def short_hostname (hostname):
return hostname.split('.')[0]
+
+####################
+# the place were other test instances tell about their not-yet-started
+# instances, that go undetected through sensing
+class Starting:
+
+ location='/root/starting'
+ def __init__ (self):
+ self.tuples=[]
+
+ def load (self):
+ try: self.tuples=[line.strip().split('@')
+ for line in file(Starting.location).readlines()]
+ except: self.tuples=[]
+
+ def vnames (self) :
+ self.load()
+ return [ x for (x,_) in self.tuples ]
+
+ def add (self, vname, bname):
+ if not vname in self.vnames():
+ file(Starting.location,'a').write("%s@%s\n"%(vname,bname))
+
+ def delete_vname (self, vname):
+ self.load()
+ if vname in self.vnames():
+ f=file(Starting.location,'w')
+ for (v,b) in self.tuples:
+ if v != vname: f.write("%s@%s\n"%(v,b))
+ f.close()
+
####################
# pool class
# allows to pick an available IP among a pool
class Pool:
- def __init__ (self, tuples,message):
+ def __init__ (self, tuples,message, substrate):
self.pool_items= [ PoolItem (hostname,userdata) for (hostname,userdata) in tuples ]
self.message=message
+ # where to send notifications upon load_starting
+ self.substrate=substrate
def list (self):
for i in self.pool_items: print i.line()
return (i.hostname,i.userdata)
return None
- # the place were other test instances tell about their not-yet-started
- # instances, that go undetected through sensing
- starting='/root/starting'
- def add_starting (self, name):
- try: items=[line.strip() for line in file(Pool.starting).readlines()]
- except: items=[]
- if not name in items:
- file(Pool.starting,'a').write(name+'\n')
+ ####################
+ # we have a starting instance of our own
+ def add_starting (self, vname, bname):
+ Starting().add(vname,bname)
for i in self.pool_items:
- if i.hostname==name: i.status='mine'
-
- # we load this after actual sensing;
+ if i.hostname==vname: i.status='mine'
+
+ # load the starting instances from the common file
+ # remember that might be ours
+ # return the list of (vname,bname) that are not ours
def load_starting (self):
- try: items=[line.strip() for line in file(Pool.starting).readlines()]
- except: items=[]
- for i in self.pool_items:
- if i.hostname in items:
- if i.status=='free' : i.status='starting'
+ starting=Starting()
+ starting.load()
+ new_tuples=[]
+ for (v,b) in starting.tuples:
+ for i in self.pool_items:
+ if i.hostname==v and i.status=='free':
+ i.status='starting'
+ new_tuples.append( (v,b,) )
+ return new_tuples
def release_my_starting (self):
for i in self.pool_items:
- if i.status=='mine':
- self.del_starting(i.hostname)
+ if i.status=='mine':
+ Starting().delete_vname (i.hostname)
i.status=None
- def del_starting (self, name):
- try: items=[line.strip() for line in file(Pool.starting).readlines()]
- except: items=[]
- if name in items:
- f=file(Pool.starting,'w')
- for item in items:
- if item != name: f.write(item+'\n')
- f.close()
-
+
##########
def _sense (self):
for item in self.pool_items:
print 'Sensing IP pool',self.message,
self._sense()
print 'Done'
- self.load_starting()
+ for (vname,bname) in self.load_starting():
+ self.substrate.add_starting_dummy (bname, vname)
print 'After starting: IP pool'
print self.line()
-
# OS-dependent ping option (support for macos, for convenience)
ping_timeout_option = None
# returns True when a given hostname/ip responds to ping
class Box:
def __init__ (self,hostname):
self.hostname=hostname
+ self._probed=None
def shortname (self):
return short_hostname(self.hostname)
def test_ssh (self): return TestSsh(self.hostname,username='root',unknown_host=False)
return result
def backquote (self, argv, trash_err=False):
+ # print 'running backquote',argv
if not trash_err:
result= subprocess.Popen(argv,stdout=subprocess.PIPE).communicate()[0]
else:
result= subprocess.Popen(argv,stdout=subprocess.PIPE,stderr=file('/dev/null','w')).communicate()[0]
return result
- def backquote_ssh (self, argv, trash_err=False):
+ def probe (self):
+ if self._probed is not None: return self._probed
# first probe the ssh link
probe_argv=self.test_ssh().actual_argv(['hostname'])
- hostname=self.backquote ( probe_argv, trash_err=True )
- if not hostname:
- print "root@%s unreachable"%self.hostname
- return ''
- else:
- return self.backquote( self.test_ssh().actual_argv(argv), trash_err)
+ self._probed=self.backquote ( probe_argv, trash_err=True )
+ if not self._probed: print "root@%s unreachable"%self.hostname
+ return self._probed
+
+ # use argv=['bash','-c',"the command line"]
+ # if you have any shell-expanded arguments like *
+ # and if there's any chance the command is adressed to the local host
+ def backquote_ssh (self, argv, trash_err=False):
+ if not self.probe(): return ''
+ return self.backquote( self.test_ssh().actual_argv(argv), trash_err)
############################################################
class BuildInstance:
# inspect box and find currently running builds
matcher=re.compile("\s*(?P<pid>[0-9]+).*-[bo]\s+(?P<buildname>[^\s]+)(\s|\Z)")
- def sense(self,options):
+ def sense(self, options):
print 'b',
self.sense_uptime()
pids=self.backquote_ssh(['pgrep','vbuild'],trash_err=True)
if not vserver_line: continue
context=vserver_line.split()[0]
if context=="CTX": continue
- longname=ctx_dict[context]
- self.add_vserver(longname,context)
-# print self.margin_outline(self.vplcname(longname)),"%(vserver_line)s [=%(longname)s]"%locals()
+ try:
+ longname=ctx_dict[context]
+ self.add_vserver(longname,context)
+ except:
+ print 'WARNING: found ctx %s in vserver_stat but was unable to figure a corresp. vserver'%context
# scan timestamps
running_vsnames = [ i.vservername for i in self.plc_instances ]
live_builds.append(buildname)
except: print 'WARNING, could not parse pid line',pid_line
# retrieve timestamps
+ if not live_builds: return
command= ['grep','.']
command += ['%s/*/timestamp'%b for b in live_builds]
command += ['/dev/null']
####################
class TestInstance:
- def __init__ (self, pid, buildname):
- self.pids=[pid]
+ def __init__ (self, buildname, pid=0):
+ self.pids=[]
+ if pid!=0: self.pid.append(pid)
self.buildname=buildname
# latest trace line
self.trace=''
# has a KO test
self.broken_steps=[]
+ self.timestamp = 0
+
+ def set_timestamp (self,timestamp): self.timestamp=timestamp
+ def set_now (self): self.timestamp=int(time.time())
+ def pretty_timestamp (self): return time.strftime("%Y-%m-%d:%H-%M",time.localtime(self.timestamp))
+
+
def add_pid (self,pid):
self.pids.append(pid)
def set_broken (self,plcindex, step):
self.broken_steps.append ( (plcindex, step,) )
def line (self):
- msg = " == %s =="%self.buildname
- msg += " (pid=%s)"%(self.pids)
+ double='=='
+ if self.pids: double='*'+double[1]
+ if self.broken_steps: double=double[0]+'B'
+ msg = " %s %s =="%(double,self.buildname)
+ if not self.pids: pass
+ elif len(self.pids)==1: msg += " (pid=%s)"%self.pids[0]
+ else: msg += " !!!pids=%s!!!"%self.pids
+ msg += " @%s"%self.pretty_timestamp()
if self.broken_steps:
- msg += "\n BROKEN IN STEPS "
- for (i,s) in self.broken_steps: msg += "step=%s,plc=%s"%(s,i)
+ msg += "\n BROKEN IN STEPS"
+ for (i,s) in self.broken_steps: msg += " %s@%s"%(s,i)
return msg
class TestBox (Box):
# can't reboot a vserver VM
self.run_ssh (['pkill','run_log'],"Terminating current runs",
dry_run=options.dry_run)
- self.run_ssh (['rm','-f','/root/starting'],"Cleaning /root/starting",
+ self.run_ssh (['rm','-f',Starting.location],"Cleaning %s"%Starting.location,
dry_run=options.dry_run)
def get_test (self, buildname):
for i in self.test_instances:
if i.buildname==buildname: return i
- def add_test (self, pid, buildname):
+ # we scan ALL remaining test results, even the ones not running
+ def add_timestamp (self, buildname, timestamp):
i=self.get_test(buildname)
- if i:
- print "WARNING: 2 concurrent tests run on same build %s"%buildname
- i.add_pid (pid)
+ if i:
+ i.set_timestamp(timestamp)
+ else:
+ i=TestInstance(buildname,0)
+ i.set_timestamp(timestamp)
+ self.test_instances.append(i)
+
+ def add_running_test (self, pid, buildname):
+ i=self.get_test(buildname)
+ if not i:
+ self.test_instances.append (TestInstance (buildname,pid))
return
- self.test_instances.append (TestInstance (pid,buildname))
+ if i.pids:
+ print "WARNING: 2 concurrent tests run on same build %s"%buildname
+ i.add_pid (pid)
+
+ def add_broken (self, buildname, plcindex, step):
+ i=self.get_test(buildname)
+ if not i:
+ i=TestInstance(buildname)
+ self.test_instances.append(i)
+ i.set_broken(plcindex, step)
matcher_proc=re.compile (".*/proc/(?P<pid>[0-9]+)/cwd.*/root/(?P<buildname>[^/]+)$")
- matcher_grep=re.compile ("/root/(?P<buildname>[^/]+)/logs/trace:TRACE:\s*(?P<plcindex>[0-9]+).*step=(?P<step>\S+).*")
+ matcher_grep=re.compile ("/root/(?P<buildname>[^/]+)/logs/trace
.*:TRACE:\s*(?P<plcindex>[0-9]+).*step=(?P<step>\S+).*")
def sense (self, options):
print 't',
self.sense_uptime()
- self.starting_ips=self.backquote_ssh(['cat','/root/starting']).strip().split('\n')
+ self.starting_ips=[x for x in self.backquote_ssh(['cat',Starting.location], trash_err=True).strip().split('\n') if x]
+
+ # scan timestamps on all tests
+ # this is likely to not invoke ssh so we need to be a bit smarter to get * expanded
+ # xxx would make sense above too
+ command=['bash','-c',"grep . /root/*/timestamp /dev/null"]
+ ts_lines=self.backquote_ssh(command,trash_err=True).split('\n')
+ for ts_line in ts_lines:
+ if not ts_line.strip(): continue
+ # expect /root/<buildname>/timestamp:<timestamp>
+ try:
+ (ts_file,timestamp)=ts_line.split(':')
+ ts_file=os.path.dirname(ts_file)
+ buildname=os.path.basename(ts_file)
+ timestamp=int(timestamp)
+ t=self.add_timestamp(buildname,timestamp)
+ except: print 'WARNING, could not parse ts line',ts_line
+
+ command=['bash','-c',"grep KO /root/*/logs/trace* /dev/null" ]
+ trace_lines=self.backquote_ssh (command).split('\n')
+ for line in trace_lines:
+ if not line.strip(): continue
+ m=TestBox.matcher_grep.match(line)
+ if m:
+ buildname=m.group('buildname')
+ plcindex=m.group('plcindex')
+ step=m.group('step')
+ self.add_broken(buildname,plcindex, step)
+ else: header("command %r returned line that failed to match\n%s"%(command,line))
+
pids = self.backquote_ssh (['pgrep','run_log'],trash_err=True)
if not pids: return
command=['ls','-ld'] + ["/proc/%s/cwd"%pid for pid in pids.split("\n") if pid]
if m:
pid=m.group('pid')
buildname=m.group('buildname')
- self.add_test(pid, buildname)
+ self.add_running_test(pid, buildname)
else: header("command %r returned line that failed to match\n%s"%(command,line))
- buildnames=[i.buildname for i in self.test_instances]
- if not buildnames: return
-
-# messy - tail has different output if one or several args
-# command=['tail','-n','1'] + [ "/root/%s/logs/trace"%b for b in buildnames ]
-# trace_lines=self.backquote_ssh (command).split('\n\n')
-# header('TAIL')
-# for line in trace_lines:
-# if not line.strip(): continue
-# print 'line [[[%s]]]'%line
- command=['grep','KO'] + [ "/root/%s/logs/trace"%b for b in buildnames ] + [ "/dev/null" ]
- trace_lines=self.backquote_ssh (command).split('\n')
- for line in trace_lines:
- if not line.strip(): continue
- m=TestBox.matcher_grep.match(line)
- if m:
- buildname=m.group('buildname')
- plcindex=m.group('plcindex')
- step=m.group('step')
- self.get_test(buildname).set_broken (plcindex, step)
- else: header("command %r returned line that failed to match\n%s"%(command,line))
-
def line (self):
return "%s (%s)"%(self.hostname,self.uptime())
def list (self):
- if not self.starting_ips:
- header ("No starting IP addresses on %s"%self.line())
- else:
- header ("IP addresses currently starting up on %s"%self.line())
- self.starting_ips.sort()
- for starting in self.starting_ips: print starting
if not self.test_instances:
- header ("No running tests on %s"%self.line())
+ header ("No known tests on %s"%self.line())
else:
- header ("Running tests on %s"%self.line())
+ header ("Known tests on %s"%self.line())
+ self.test_instances.sort(timestamp_sort)
for i in self.test_instances: print i.line()
+ if self.starting_ips:
+ header ("Starting IP addresses on %s"%self.line())
+ self.starting_ips.sort()
+ for starting in self.starting_ips: print starting
############################################################
class Options: pass
self.build_boxes = [ BuildBox(h) for h in self.build_boxes_spec() ]
self.plc_boxes = [ PlcBox (h,m) for (h,m) in self.plc_boxes_spec ()]
self.qemu_boxes = [ QemuBox (h,m) for (h,m) in self.qemu_boxes_spec ()]
- self.all_boxes = self.plc_boxes + self.qemu_boxes
+ self.default_boxes = self.plc_boxes + self.qemu_boxes
+ self.all_boxes = self.build_boxes + [ self.test_box ] + self.plc_boxes + self.qemu_boxes
self._sensed=False
- self.vplc_pool = Pool (self.vplc_ips(),"for vplcs")
- self.vnode_pool = Pool (self.vnode_ips(),"for vnodes")
+ self.vplc_pool = Pool (self.vplc_ips(),"for vplcs",self)
+ self.vnode_pool = Pool (self.vnode_ips(),"for vnodes",self)
def fqdn (self, hostname):
if hostname.find('.')<0: return "%s.%s"%(hostname,self.domain())
def sense (self,force=False):
if self._sensed and not force: return False
print 'Sensing local substrate...',
- for b in self.all_boxes: b.sense(self.options)
+ for b in self.default_boxes: b.sense(self.options)
print 'Done'
self._sensed=True
return True
+ def list (self):
+ for b in self.default_boxes:
+ b.list()
+
def add_dummy_plc (self, plc_boxname, plcname):
for pb in self.plc_boxes:
if pb.hostname==plc_boxname:
pb.add_dummy(plcname)
+ return True
def add_dummy_qemu (self, qemu_boxname, qemuname):
for qb in self.qemu_boxes:
if qb.hostname==qemu_boxname:
qb.add_dummy(qemuname)
+ return True
+
+ def add_starting_dummy (self, bname, vname):
+ return self.add_dummy_plc (bname, vname) or self.add_dummy_qemu (bname, vname)
##########
def provision (self,plcs,options):
plcs = [ self.provision_qemus (plc,options) for plc in plcs ]
# update the SFA spec accordingly
plcs = [ self.localize_sfa_rspec(plc,options) for plc in plcs ]
+ self.list()
return plcs
except Exception, e:
print '* Could not provision this test on current substrate','--',e,'--','exiting'
plc_boxname = options.ips_bplc.pop()
vplc_hostname=options.ips_vplc.pop()
else:
- if self.sense(): self.list_all()
+ if self.sense(): self.list()
plc_boxname=None
vplc_hostname=None
# try to find an available IP
#
self.add_dummy_plc(plc_boxname,plc['name'])
vplc_ip = self.vplc_pool.get_ip(vplc_hostname)
- self.vplc_pool.add_starting(vplc_hostname)
+ self.vplc_pool.add_starting(vplc_hostname, plc_boxname)
#### compute a helpful vserver name
# remove domain in hostname
qemu_boxname=options.ips_bnode.pop()
vnode_hostname=options.ips_vnode.pop()
else:
- if self.sense(): self.list_all()
+ if self.sense(): self.list()
qemu_boxname=None
vnode_hostname=None
# try to find an available IP
vnode_hostname=freed_vnode_hostname
self.vnode_pool.set_mine(vnode_hostname)
- self.add_dummy_qemu (qemu_boxname,nodename)
+ self.add_dummy_qemu (qemu_boxname,vnode_hostname)
mac=self.vnode_pool.retrieve_userdata(vnode_hostname)
ip=self.vnode_pool.get_ip (vnode_hostname)
- self.vnode_pool.add_starting(vnode_hostname)
+ self.vnode_pool.add_starting(vnode_hostname,qemu_boxname)
vnode_fqdn = self.fqdn(vnode_hostname)
nodemap={'host_box':qemu_boxname,
plc['sfa']['SFA_SM_HOST'] = plc['PLC_DB_HOST']
plc['sfa']['SFA_PLC_DB_HOST'] = plc['PLC_DB_HOST']
plc['sfa']['SFA_PLC_URL'] = 'https://' + plc['PLC_API_HOST'] + ':443/PLCAPI/'
- for site in plc['sites']:
- for node in site['nodes']:
- plc['sfa']['sfa_slice_rspec']['part4'] = node['node_fields']['hostname']
return plc
#################### release:
print "Could not find box %s"%boxname
return None
- def list_boxes(self,boxes):
+ def list_boxes(self,box_or_names):
print 'Sensing',
- for box in boxes:
- b=self.get_box(box)
- if not b: continue
- b.sense(self.options)
+ for box in box_or_names:
+ if not isinstance(box,Box): box=self.get_box(box)
+ if not box: continue
+ box.sense(self.options)
print 'Done'
- for box in boxes:
- b=self.get_box(box)
- if not b: continue
- b.list()
+ for box in box_or_names:
+ if not isinstance(box,Box): box=self.get_box(box)
+ if not box: continue
+ box.list()
- def reboot_boxes(self,boxes):
- for box in boxes:
- b=self.get_box(box)
- if not b: continue
- b.reboot(self.options)
+ def reboot_boxes(self,box_or_names):
+ for box in box_or_names:
+ if not isinstance(box,Box): box=self.get_box(box)
+ if not box: continue
+ box.reboot(self.options)
####################
# can be run as a utility to manage the local infrastructure
help='add plc boxes')
parser.add_option ('-q',"--qemu",action='store_true',dest='qemus',default=False,
help='add qemu boxes')
+ parser.add_option ('-a',"--all",action='store_true',dest='all',default=False,
+ help='address all known boxes, like -b -t -p -q')
parser.add_option ('-v',"--verbose",action='store_true',dest='verbose',default=False,
help='verbose mode')
parser.add_option ('-n',"--dry_run",action='store_true',dest='dry_run',default=False,
(self.options,args)=parser.parse_args()
boxes=args
- if self.options.testbox: boxes += [self.test_box.hostname]
- if self.options.builds: boxes += [b.hostname for b in self.build_boxes]
- if self.options.plcs: boxes += [b.hostname for b in self.plc_boxes]
- if self.options.qemus: boxes += [b.hostname for b in self.qemu_boxes]
- boxes=list(set(boxes))
+ if self.options.testbox: boxes += [self.test_box]
+ if self.options.builds: boxes += self.build_boxes
+ if self.options.plcs: boxes += self.plc_boxes
+ if self.options.qemus: boxes += self.qemu_boxes
+ if self.options.all: boxes += self.all_boxes
- # default scope
+ # default scope is -b -p -q
if not boxes:
- boxes = [ b.hostname for b in \
- self.build_boxes + [ self.test_box ] + \
- self.plc_boxes + self.qemu_boxes ]
+ boxes = self.build_boxes + self.plc_boxes + self.qemu_boxes
if self.options.reboot: self.reboot_boxes (boxes)
else: self.list_boxes (boxes)
git://git.onelab.eu / tests.git / blobdiff