#! /usr/bin/env python3
-# The goal is to introduce random failures in a simulation, to test simgrid under extreme conditions.
+# The goal is to introduce random failures in a simulation, to test SimGrid under extreme conditions.
#
# It is made of several components.
#
# Kill the link #0 after 42 seconds (using a kernel::Timer)
#
# * a python script: tools/simgrid-monkey (this file)
-# * It takes a regular simgrid simulation as a parameter, use the cmonkey plugin to get the information about it,
+# * It takes a regular SimGrid simulation as a parameter, use the cmonkey plugin to get the information about it,
# and then restart many runs, with one resource being turn_off() + turn_on() in each run.
# * Each resource gets killed between each timestamps, and on each timestamp.
# * So the amount of simulations is: 1 + (host_c+link_c) * timestamps * 2
#
# * Test program, written to resist these extreme conditions:
-# * teshsuite/s4u/monkey-masterworkers: tests synchronous comms and execs
+# * teshsuite/s4u/monkey-masterworkers: tests synchronous comms and execs (C++ and python)
+# * teshsuite/s4u/monkey-semaphore: tests async semaphores (C++ only)
-import multiprocessing as mp
import sys
import os
import argparse
#print(f"hosts:{host_count} links:{link_count} timestamps:{' '.join(([str(i) for i in timestamps]))}")
return (host_count, link_count, timestamps)
-parser = argparse.ArgumentParser(description='Run a simgrid simulation, and turn off/on resources at random.')
+parser = argparse.ArgumentParser(description='Run a SimGrid simulation, and turn off/on resources at random.')
parser.add_argument('--valgrind', help="Run the simulations in valgrind")
parser.add_argument('command', nargs='*')
args = parser.parse_args()
(host_count, link_count, timestamps) = get_info(args.command)
+timestamps = sorted([*{*timestamps}]) # kill duplicates
print(f"Monkey informations: hosts:{host_count} links:{link_count} timestamps:{' '.join(([str(i) for i in timestamps]))}")
-def do_run(cmd, extra_params):
+error_count = 0
+test_count = 0
+def do_run(cmd, extra_params, test_todo):
+ global test_count, error_count
+ test_count = test_count + 1
cmd = copy.deepcopy(cmd)
cmd.append("--cfg=plugin:cmonkey")
for p in extra_params:
cmd.append(p)
print(f"Start {' '.join(cmd)}")
+ sys.stdout.flush()
+
run = subprocess.run(cmd, shell=False, cwd=os.getcwd(), stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
+ out = str(run.stdout, errors='replace')
if run.returncode != 0:
msg = f"ERROR (retcode: {run.returncode}). Output:\n"
- msg += str(run.stdout, errors='replace')
+ msg += out
print(msg)
- os.exit(1)
- print ("Success.\n")
+ sys.exit(1)
+ for line in out.split("\n"):
+ if re.match("==.* in use at exit: ", line) and not re.match("==.* in use at exit: 0 bytes in 0 blocks", line):
+ m = re.match("==.* in use at exit: (.*)", line)
+ print(f"LEAK SUMMARY: {m.group(1)} in use at exit")
+ error_count += 1
+
+ if re.match("==.* ERROR SUMMARY: ", line):
+ m = re.match("==.* ERROR SUMMARY: (.*)", line)
+ print(f"valgrind summary: {m.group(1)}")
+ if not re.match("==.* 0 errors from 0 contexts", line):
+ error_count += 1
+ print (f"Test {test_count} out of {test_todo} succeded.\n")
+
def doit():
- prev = 0
+ prev_time = 0
+ test_todo = 2 * len(timestamps) * (host_count + link_count)
for pos in range(len(timestamps)):
now = timestamps[pos]
for host in range(host_count):
- do_run(args.command, [f"--cfg=cmonkey/time:{(now-prev)/2}", f"--cfg=cmonkey/host:{host}"])
- do_run(args.command, [f"--cfg=cmonkey/time:{now}", f"--cfg=cmonkey/host:{host}"])
+ do_run(args.command, [f"--cfg=cmonkey/time:{(now-prev_time)/2}", f"--cfg=cmonkey/host:{host}"], test_todo)
+ for link in range(link_count):
+ do_run(args.command, [f"--cfg=cmonkey/time:{(now-prev_time)/2}", f"--cfg=cmonkey/link:{link}"], test_todo)
+ for host in range(host_count):
+ do_run(args.command, [f"--cfg=cmonkey/time:{now}", f"--cfg=cmonkey/host:{host}"], test_todo)
for link in range(link_count):
- do_run(args.command, [f"--cfg=cmonkey/time:{(now-prev)/2}", f"--cfg=cmonkey/link:{link}"])
- do_run(args.command, [f"--cfg=cmonkey/time:{now}", f"--cfg=cmonkey/link:{link}"])
+ do_run(args.command, [f"--cfg=cmonkey/time:{now}", f"--cfg=cmonkey/link:{link}"], test_todo)
doit()
+
+print(f"In total, the monkey found {error_count} errors.")
+sys.exit(error_count)
