| 1 |
from PostMortem import PostMortem
|
| 2 |
|
| 3 |
from crab_util import *
|
| 4 |
import common
|
| 5 |
import string, os
|
| 6 |
|
| 7 |
from ProdCommon.Storage.SEAPI.SElement import SElement
|
| 8 |
from ProdCommon.Storage.SEAPI.SBinterface import SBinterface
|
| 9 |
|
| 10 |
class PostMortemServer(PostMortem):
|
| 11 |
def __init__(self, cfg_params, nj_list):
|
| 12 |
|
| 13 |
PostMortem.__init__(self, cfg_params, nj_list)
|
| 14 |
|
| 15 |
# init client server params...
|
| 16 |
CliServerParams(self)
|
| 17 |
|
| 18 |
if self.storage_path[0]!='/':
|
| 19 |
self.storage_path = '/'+self.storage_path
|
| 20 |
|
| 21 |
return
|
| 22 |
|
| 23 |
def collectLogging(self):
|
| 24 |
# get updated status from server
|
| 25 |
try:
|
| 26 |
from StatusServer import StatusServer
|
| 27 |
stat = StatusServer(self.cfg_params)
|
| 28 |
stat.resynchClientSide()
|
| 29 |
except:
|
| 30 |
pass
|
| 31 |
|
| 32 |
#create once storage interaction object
|
| 33 |
seEl = None
|
| 34 |
loc = None
|
| 35 |
try:
|
| 36 |
seEl = SElement(self.storage_name, self.storage_proto, self.storage_port)
|
| 37 |
except Exception, ex:
|
| 38 |
common.logger.debug(1, str(ex))
|
| 39 |
msg = "ERROR: Unable to create SE source interface \n"
|
| 40 |
raise CrabException(msg)
|
| 41 |
try:
|
| 42 |
loc = SElement("localhost", "local")
|
| 43 |
except Exception, ex:
|
| 44 |
common.logger.debug(1, str(ex))
|
| 45 |
msg = "ERROR: Unable to create destination interface \n"
|
| 46 |
raise CrabException(msg)
|
| 47 |
|
| 48 |
## coupling se interfaces
|
| 49 |
sbi = SBinterface( seEl, loc )
|
| 50 |
|
| 51 |
## get the list of jobs to get logging.info skimmed by failed status
|
| 52 |
logginable = self.skimDeadList()
|
| 53 |
|
| 54 |
## iter over each asked job and print warning if not in skimmed list
|
| 55 |
for id in self.nj_list:
|
| 56 |
if id not in self.all_jobs:
|
| 57 |
common.logger.message('Warning: job # ' + str(id) + ' does not exist! Not possible to ask for postMortem ')
|
| 58 |
continue
|
| 59 |
elif id in logginable:
|
| 60 |
fname = self.fname_base + str(id) + '.LoggingInfo'
|
| 61 |
if os.path.exists(fname):
|
| 62 |
common.logger.message('Logging info for job ' + str(id) + ' already present in '+fname+'\nRemove it for update')
|
| 63 |
continue
|
| 64 |
## retrieving & processing logging info
|
| 65 |
if self.retrieveFile( sbi, id, fname):
|
| 66 |
## decode logging info
|
| 67 |
fl = open(fname, 'r')
|
| 68 |
out = "".join(fl.readlines())
|
| 69 |
fl.close()
|
| 70 |
reason = self.decodeLogging(out)
|
| 71 |
common.logger.message('Logging info for job '+ str(id) +': '+str(reason)+'\n written to '+str(fname)+' \n' )
|
| 72 |
else:
|
| 73 |
common.logger.message('Logging info for job '+ str(id) +' not retrieved')
|
| 74 |
else:
|
| 75 |
common.logger.message('Warning: job # ' + str(id) + ' not killed or aborted! Will get loggingInfo manually ')
|
| 76 |
PostMortem.collectLogging(self)
|
| 77 |
return
|
| 78 |
|
| 79 |
|
| 80 |
def skimDeadList(self):
|
| 81 |
"""
|
| 82 |
__skimDeadList__
|
| 83 |
return the list of jobs really failed: K, A
|
| 84 |
"""
|
| 85 |
skimmedlist = []
|
| 86 |
self.up_task = common._db.getTask( self.nj_list )
|
| 87 |
for job in self.up_task.jobs:
|
| 88 |
if job.runningJob['status'] in ['K','A']:
|
| 89 |
skimmedlist.append(job['jobId'])
|
| 90 |
return skimmedlist
|
| 91 |
|
| 92 |
def retrieveFile(self, sbi, jobid, destlog):
|
| 93 |
"""
|
| 94 |
__retrieveFile__
|
| 95 |
|
| 96 |
retrieves logging.info file from the server storage area
|
| 97 |
"""
|
| 98 |
self.taskuuid = str(common._db.queryTask('name'))
|
| 99 |
common.logger.debug(3, "Task name: " + self.taskuuid)
|
| 100 |
|
| 101 |
# full remote dir
|
| 102 |
remotedir = os.path.join(self.storage_path, self.taskuuid)
|
| 103 |
remotelog = remotedir + '/loggingInfo_'+str(jobid)+'.log'
|
| 104 |
|
| 105 |
common.logger.message("Starting retrieving logging-info from server " \
|
| 106 |
+ str(self.storage_name) + " for job " \
|
| 107 |
+ str(jobid) + "...")
|
| 108 |
|
| 109 |
# retrieve logging info from storage
|
| 110 |
common.logger.debug(1, "retrieving "+ str(remotelog) +" to "+ str(destlog) )
|
| 111 |
try:
|
| 112 |
sbi.copy( remotelog, destlog)
|
| 113 |
except Exception, ex:
|
| 114 |
msg = "WARNING: Unable to retrieve logging-info file %s \n"%remotelog
|
| 115 |
msg += str(ex)
|
| 116 |
common.logger.debug(1,msg)
|
| 117 |
return False
|
| 118 |
# cleaning remote logging info file
|
| 119 |
try:
|
| 120 |
common.logger.debug(5, "Cleaning remote file [%s] " + str(remotelog) )
|
| 121 |
sbi.delete(remotelog)
|
| 122 |
except Exception, ex:
|
| 123 |
msg = "WARNING: Unable to clean remote logging-info file %s \n"%remotelog
|
| 124 |
msg += str(ex)
|
| 125 |
common.logger.debug(5,msg)
|
| 126 |
return True
|
