munet: 0.14.8: dont kill other munet procs by default

- add `--kill` option to cleanup existing processes using same
  rundir.
- detect if something is running using the same rundir and exit
  mentioning `--kill` option to cleanup

Signed-off-by: Christian Hopps <chopps@labn.net>

munet/__main__.py

@@ -19,6 +19,7 @@
 from .args import add_launch_args
 from .base import get_event_loop
 from .cleanup import cleanup_previous
+from .cleanup import is_running_in_rundir
 from .compat import PytestConfig
 
 
@@ -139,10 +140,11 @@ def main(*args):
     eap = ap.add_argument_group(title="Uncommon", description="uncommonly used options")
     eap.add_argument("--log-config", help="logging config file (yaml, toml, json, ...)")
     eap.add_argument(
-        "--no-kill",
+        "--kill",
         action="store_true",
-        help="Do not kill previous running processes",
+        help="Kill previous running processes using same rundir and exit",
     )
+    eap.add_argument("--no-kill", action="store_true", help=argparse.SUPPRESS)
     eap.add_argument(
         "--no-cli", action="store_true", help="Do not run the interactive CLI"
     )
@@ -157,7 +159,18 @@ def main(*args):
         sys.exit(0)
 
     rundir = args.rundir if args.rundir else "/tmp/munet"
+    rundir = os.path.abspath(rundir)
     args.rundir = rundir
+
+    if args.kill:
+        logging.info("Killing any previous run using rundir: {rundir}")
+        cleanup_previous(args.rundir)
+    elif is_running_in_rundir(args.rundir):
+        logging.fatal(
+            "Munet processes using rundir: %s, use `--kill` to cleanup first", rundir
+        )
+        return 1
+
     if args.cleanup:
         if os.path.exists(rundir):
             if not os.path.exists(f"{rundir}/config.json"):
@@ -169,6 +182,10 @@ def main(*args):
                 sys.exit(1)
             else:
                 subprocess.run(["/usr/bin/rm", "-rf", rundir], check=True)
+
+    if args.kill:
+        return 0
+
     subprocess.run(f"mkdir -p {rundir} && chmod 755 {rundir}", check=True, shell=True)
     os.environ["MUNET_RUNDIR"] = rundir
 
@@ -183,9 +200,6 @@ def main(*args):
         logger.critical("No nodes defined in config file")
         return 1
 
-    if not args.no_kill:
-        cleanup_previous()
-
     loop = None
     status = 4
     try:

munet/base.py

@@ -469,6 +469,8 @@ def _get_sub_args(self, cmd_list, defaults, use_pty=False, ns_only=False, **kwar
         env = {**(kwargs["env"] if "env" in kwargs else os.environ)}
         if "MUNET_NODENAME" not in env:
             env["MUNET_NODENAME"] = self.name
+        if "MUNET_PID" not in env and "MUNET_PID" in os.environ:
+            env["MUNET_PID"] = os.environ["MUNET_PID"]
         kwargs["env"] = env
 
         defaults.update(kwargs)
@@ -1226,7 +1228,13 @@ def run_in_window(
         if self.is_vm and self.use_ssh and not ns_only:  # pylint: disable=E1101
             if isinstance(cmd, str):
                 cmd = shlex.split(cmd)
-            cmd = ["/usr/bin/env", f"MUNET_NODENAME={self.name}"] + cmd
+            cmd = [
+                "/usr/bin/env",
+                f"MUNET_NODENAME={self.name}",
+            ]
+            if "MUNET_PID" in os.environ:
+                cmd.append(f"MUNET_PID={os.environ.get('MUNET_PID')}")
+            cmd += cmd
 
             # get the ssh cmd
             cmd = self._get_pre_cmd(False, True, ns_only=ns_only) + [shlex.join(cmd)]
@@ -1246,6 +1254,8 @@ def run_in_window(
             envvars = f"MUNET_NODENAME={self.name} NODENAME={self.name}"
             if hasattr(self, "rundir"):
                 envvars += f" RUNDIR={self.rundir}"
+            if "MUNET_PID" in os.environ:
+                envvars += f" MUNET_PID={os.environ.get('MUNET_PID')}"
             if hasattr(self.unet, "config_dirname") and self.unet.config_dirname:
                 envvars += f" CONFIGDIR={self.unet.config_dirname}"
             elif "CONFIGDIR" in os.environ:
@@ -2650,10 +2660,6 @@ def __init__(
 
         self.cfgopt = munet_config.ConfigOptionsProxy(pytestconfig)
 
-        super().__init__(
-            name, mount=True, net=isolated, uts=isolated, pid=pid, unet=None, **kwargs
-        )
-
         # This allows us to cleanup any leftover running munet's
         if "MUNET_PID" in os.environ:
             if os.environ["MUNET_PID"] != str(our_pid):
@@ -2664,6 +2670,10 @@ def __init__(
                 )
         os.environ["MUNET_PID"] = str(our_pid)
 
+        super().__init__(
+            name, mount=True, net=isolated, uts=isolated, pid=pid, unet=None, **kwargs
+        )
+
         # this is for testing purposes do not use
         if not BaseMunet.g_unet:
             BaseMunet.g_unet = self

munet/cleanup.py

@@ -59,25 +59,33 @@ def _get_our_pids():
     return {}
 
 
-def _get_other_pids():
-    piddict = get_pids_with_env("MUNET_PID")
-    unet_pids = {d["MUNET_PID"] for d in piddict.values()}
+def _get_other_pids(rundir):
+    if rundir:
+        # get only munet pids using the given rundir
+        piddict = get_pids_with_env("MUNET_RUNDIR", str(rundir))
+    else:
+        # Get all munet pids
+        piddict = get_pids_with_env("MUNET_PID")
+    unet_pids = {d["MUNET_PID"] for d in piddict.values() if "MUNET_PID" in d}
     pids_by_upid = {p: set() for p in unet_pids}
     for pid, envdict in piddict.items():
+        if "MUNET_PID" not in envdict:
+            continue
         unet_pid = envdict["MUNET_PID"]
         pids_by_upid[unet_pid].add(pid)
     # Filter out any child pid sets whos munet pid is still running
     return {x: y for x, y in pids_by_upid.items() if x not in y}
 
 
-def _get_pids_by_upid(ours):
+def _get_pids_by_upid(ours, rundir):
     if ours:
+        assert rundir is None
         return _get_our_pids()
-    return _get_other_pids()
+    return _get_other_pids(rundir)
 
 
-def _cleanup_pids(ours):
-    pids_by_upid = _get_pids_by_upid(ours).items()
+def _cleanup_pids(ours, rundir):
+    pids_by_upid = _get_pids_by_upid(ours, rundir).items()
     if not pids_by_upid:
         return
 
@@ -94,7 +102,7 @@ def _cleanup_pids(ours):
     #         return
     #     time.sleep(1)
 
-    pids_by_upid = _get_pids_by_upid(ours).items()
+    pids_by_upid = _get_pids_by_upid(ours, rundir).items()
     _kill_piddict(pids_by_upid, signal.SIGKILL)
 
 
@@ -103,12 +111,16 @@ def cleanup_current():
 
     Currently this only scans for old processes.
     """
-    _cleanup_pids(True)
+    _cleanup_pids(True, None)
 
 
-def cleanup_previous():
+def cleanup_previous(rundir=None):
     """Attempt to cleanup preview runs.
 
     Currently this only scans for old processes.
     """
-    _cleanup_pids(False)
+    _cleanup_pids(False, rundir)
+
+
+def is_running_in_rundir(rundir):
+    return bool(get_pids_with_env("MUNET_RUNDIR", str(rundir)))

munet/mutest/__main__.py

@@ -450,8 +450,9 @@ def main():
         sys.exit(0)
 
     rundir = args.rundir if args.rundir else "/tmp/mutest"
-    args.rundir = Path(rundir)
-    os.environ["MUNET_RUNDIR"] = rundir
+    rundir = Path(rundir).absolute()
+    args.rundir = rundir
+    os.environ["MUNET_RUNDIR"] = str(rundir)
     subprocess.run(f"mkdir -p {rundir} && chmod 755 {rundir}", check=True, shell=True)
 
     config = parser.setup_logging(args, config_base="logconf-mutest")

munet/testing/fixtures.py

@@ -25,7 +25,6 @@
 from ..base import Bridge
 from ..base import get_event_loop
 from ..cleanup import cleanup_current
-from ..cleanup import cleanup_previous
 from ..native import L3NodeMixin
 from ..parser import async_build_topology
 from ..parser import get_config
@@ -130,9 +129,12 @@ def session_autouse():
     else:
         is_worker = True
 
-    if not is_worker:
-        # This is unfriendly to multi-instance
-        cleanup_previous()
+    # We dont want to kill all munet and we don't have the rundir here yet
+    # This was more useful back when we used to leave processes around a lot
+    # more.
+    # if not is_worker:
+    #     # This is unfriendly to multi-instance
+    #     cleanup_previous()
 
     # We never pop as we want to keep logging
     _push_log_handler("session", "/tmp/unet-test/pytest-session.log")
@@ -150,8 +152,9 @@ def session_autouse():
 
 @pytest.fixture(autouse=True, scope="module")
 def module_autouse(request):
+    root_path = os.environ.get("MUNET_RUNDIR", "/tmp/unet-test")
     logpath = get_test_logdir(request.node.nodeid, True)
-    logpath = os.path.join("/tmp/unet-test", logpath, "pytest-exec.log")
+    logpath = os.path.join(root_path, logpath, "pytest-exec.log")
     with log_handler("module", logpath):
         sdir = os.path.dirname(os.path.realpath(request.fspath))
         with chdir(sdir, "module autouse fixture"):
@@ -174,7 +177,8 @@ def event_loop():
 
 @pytest.fixture(scope="module")
 def rundir_module():
-    d = os.path.join("/tmp/unet-test", get_test_logdir(module=True))
+    root_path = os.environ.get("MUNET_RUNDIR", "/tmp/unet-test")
+    d = os.path.join(root_path, get_test_logdir(module=True))
     logging.debug("conftest: test module rundir %s", d)
     return d
 
@@ -375,17 +379,17 @@ async def stepfunction(desc=""):
 
 @pytest.fixture(scope="function")
 def rundir():
-    d = os.path.join("/tmp/unet-test", get_test_logdir(module=False))
+    root_path = os.environ.get("MUNET_RUNDIR", "/tmp/unet-test")
+    d = os.path.join(root_path, get_test_logdir(module=False))
     logging.debug("conftest: test function rundir %s", d)
     return d
 
 
 # Configure logging
 @pytest.hookimpl(hookwrapper=True, tryfirst=True)
 def pytest_runtest_setup(item):
-    d = os.path.join(
-        "/tmp/unet-test", get_test_logdir(nodeid=item.nodeid, module=False)
-    )
+    root_path = os.environ.get("MUNET_RUNDIR", "/tmp/unet-test")
+    d = os.path.join(root_path, get_test_logdir(nodeid=item.nodeid, module=False))
     config = item.config
     logging_plugin = config.pluginmanager.get_plugin("logging-plugin")
     filename = Path(d, "pytest-exec.log")

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "munet"
-version = "0.14.7"
+version = "0.14.8"
 description = "A package to facilitate network simulations"
 authors = ["Christian Hopps <chopps@labn.net>"]
 license = "GPL-2.0-or-later"