Add nilmrun-ps, -kill, and -run commands

README.txt

@@ -7,7 +7,7 @@ Prerequisites:
   sudo apt-get install python2.7 python-setuptools
 
   # Plus nilmdb and its dependencies
-  nilmdb (1.8.2+)
+  nilmdb (1.9.5+)
 
 Install:
 

nilmrun/processmanager.py

@@ -92,6 +92,12 @@ class Process(object):
                 except OSError: # pragma: no cover
                     return None
 
+            def kill(pid, sig):
+                try:
+                    return os.kill(pid, sig)
+                except OSError: # pragma: no cover
+                    return
+
             # Find all children
             group = getpgid(self._process.pid)
             main = psutil.Process(self._process.pid)
@@ -100,7 +106,7 @@ class Process(object):
             # Kill with SIGTERM, if they're still in this process group
             for proc in allproc:
                 if getpgid(proc.pid) == group:
-                    os.kill(proc.pid, signal.SIGTERM)
+                    kill(proc.pid, signal.SIGTERM)
 
             # Wait for it to die again
             if self._join(timeout):
@@ -109,7 +115,7 @@ class Process(object):
             # One more try with SIGKILL
             for proc in allproc:
                 if getpgid(proc.pid) == group:
-                    os.kill(proc.pid, signal.SIGKILL)
+                    kill(proc.pid, signal.SIGKILL)
 
             # See if it worked
             return self._join(timeout)
@@ -178,9 +184,27 @@ class ProcessManager(object):
     """Track and manage a collection of Process objects"""
     def __init__(self):
         self.processes = {}
-        self.tmpfiles = {}
+        self.tmpdirs = {}
-        self.tmpdir = tempfile.mkdtemp(prefix = "nilmrun-usercode-")
+        atexit.register(self._atexit)
-        atexit.register(shutil.rmtree, self.tmpdir)
+
+    def _cleanup_tmpdir(self, pid):
+        if pid in self.tmpdirs:
+            try:
+                shutil.rmtree(self.tmpdirs[pid])
+            except OSError: # pragma: no cover
+                pass
+            del self.tmpdirs[pid]
+
+    def _atexit(self):
+        # Kill remaining processes, remove their dirs
+        for pid in self.processes.keys():
+            try:
+                self.processes[pid].terminate()
+                del self.processes[pid]
+                shutil.rmtree(self.tmpdirs[pid])
+                del self.tmpdirs[pid]
+            except Exception: # pragma: no cover
+                pass
 
     def __iter__(self):
         return iter(self.processes.keys())
@@ -193,15 +217,33 @@ class ProcessManager(object):
         executed.  The arguments, which must be strings, will be
         accessible in the code as sys.argv[1:]."""
         # The easiest way to do this, by far, is to just write the
-        # code to a file.
+        # code to a file.  Make a directory to put it in.
-        (fd, path) = tempfile.mkstemp(prefix = "nilmrun-usercode-",
+        tmpdir = tempfile.mkdtemp(prefix = "nilmrun-usercode-")
-                                      suffix = ".py", dir=self.tmpdir)
+        try:
-        with os.fdopen(fd, 'w') as f:
+            # Write the code
-            f.write(code)
+            codepath = os.path.join(tmpdir, "usercode.py")
-        argv = [ sys.executable, "-B", "-s", "-u", path ] + args
+            with open(codepath, "w") as f:
-        pid = self.run_command(argv)
+                f.write(code)
-        self.tmpfiles[pid] = path
+            # Save the args too, for debugging purposes
-        return pid
+            with open(os.path.join(tmpdir, "args.txt"), "w") as f:
+                f.write(repr(args))
+
+            # Run the code
+            argv = [ sys.executable, "-B", "-s", "-u", codepath ] + args
+            pid = self.run_command(argv)
+
+            # Save the temp dir
+            self.tmpdirs[pid] = tmpdir
+            tmpdir = None  # Don't need to remove it anymore
+
+            return pid
+        finally:
+            # Clean up tempdir if we didn't finish
+            if tmpdir is not None:
+                try:
+                    shutil.rmtree(tmpdir)
+                except OSError: # pragma: no cover
+                    pass
 
     def run_command(self, argv):
         """Execute a command line program"""
@@ -213,12 +255,7 @@ class ProcessManager(object):
         return self.processes[pid].terminate()
 
     def remove(self, pid):
-        if pid in self.tmpfiles:
+        self._cleanup_tmpdir(pid)
-            try:
-                os.unlink(self.tmpfiles[pid])
-            except OSError: # pragma: no cover
-                pass
-            del self.tmpfiles[pid]
         del self.processes[pid]
 
     def get_info(self):

nilmrun/server.py

@@ -22,6 +22,7 @@ from nilmdb.server.serverutil import (
         cherrypy_stop,
         bool_param,
         )
+from nilmdb.utils import serializer_proxy
 import nilmrun
 import nilmrun.testfilter
 
@@ -204,8 +205,12 @@ class Server(object):
         # error messages.
         cherrypy._cperror._ie_friendly_error_sizes = {}
 
+        # The manager maintains internal state and isn't necessarily
+        # thread-safe, so wrap it in the serializer.
+        manager = serializer_proxy(nilmrun.processmanager.ProcessManager)()
+
         # Build up the application and mount it
-        manager = nilmrun.processmanager.ProcessManager()
+        self._manager = manager
         root = App()
         root.process = AppProcess(manager)
         root.run = AppRun(manager)

scripts/kill.py

@@ -0,0 +1,51 @@
+#!/usr/bin/python
+
+from nilmdb.client.httpclient import HTTPClient, ClientError, ServerError
+from nilmdb.utils.printf import *
+import nilmrun
+
+import argparse
+import os
+import sys
+
+def main():
+    """Kill/remove a process from the NilmRun server"""
+    def_url = os.environ.get("NILMRUN_URL", "http://localhost/nilmrun/")
+    parser = argparse.ArgumentParser(
+        description = 'Kill/remove a process from the NilmRun server',
+        formatter_class = argparse.ArgumentDefaultsHelpFormatter)
+    parser.add_argument("-V", "--version", action="version",
+                        version = nilmrun.__version__)
+    group = parser.add_argument_group("Standard options")
+    group.add_argument('-u', '--url',
+                       help = 'NilmRun server URL', default = def_url)
+    group.add_argument('-n', '--noverify', action="store_true",
+                       help = 'Disable SSL certificate verification')
+    group = parser.add_argument_group("Program")
+    group.add_argument('-q', '--quiet', action="store_true",
+                       help = "Don't print out the final log contents")
+    group.add_argument('pid', nargs='+', help="PIDs to kill")
+    args = parser.parse_args()
+
+    client = HTTPClient(baseurl = args.url, verify_ssl = not args.noverify)
+
+    # Kill or remove process
+    all_failed = True
+    for pid in args.pid:
+        try:
+            s = client.post("process/remove", { "pid": pid })
+            if not args.quiet:
+                sys.stdout.write(s['log'])
+            all_failed = False
+        except ClientError as e:
+            if "404" in e.status:
+                fprintf(sys.stderr, "no such pid: %s\n", pid)
+            else:
+                raise
+
+    # Return error if we failed to remove any of them
+    if all_failed:
+        raise SystemExit(1)
+
+if __name__ == "__main__":
+    main()

scripts/ps.py

@@ -0,0 +1,66 @@
+#!/usr/bin/python
+
+from nilmdb.client.httpclient import HTTPClient, ClientError, ServerError
+from nilmdb.utils.printf import *
+from nilmdb.utils import datetime_tz
+import nilmrun
+
+import argparse
+import os
+
+def main():
+    """List NilmRun processes"""
+    def_url = os.environ.get("NILMRUN_URL", "http://localhost/nilmrun/")
+    parser = argparse.ArgumentParser(
+        description = 'List NilmRun processes',
+        formatter_class = argparse.ArgumentDefaultsHelpFormatter)
+    parser.add_argument("-V", "--version", action="version",
+                        version = nilmrun.__version__)
+    group = parser.add_argument_group("Standard options")
+    group.add_argument('-u', '--url',
+                       help = 'NilmRun server URL', default = def_url)
+    group.add_argument('-n', '--noverify', action="store_true",
+                       help = 'Disable SSL certificate verification')
+    args = parser.parse_args()
+
+    client = HTTPClient(baseurl = args.url, verify_ssl = not args.noverify)
+    # Print overall system info
+    info = client.get("process/info")
+    total = info['total']
+    system = info['system']
+    printf(" procs: %d nilm, %d other\n", info['total']['procs'],
+           info['system']['procs'] - info['total']['procs'])
+    printf("   cpu: %d%% nilm, %d%% other, %d%% max\n",
+           round(info['total']['cpu_percent']),
+           round(info['system']['cpu_percent'] - info['total']['cpu_percent']),
+           round(info['system']['cpu_max']))
+    printf("   mem: %d MiB used, %d MiB total, %d%%\n",
+           round(info['system']['mem_used'] / 1048576.0),
+           round(info['system']['mem_total'] / 1048576.0),
+           round(info['system']['mem_used'] * 100.0
+                 / info['system']['mem_total']))
+
+    if len(info['pids']) == 0:
+        printf("No running processes\n")
+        raise SystemExit(0)
+
+    # Print process detail for each managed process
+    fmt = "%-36s %-6s %-15s %-4s %-3s %-5s\n"
+    printf(fmt, "PID", "STATE", "SINCE", "PROC", "CPU", "LOG")
+    for pid in sorted(info['pids'].keys()):
+        pidinfo = client.get("process/status", { "pid": pid })
+        if pidinfo['alive']:
+            status = "alive"
+        else:
+            if pidinfo['exitcode']:
+                status = "error"
+            else:
+                status = "done"
+        dt = datetime_tz.datetime_tz.fromtimestamp(pidinfo['start_time'])
+        since = dt.strftime("%m/%d-%H:%M:%S")
+        printf(fmt, pid, status, since, info['pids'][pid]['procs'],
+               str(int(round(info['pids'][pid]['cpu_percent']))),
+               len(pidinfo['log']))
+
+if __name__ == "__main__":
+    main()

scripts/run.py

@@ -0,0 +1,60 @@
+#!/usr/bin/python
+
+from nilmdb.client.httpclient import HTTPClient, ClientError, ServerError
+from nilmdb.utils.printf import *
+import nilmrun
+
+import argparse
+import os
+import time
+import sys
+
+def main():
+    """Run a command on the NilmRun server"""
+    def_url = os.environ.get("NILMRUN_URL", "http://localhost/nilmrun/")
+    parser = argparse.ArgumentParser(
+        description = 'Run a command on the NilmRun server',
+        formatter_class = argparse.ArgumentDefaultsHelpFormatter)
+    parser.add_argument("-V", "--version", action="version",
+                        version = nilmrun.__version__)
+    group = parser.add_argument_group("Standard options")
+    group.add_argument('-u', '--url',
+                       help = 'NilmRun server URL', default = def_url)
+    group.add_argument('-n', '--noverify', action="store_true",
+                       help = 'Disable SSL certificate verification')
+    group = parser.add_argument_group("Program")
+    group.add_argument('-d', '--detach', action="store_true",
+                       help = 'Run process and return immediately without '
+                       'printing its output')
+    group.add_argument('cmd', help="Command to run")
+    group.add_argument('arg', nargs=argparse.REMAINDER,
+                       help="Arguments for command")
+    args = parser.parse_args()
+
+    client = HTTPClient(baseurl = args.url, verify_ssl = not args.noverify)
+
+    # Run command
+    pid = client.post("run/command", { "argv": [ args.cmd ] + args.arg })
+
+    # If we're detaching, just print the PID
+    if args.detach:
+        print pid
+        raise SystemExit(0)
+
+    # Otherwise, watch the log output, and kill the process when it's done
+    # or when this script terminates.
+    try:
+        while True:
+            s = client.get("process/status", { "pid": pid, "clear": 1 })
+            sys.stdout.write(s['log'])
+            sys.stdout.flush()
+            if not s['alive']:
+                break
+            time.sleep(1)
+    finally:
+        s = client.post("process/remove", { "pid": pid })
+
+    raise SystemExit(s['exitcode'])
+
+if __name__ == "__main__":
+    main()

setup.py

@@ -61,7 +61,7 @@ setup(name='nilmrun',
       long_description = "NILM Database Filter Runner",
       license = "Proprietary",
       author_email = 'jim@jtan.com',
-      install_requires = [ 'nilmdb >= 1.8.2',
+      install_requires = [ 'nilmdb >= 1.9.5',
                            'psutil >= 0.3.0',
                            'cherrypy >= 3.2',
                            'simplejson',
@@ -75,6 +75,9 @@ setup(name='nilmrun',
       entry_points = {
           'console_scripts': [
               'nilmrun-server = nilmrun.scripts.nilmrun_server:main',
+              'nilmrun-ps = nilmrun.scripts.ps:main',
+              'nilmrun-run = nilmrun.scripts.run:main',
+              'nilmrun-kill = nilmrun.scripts.kill:main',
               ],
           },
       zip_safe = False,

tests/test_nilmrun.py

@@ -390,3 +390,20 @@ class TestClient(object):
         # Programs that spit out invalid UTF-8 should get replacement
         # markers
         verify("echo -ne \\\\xae", u"\ufffd")
+
+    def test_client_11_atexit(self):
+        # Leave a directory and running process behind, for the atexit
+        # handler to clean up.  Here we trigger the atexit manually,
+        # since it's hard to trigger it as part of the test suite.
+        client = HTTPClient(baseurl = testurl, post_json = True)
+        code = textwrap.dedent("""
+        import time
+        time.sleep(10)
+        """)
+        client.post("run/code", { "code": code, "args": [ "hello"] })
+
+        # Trigger atexit function
+        test_server._manager._atexit()
+
+        # Ensure no processes exit
+        eq_(client.get("process/list"), [])