
+-const '(GLIDEIN_ToRetire=!=UNDEFINED)&&(CurrentTime>GLIDEIN_ToRetire)' \
+-format '%s ' State -format '%s\n' Activity | sort | uniq -c | \
+awk '{printf("        [\"%s\",\"%s\",%i],\n",$2,$3,$1)}' 
+echo "        [null,null,0]"
+echo "      ]"
+echo "    },"
+
+
+echo "    \"Dynamic retiring glidein Cpus\": {"
+echo "      \"header\": [\"State\",\"Activity\",\"Cpus\"],"
+echo "      \"data\": ["
+condor_status -pool $COLLECTOR        \
+-const '(regexp("T1_",GLIDEIN_CMSSite)=?=True)' \
+-const '(SlotType=?="Dynamic")' \
+-const '(GLIDEIN_ToRetire=!=UNDEFINED)&&(CurrentTime>GLIDEIN_ToRetire)' \
+-format '%s ' Cpus  -format '%s ' State -format '%s\n' Activity | \
+awk ' { for (i=$1; i>0; i--) { print $2 " " $3 } }' | sort | uniq -c | \
+awk '{printf("        [\"%s\",\"%s\",%i],\n",$2,$3,$1)}' 
+echo "        [null,null,0]"
+echo "      ]"
+echo "    },"
+
+
+echo "    \"Static glideins\": {"
+echo "      \"header\": [\"State\",\"Activity\",\"glideins\"],"
+echo "      \"data\": ["
+condor_status -pool $COLLECTOR        \
+-const '(regexp("T1_",GLIDEIN_CMSSite)=?=True)' \
+-const '(SlotType=?="Static")' \
+-format '%s ' State -format '%s\n' Activity | sort | uniq -c | \
+awk '{printf("        [\"%s\",\"%s\",%i],\n",$2,$3,$1)}' 
+echo "        [null,null,0]"
+echo "      ]"
+echo "    },"
+
+
+echo "    \"Static glidein Cpus\": {"
+echo "      \"header\": [\"State\",\"Activity\",\"Cpus\"],"
+echo "      \"data\": ["
+condor_status -pool $COLLECTOR        \
+-const '(regexp("T1_",GLIDEIN_CMSSite)=?=True)' \
+-const '(SlotType=?="Static")' \
+-format '%s ' Cpus  -format '%s ' State -format '%s\n' Activity | \
+awk ' { for (i=$1; i>0; i--) { print $2 " " $3 } }' | sort | uniq -c | \
+awk '{printf("        [\"%s\",\"%s\",%i],\n",$2,$3,$1)}' 
+echo "        [null,null,0]"
+echo "      ]"
+echo "    },"
+
+
+echo "    \"Static retiring glideins\": {"
+echo "      \"header\": [\"State\",\"Activity\",\"glideins\"],"
+echo "      \"data\": ["
+condor_status -pool $COLLECTOR        \
+-const '(regexp("T1_",GLIDEIN_CMSSite)=?=True)' \
+-const '(SlotType=?="Static")' \
+-const '(GLIDEIN_ToRetire=!=UNDEFINED)&&(CurrentTime>GLIDEIN_ToRetire)' \
+-format '%s ' State -format '%s\n' Activity | sort | uniq -c | \
+awk '{printf("        [\"%s\",\"%s\",%i],\n",$2,$3,$1)}' 
+echo "        [null,null,0]"
+echo "      ]"
+echo "    },"
+
+
+echo "    \"Static retiring glidein Cpus\": {"
+echo "      \"header\": [\"State\",\"Activity\",\"Cpus\"],"
+echo "      \"data\": ["
+condor_status -pool $COLLECTOR        \
+-const '(regexp("T1_",GLIDEIN_CMSSite)=?=True)' \
+-const '(SlotType=?="Static")' \
+-const '(GLIDEIN_ToRetire=!=UNDEFINED)&&(CurrentTime>GLIDEIN_ToRetire)' \
+-format '%s ' Cpus  -format '%s ' State -format '%s\n' Activity | \
+awk ' { for (i=$1; i>0; i--) { print $2 " " $3 } }' | sort | uniq -c | \
+awk '{printf("        [\"%s\",\"%s\",%i],\n",$2,$3,$1)}' 
+echo "        [null,null,0]"
+echo "      ]"
+echo "    },"
+
+
+echo "    \"Static multi-core glideins\": {"
+echo "      \"header\": [\"State\",\"Activity\",\"glideins\"],"
+echo "      \"data\": ["
+condor_status -pool $COLLECTOR        \
+-const '(regexp("T1_",GLIDEIN_CMSSite)=?=True)' \
+-const '(SlotType=?="Static")' \
+-format '%s ' Name  -format '%s ' State -format '%s\n' Activity | \
+grep ^slot[0-9]*\@ | awk '{print $2 " " $3 }' | sort | uniq -c | \
+awk '{printf("        [\"%s\",\"%s\",%i],\n",$2,$3,$1)}' 
+echo "        [null,null,0]"
+echo "      ]"
+echo "    },"
+
+
+echo "    \"Static multi-core glidein Cpus\": {"
+echo "      \"header\": [\"State\",\"Activity\",\"Cpus\"],"
+echo "      \"data\": ["
+condor_status -pool $COLLECTOR        \
+-const '(regexp("T1_",GLIDEIN_CMSSite)=?=True)' \
+-const '(SlotType=?="Static")' \
+-format '%s ' Name  -format '%s ' Cpus -format '%s ' State -format '%s\n' Activity | \
+grep ^slot[0-9]*\@ | awk '{print $2 " " $3 " " $4 }' | \
+awk ' { for (i=$1; i>0; i--) { print $2 " " $3 } }' | sort | uniq -c | \
+awk '{printf("        [\"%s\",\"%s\",%i],\n",$2,$3,$1)}' 
+echo "        [null,null,0]"
+echo "      ]"
+echo "    },"
+
+
+echo "    \"Static multi-core retiring glideins\": {"
+echo "      \"header\": [\"State\",\"Activity\",\"glideins\"],"
+echo "      \"data\": ["
+condor_status -pool $COLLECTOR        \
+-const '(regexp("T1_",GLIDEIN_CMSSite)=?=True)' \
+-const '(SlotType=?="Static")' \
+-const '(GLIDEIN_ToRetire=!=UNDEFINED)&&(CurrentTime>GLIDEIN_ToRetire)' \
+-format '%s ' Name  -format '%s ' State -format '%s\n' Activity | \
+grep ^slot[0-9]*\@ | awk '{print $2 " " $3 }' | sort | uniq -c | \
+awk '{printf("        [\"%s\",\"%s\",%i],\n",$2,$3,$1)}' 
+echo "        [null,null,0]"
+echo "      ]"
+echo "    },"
+
+
+echo "    \"Static multi-core retiring glidein Cpus\": {"
+echo "      \"header\": [\"State\",\"Activity\",\"Cpus\"],"
+echo "      \"data\": ["
+condor_status -pool $COLLECTOR        \
+-const '(regexp("T1_",GLIDEIN_CMSSite)=?=True)' \
+-const '(SlotType=?="Static")' \
+-const '(GLIDEIN_ToRetire=!=UNDEFINED)&&(CurrentTime>GLIDEIN_ToRetire)' \
+-format '%s ' Name  -format '%s ' Cpus -format '%s ' State -format '%s\n' Activity | \
+grep ^slot[0-9]*\@ | awk '{print $2 " " $3 " " $4 }' | \
+awk ' { for (i=$1; i>0; i--) { print $2 " " $3 } }' | sort | uniq -c | \
+awk '{printf("        [\"%s\",\"%s\",%i],\n",$2,$3,$1)}' 
+echo "        [null,null,0]"
+echo "      ]"
+echo "    },"
+
+
+echo "    \"Total glideins\": {"
+echo "      \"header\": [\"State\",\"Activity\",\"glideins\"],"
+echo "      \"data\": ["
+condor_status -pool $COLLECTOR \
+-const '(regexp("T1_",GLIDEIN_CMSSite)=?=True)' \
+-format '%s ' State -format '%s\n' Activity \
+| sort |uniq -c \
+| awk '{printf("        [\"%s\",\"%s\",%i],\n",$2,$3,$1)}'
+echo "        [null,null,0]"
+echo "      ]"
+echo "    },"
+
+
+echo "    \"Total Cpus\": {"
+echo "      \"header\": [\"State\",\"Activity\",\"Cpus\"],"
+echo "      \"data\": ["
+condor_status -pool $COLLECTOR \
+-const '(regexp("T1_",GLIDEIN_CMSSite)=?=True)' \
+-format '%s ' Cpus  -format '%s ' State -format '%s\n' Activity \
+| awk ' { for (i=$1; i>0; i--) { print $2 " " $3 } }' | sort |uniq -c \
+| awk '{printf("        [\"%s\",\"%s\",%i],\n",$2,$3,$1)}'
+echo "        [null,null,0]"
+echo "      ]"
+echo "    }"
+
+
+echo "  }"
+echo "}"
+
+
+return
+}
+
+JSONFILE=${glideinWMSMonitor_OUTPUT_DIR}-json/monitor-multicore-t1-global-`/bin/date +%F-Z%R -u`.json
+writeoutjsonfile vocms097.cern.ch > $JSONFILE
+
+exit