fragattacks/tests/hwsim/vm/parallel-vm.py

244 lines
8.0 KiB
Python
Raw Normal View History

#!/usr/bin/env python2
#
# Parallel VM test case executor
# Copyright (c) 2014, Jouni Malinen <j@w1.fi>
#
# This software may be distributed under the terms of the BSD license.
# See README for more details.
import curses
import fcntl
import os
import subprocess
import sys
import time
def get_results():
global vm
started = []
passed = []
failed = []
skipped = []
for i in range(0, num_servers):
lines = vm[i]['out'].splitlines()
started += [ l for l in lines if l.startswith('START ') ]
passed += [ l for l in lines if l.startswith('PASS ') ]
failed += [ l for l in lines if l.startswith('FAIL ') ]
skipped += [ l for l in lines if l.startswith('SKIP ') ]
return (started, passed, failed, skipped)
def show_progress(scr):
global num_servers
global vm
global dir
global timestamp
global tests
total_tests = len(tests)
scr.leaveok(1)
scr.addstr(0, 0, "Parallel test execution status", curses.A_BOLD)
for i in range(0, num_servers):
scr.addstr(i + 1, 0, "VM %d:" % (i + 1), curses.A_BOLD)
scr.addstr(i + 1, 10, "starting VM")
scr.addstr(num_servers + 1, 0, "Total:", curses.A_BOLD)
scr.addstr(num_servers + 1, 20, "TOTAL={} STARTED=0 PASS=0 FAIL=0 SKIP=0".format(total_tests))
scr.refresh()
while True:
running = False
updated = False
for i in range(0, num_servers):
if not vm[i]['proc']:
continue
if vm[i]['proc'].poll() is not None:
vm[i]['proc'] = None
scr.move(i + 1, 10)
scr.clrtoeol()
log = '{}/{}.srv.{}/console'.format(dir, timestamp, i + 1)
with open(log, 'r') as f:
if "Kernel panic" in f.read():
scr.addstr("kernel panic")
else:
scr.addstr("completed run")
updated = True
continue
running = True
try:
err = vm[i]['proc'].stderr.read()
vm[i]['err'] += err
except:
pass
try:
out = vm[i]['proc'].stdout.read()
if "READY" in out or "PASS" in out or "FAIL" in out or "SKIP" in out:
if not tests:
vm[i]['proc'].stdin.write('\n')
else:
name = tests.pop(0)
vm[i]['proc'].stdin.write(name + '\n')
except:
continue
#print("VM {}: '{}'".format(i, out))
vm[i]['out'] += out
lines = vm[i]['out'].splitlines()
last = [ l for l in lines if l.startswith('START ') ]
if len(last) > 0:
try:
info = last[-1].split(' ')
scr.move(i + 1, 10)
scr.clrtoeol()
scr.addstr(info[1])
updated = True
except:
pass
if not running:
break
if updated:
(started, passed, failed, skipped) = get_results()
scr.move(num_servers + 1, 10)
scr.clrtoeol()
scr.addstr("{} %".format(int(100.0 * (len(passed) + len(failed) + len(skipped)) / total_tests)))
scr.addstr(num_servers + 1, 20, "TOTAL={} STARTED={} PASS={} FAIL={} SKIP={}".format(total_tests, len(started), len(passed), len(failed), len(skipped)))
if len(failed) > 0:
scr.move(num_servers + 2, 0)
scr.clrtoeol()
scr.addstr("Failed test cases: ")
for f in failed:
scr.addstr(f.split(' ')[1])
scr.addstr(' ')
scr.refresh()
time.sleep(0.5)
scr.refresh()
time.sleep(0.3)
def main():
global num_servers
global vm
global dir
global timestamp
global tests
if len(sys.argv) < 2:
sys.exit("Usage: %s <number of VMs> [--codecov] [params..]" % sys.argv[0])
num_servers = int(sys.argv[1])
if num_servers < 1:
sys.exit("Too small number of VMs")
timestamp = int(time.time())
if len(sys.argv) > 2 and sys.argv[2] == "--codecov":
idx = 3
print "Code coverage - build separate binaries"
logdir = "/tmp/hwsim-test-logs/" + str(timestamp)
os.makedirs(logdir)
subprocess.check_call(['./build-codecov.sh', logdir])
codecov_args = ['--codecov_dir', logdir]
codecov = True
else:
idx = 2
codecov_args = []
codecov = False
tests = []
cmd = [ '../run-tests.py', '-L' ] + sys.argv[idx:]
lst = subprocess.Popen(cmd, stdout=subprocess.PIPE)
for l in lst.stdout.readlines():
name = l.split(' ')[0]
tests.append(name)
if len(tests) == 0:
sys.exit("No test cases selected")
if '-f' in sys.argv[idx:]:
extra_args = sys.argv[idx:]
else:
extra_args = [x for x in sys.argv[idx:] if x not in tests]
dir = '/tmp/hwsim-test-logs'
try:
os.mkdir(dir)
except:
pass
if num_servers > 2 and len(tests) > 100:
# Move test cases with long duration to the beginning as an
# optimization to avoid last part of the test execution running a long
# duration test case on a single VM while all other VMs have already
# completed their work.
long = [ "ap_wps_pbc_timeout",
"grpform_cred_ready_timeout",
"grpform_cred_ready_timeout2",
"discovery_pd_retries",
"ibss_wpa_none",
"concurrent_p2pcli",
"wpas_ap_wps",
"ibss_rsn",
"wext_pmksa_cache",
"ap_ht_40mhz_intolerant_ap",
"ap_wps_setup_locked_timeout",
"ap_vht160",
"dfs_radar",
"dfs" ]
for l in long:
if l in tests:
tests.remove(l)
tests.insert(0, l)
vm = {}
for i in range(0, num_servers):
print("\rStarting virtual machine {}/{}".format(i + 1, num_servers)),
cmd = ['./vm-run.sh', '--timestamp', str(timestamp),
'--ext', 'srv.%d' % (i + 1),
'-i'] + codecov_args + extra_args
vm[i] = {}
vm[i]['proc'] = subprocess.Popen(cmd,
stdin=subprocess.PIPE,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE)
vm[i]['out'] = ""
vm[i]['err'] = ""
for stream in [ vm[i]['proc'].stdout, vm[i]['proc'].stderr ]:
fd = stream.fileno()
fl = fcntl.fcntl(fd, fcntl.F_GETFL)
fcntl.fcntl(fd, fcntl.F_SETFL, fl | os.O_NONBLOCK)
print
curses.wrapper(show_progress)
with open('{}/{}-parallel.log'.format(dir, timestamp), 'w') as f:
for i in range(0, num_servers):
f.write('VM {}\n{}\n{}\n'.format(i, vm[i]['out'], vm[i]['err']))
(started, passed, failed, skipped) = get_results()
if len(failed) > 0:
print "Failed test cases:"
for f in failed:
print f.split(' ')[1],
print
print("TOTAL={} PASS={} FAIL={} SKIP={}".format(len(started), len(passed), len(failed), len(skipped)))
print "Logs: " + dir + '/' + str(timestamp)
for i in range(0, num_servers):
log = '{}/{}.srv.{}/console'.format(dir, timestamp, i + 1)
with open(log, 'r') as f:
if "Kernel panic" in f.read():
print "Kernel panic in " + log
if codecov:
print "Code coverage - preparing report"
for i in range(num_servers):
subprocess.check_call(['./process-codecov.sh',
logdir + ".srv.%d" % (i + 1),
str(i)])
subprocess.check_call(['./combine-codecov.sh', logdir])
print "file://%s/index.html" % logdir
if __name__ == "__main__":
main()