Changeset 667 for trunk

Timestamp:

05/11/12 18:16:10 (12 years ago)

Author:

sdipsl

Message:

• Curie : to overcome I/O issue during copy we loop over cp command. Only IGCM_sys_Put_Out and IGCM_sys_Get for now.

:Interrupted system call
:Cannot send after transport endpoint shutdown

File:

• trunk/libIGCM/libIGCM_sys/libIGCM_sys_curie.ksh (modified) (5 diffs)

trunk/libIGCM/libIGCM_sys/libIGCM_sys_curie.ksh

@@ -1180 +1180 @@
     echo "IGCM_sys_Put_Out :" $@
   fi
+
+  typeset NB_ESSAI DELAI RET i
+
+  # number of tentative
+  NB_ESSAI=10
+  # time delay between tentative
+  DELAI=10
+
   if [ $DRYRUN = 0 ]; then
     if [ ! -f ${1} ] ; then
@@ -1189 +1197 @@
     IGCM_sys_MkdirArchive $( dirname $2 )
     #
-    typeset RET
 
     #=====================================================
@@ -1211 +1218 @@
     fi
     #
+
+    (( i = 0 ))
+    while [ $i -lt $NB_ESSAI ] ; do
     # USUAL WAY
-    \cp $1 $2 > out_rsync 2>&1
-    RET=$?
+      \cp $1 $2 > out_rsync 2>&1
+      RET=$?
+      if [ ${RET} -gt 0 ]; then
+        IGCM_debug_Print 2 "IGCM_sys_Put_Out : cp failed ${i}/${NB_ESSAI}"
+        IGCM_debug_Print 2 "IGCM_sys_Put_Out : sleep ${DELAI} seconds and try again."
+        sleep $DELAI
+      else
+        break
+      fi
+      (( i = i + 1 ))
+    done
 
 #       #RSYNC WITH NETWORK SSH CALL
@@ -1297 +1316 @@
 
   typeset DEST RET dm_liste ifile target file_work
+  typeset NB_ESSAI DELAI status i
 
   if ( $DEBUG_sys ) ; then
     echo "IGCM_sys_Get :" $@
   fi
+
+  # number of tentative
+  NB_ESSAI=10
+  # time delay between tentative
+  DELAI=10
+
   if [ $DRYRUN -le 2 ]; then
     if [ X${1} = X'/l' ] ; then
@@ -1368 +1394 @@
       for target in ${dm_liste[*]} ; do
         local_file=$( basename ${target} )
-        \cp ${target} ${DEST}/${local_file} >> out_rsync 2>&1
-        (( RET = RET + $? ))
+        (( i = 0 ))
+        while [ $i -lt $NB_ESSAI ] ; do
+          \cp ${target} ${DEST}/${local_file} >> out_rsync 2>&1
+          status=$?
+          if [ ${status} -gt 0 ]; then
+            IGCM_debug_Print 2 "IGCM_sys_Get : cp failed ${i}/${NB_ESSAI}"
+            IGCM_debug_Print 2 "IGCM_sys_Get : sleep ${DELAI} seconds and try again."
+            sleep $DELAI
+          else
+            break
+          fi
+          (( i = i + 1 ))
+        done
+        (( RET = RET + status ))
       done
     else