#!/usr/bin/python
-import os.path
+import os.path, sys
import re
import subprocess
from optparse import OptionParser
class BuildBoxes:
# everything in the onelab.eu domain
- domain = 'onelab.eu'
+ domain = 'pl.sophia.inria.fr'
testmaster = 'testmaster'
build_boxes = [ "mirror", "liquid", "reed", "velvet", ]
plc_boxes = [ "testplc" ]
qemu_boxes = \
- [ "testqemu%d"%i for i in range (1,4) ] + \
- [ "testqemu32-%d"%i for i in range (1,6) ]
+ [ "qemu64-%d"%i for i in range (1,4) ] + \
+ [ "qemu32-%d"%i for i in range (1,6) ]
test_boxes = plc_boxes + qemu_boxes
+ testmaster_boxes = [ testmaster ]
def __init__ (self):
# dummy defaults
def fqdn (self, box):
return "%s.%s"%(box,self.domain)
+
+ ssh_command=['ssh','-o','ConnectTimeout=3']
@staticmethod
def root (box): return "root@%s"%box
+ @staticmethod
+ def ssh(box):
+ return BuildBoxes.ssh_command + [ BuildBoxes.root(box) ]
+
def header (self,message):
print "===============",message
+ sys.stdout.flush()
def run (self,argv,message, trash_err=False):
if self.options.dry_run:
print 'DRY_RUN:',
print " ".join(argv)
+ return 0
else:
if message: self.header(message)
if not trash_err:
- subprocess.call(argv)
+ return subprocess.call(argv)
else:
- subprocess.call(argv,stderr=file('/dev/null','w'))
+ return subprocess.call(argv,stderr=file('/dev/null','w'))
+ def run_ssh (self, box, argv, message, trash_err=False):
+ result=self.run (self.ssh(box) + argv, message, trash_err)
+ if result!=0:
+ print "WARNING: failed to run %s on %s"%(" ".join(argv),box)
+ return result
+
def backquote (self, argv, trash_err=False):
if not trash_err:
return subprocess.Popen(argv,stdout=subprocess.PIPE).communicate()[0]
else:
- null = open('/dev/null','w')
- result = subprocess.Popen(argv,stdout=subprocess.PIPE,stderr=null).communicate()[0]
- null.close()
- return result
+ return subprocess.Popen(argv,stdout=subprocess.PIPE,stderr=file('/dev/null','w')).communicate()[0]
+
+ def backquote_ssh (self, box, argv, trash_err=False):
+ # first probe the ssh link
+ hostname=self.backquote ( self.ssh(box) + [ "hostname"], trash_err=True )
+ if not hostname:
+ print "%s unreachable"%self.root(box)
+ return ''
+ else:
+ return self.backquote( ['ssh',self.root(box)] + argv, trash_err)
def reboot (self,box):
command=['ssh',self.root(box),'shutdown','-r','now']
self.run (command,"Rebooting %s"%box)
- def handle_tracker_file (self,filename):
+ def handle_tracker_plcs (self):
box = self.fqdn (self.testmaster)
- if self.options.probe:
- command=['ssh',self.root(box),"head","-v","--lines=100",filename]
- self.run(command,"++++++++++ Inspecting %s on %s"%(filename,box))
+ filename="tracker-plcs"
+ if not self.options.probe:
+ command=["rm","-rf",filename]
+ self.run_ssh(box,command,"Cleaning up %s on %s"%(filename,box))
else:
- command=['ssh',self.root(box),"rm","-rf",filename]
- self.run(command,"Cleaning up %s on %s"%(filename,box))
+ self.header ("++++++++++ Inspecting %s on %s"%(filename,box))
+ read_command = ["cat",filename]
+ trackers=self.backquote_ssh(box,read_command)
+ for tracker in trackers.split('\n'):
+ if not tracker: continue
+ try:
+ tracker=tracker.strip()
+ [hostname,buildname]=tracker.split('@')
+ [left,plcname]=buildname.rsplit('-',1)
+ print self.margin_outline(plcname),tracker
+ except:
+ print self.margin(""),tracker
- def handle_tracker_plcs (self):
- self.handle_tracker_file("tracker-plcs")
def handle_tracker_qemus (self):
- self.handle_tracker_file("tracker-qemus")
- def handle_trackers (self):
- self.handle_tracker_plcs()
- self.handle_tracker_qemus()
+ box = self.fqdn (self.testmaster)
+ filename="tracker-qemus"
+ if not self.options.probe:
+ command=["rm","-rf",filename]
+ self.run_ssh(box,command,"Cleaning up %s on %s"%(filename,box))
+ else:
+ self.header ("++++++++++ Inspecting %s on %s"%(filename,box))
+ read_command = ["cat",filename]
+ trackers=self.backquote_ssh(box,read_command)
+ for tracker in trackers.split('\n'):
+ if not tracker: continue
+ try:
+ tracker=tracker.strip()
+ [hostname,buildname,nodename]=tracker.split('@')
+ nodename=nodename.split('.')[0]
+ print self.margin_outline(nodename),tracker
+ except:
+ print self.margin(""),tracker
def handle_build_box (self,box):
if not self.options.probe:
self.reboot(box)
else:
- command=['ssh',self.root(box),'uptime']
- uptime=self.backquote(command,True).strip()
+ command=['uptime']
+ uptime=self.backquote_ssh(box,command,True).strip()
- command=['ssh',self.root(box),'pgrep','build']
+ command=['pgrep','build']
if self.options.dry_run:
- self.run(command,None)
+ self.run_ssh(box,command,None)
else:
- pids=self.backquote(command,True)
+ pids=self.backquote_ssh(box,command,True)
if not pids:
self.header ('No build process on %s (%s)'%(box,uptime))
else:
- command=['ssh',self.root(box),'ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
- self.run(command,"Active build processes on %s (%s)"%(box,uptime),True)
+ command=['ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
+ self.run_ssh(box,command,"Active build processes on %s (%s)"%(box,uptime),True)
vplc_matcher = re.compile(".*(vplc[0-9]+$)")
def vplcname (self, vservername):
if match: return match.groups(0)
else: return ""
+ margin_format="%-14s"
+ def margin(self,string): return self.margin_format%string
+ def outline (self, string): return '== %s =='%string
+ def margin_outline (self, string): return self.margin(self.outline(string))
+
def handle_plc_box (self,box):
if not self.options.probe:
self.reboot(box)
else:
- command=['ssh',self.root(box),'vserver-stat']
+ command=['vserver-stat']
if self.options.dry_run:
- self.run(command,"Active vservers on %s"%box)
+ self.run_ssh(box,command,"Active vservers on %s"%box)
else:
# try to find fullname (vserver_stat truncates to a ridiculously short name)
try:
self.header ("vserver map on %s"%box)
# fetch the contexts for all vservers on that box
- map_command=['ssh',self.root(box),'grep','.','/etc/vservers/*/context','/dev/null',]
- context_map=self.backquote (map_command)
+ map_command=['grep','.','/etc/vservers/*/context','/dev/null',]
+ context_map=self.backquote_ssh (box,map_command)
# at this point we have a set of lines like
# /etc/vservers/2010.01.20--k27-f12-32-vplc03/context:40144
ctx_dict={}
ctx_dict[xid]=os.path.basename(os.path.dirname(path))
# at this point ctx_id maps context id to vservername
- vserver_stat = self.backquote (command)
- format="%-14s"
+ vserver_stat = self.backquote_ssh (box,command)
for vserver_line in vserver_stat.split("\n"):
if not vserver_line: continue
context=vserver_line.split()[0]
if context=="CTX":
- print format%"",vserver_line
+ print self.margin(""),vserver_line
continue
longname=ctx_dict[context]
- plcname=self.vplcname(longname)
- header="== %s =="%plcname
- print format%header,"%(vserver_line)s [=%(longname)s]"%locals()
+ print self.margin_outline(self.vplcname(longname)),"%(vserver_line)s [=%(longname)s]"%locals()
except:
- self.run(command,"Fine-grained method failed - fallback to plain vserver-stat")
+ self.run_ssh(box,command,"Fine-grained method failed - fallback to plain vserver-stat")
vnode_matcher = re.compile(".*(vnode[0-9]+)")
def vnodename (self, ps_line):
if match: return match.groups(0)
else: return ""
-
def handle_qemu_box (self,box):
if not self.options.probe:
self.reboot(box)
else:
- command=['ssh',self.root(box),'pgrep','qemu']
+ command=['lsmod']
+ modules=self.backquote_ssh(box,command).split('\n')
+ kqemu_msg='*NO kqemu MODULE LOADED*'
+ for module in modules:
+ if module.find('kqemu')==0:
+ kqemu_msg='kqemu OK'
+
+ command=['pgrep','qemu']
if self.options.dry_run:
- self.run(command,None)
+ self.run_ssh(box,command,None)
else:
- pids=self.backquote(command)
+ pids=self.backquote_ssh(box,command)
if not pids:
self.header ('No qemu process on %s'%box)
else:
- format="%-14s"
- self.header ("Active qemu processes on %s"%box)
- command=['ssh',self.root(box),'ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
- ps_lines = self.backquote (command).split("\n")
+ self.header ("Active qemu processes on %s (%s)"%(box,kqemu_msg))
+ command=['ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
+ ps_lines = self.backquote_ssh (box,command).split("\n")
for ps_line in ps_lines:
if not ps_line or ps_line.find('PID') >=0 : continue
- header=self.vnodename(ps_line)
- print format%header, ps_line
+ print self.margin_outline(self.vnodename(ps_line)), ps_line
+
+ # the ouput of ps -o pid,command gives us <pid> bash <buildname>/run_log
+ def testmaster_buildname (self, ps_line):
+ chunks=ps_line.split()
+ path=chunks[2]
+ [buildname,command]=path.split('/')
+ return buildname
+
+ def handle_testmaster_box (self, box):
+ command=['pgrep','run_log']
+ if self.options.dry_run:
+ self.run_ssh(box,command,None)
+ else:
+ pids=self.backquote_ssh(box,command)
+ if not pids:
+ self.header ('No run_log process on %s'%box)
+ else:
+ self.header ("Active run_log processes on %s"%(box))
+ command=['ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
+ ps_lines = self.backquote_ssh (box,command).split("\n")
+ for ps_line in ps_lines:
+ if not ps_line or ps_line.find('PID') >=0 : continue
+ print self.margin_outline(self.testmaster_buildname(ps_line)), ps_line
+
def handle_box(self,box,type):
if box in self.qemu_boxes:
if type=="qemu": self.handle_qemu_box(self.fqdn(box))
elif box in self.plc_boxes:
if type=="plc": self.handle_plc_box(self.fqdn(box))
+ elif box in self.testmaster_boxes:
+ if type=='testmaster': self.handle_testmaster_box(self.fqdn(box))
elif type=="build":
self.handle_build_box(self.fqdn(box))
+ def handle_disk (self,box):
+ box=self.fqdn(box)
+ return self.run_ssh(box,["df","-h",],"Disk space on %s"%box)
+
def main (self):
usage="""%prog [options] [hostname..(s)]
-Default is to act on test boxes only (with trackers clean)"""
+Default is to act on test boxes only"""
parser = OptionParser (usage=usage)
parser.add_option ("-n","--dry-run",action="store_true",dest="dry_run",default=False,
help="Dry run")
help="Only acts on the qemu boxes")
parser.add_option ("-t","--trackers",action="store_true",dest="trackers_only",default=False,
help="Only wipes trackers")
+ parser.add_option ("-m","--master",action="store_true",dest="testmaster_only",default=False,
+ help="Display the testmaster status")
+ parser.add_option ("-d","--disk",action="store_true",dest="show_disk",default=False,
+ help="Only inspects disk status")
(self.options,args) = parser.parse_args()
self.boxes=self.plc_boxes
self.do_tracker_plcs = True
self.do_tracker_qemus = False
+ elif self.options.testmaster_only:
+ self.boxes=self.testmaster_boxes
+ self.do_tracker_plcs = False
+ self.do_tracker_qemus = False
elif self.options.trackers_only:
self.boxes = []
self.do_tracker_plcs = True
self.do_tracker_plcs = True
self.do_tracker_qemus = True
+ if self.options.show_disk:
+ for box in self.boxes: self.handle_disk(box)
+ return
+
# ALL OTHERS
for box in self.boxes: self.handle_box (box,"build")
+ # TESTMASTER
+ for box in self.boxes: self.handle_box (box,"testmaster")
# PLCS
if self.do_tracker_plcs:self.handle_tracker_plcs ()
for box in self.boxes: self.handle_box (box,"plc")