X-Git-Url: https://git.llucax.com/software/pymin.git/blobdiff_plain/c6010d17b74d9b70ba68b8f4567229c92a868f46..2e8aebbf35feff4c6e68c850193895c57acdd92f:/pymin/procman.py?ds=inline

diff --git a/pymin/procman.py b/pymin/procman.py
index 423ace0..0d67b9b 100644
--- a/pymin/procman.py
+++ b/pymin/procman.py
@@ -2,48 +2,185 @@
 
 import os
 import errno
+import signal
 import subprocess
+import logging ; log = logging.getLogger('pymin.procman')
+
+__all__ = ('ProcessManager', 'manager', 'register', 'unregister', 'once',
+           'start', 'stop', 'restart', 'kill', 'get', 'has', 'sigchild_handler')
 
 class ProcessInfo:
-    def __init__(self, name, process, args, kw, callback=None, persist=False):
-        self.name = name
-        self.process = process
-        self.args = args
-        self.kw = kw
+    def __init__(self, name, command, callback=None, persist=False,
+                 max_errors=3, args=None, kwargs=None):
+        self._name = name
+        self.command = command
         self.callback = callback
+        if args is None: args = list()
+        self.args = args
+        if kwargs is None: kwargs = dict()
+        self.kwargs = kwargs
         self.persist = persist
+        self.max_errors = max_errors
+        self.clear()
+    def clear(self):
+        self._dont_run = False
+        self._signal = None
+        self._process = None
+        self._error_count = 0
+    def start(self):
+        assert self.process is None
+        self.restart()
+    def restart(self):
+        self.clear()
+        log.debug(u'ProcessInfo.restart(): executing %s', self.command)
+        self._process = subprocess.Popen(self.command,
+                                         *self.args, **self.kwargs)
+    def stop(self):
+        assert self.process is not None
+        self._dont_run = True
+        if self._signal == signal.SIGTERM or self._signal == signal.SIGKILL:
+            # Allready stopped, kill it
+            self.kill(signal.SIGKILL)
+        else:
+            # Stop it
+            self.kill(signal.SIGTERM)
+    def kill(self, signum):
+        log.debug(u'ProcessInfo.kill(): killing pid %s with signal %s',
+                      self.process.pid, signum)
+        assert self.process is not None
+        os.kill(self.process.pid, signum)
+        self._signal = signum
+    @property
+    def running(self):
+        return self.process is not None and self.process.poll() is None
+    @property
+    def name(self):
+        return self._name
+    @property
+    def process(self):
+        return self._process
+    @property
+    def error_count(self):
+        return self._error_count
     def __repr__(self):
-        return 'ProcessInfo(name=%s, pid=%s, persist=%s, cb=%s, args=%s)' % (
-                    self.name, self.process.pid, self.persist,
-                    self.callback.__name__, self.args)
+        pid = None
+        if self.process is not None:
+            pid = self.process.pid
+        return 'ProcessInfo(name=%s, pid=%s command=%s, persist=%s, cb=%s)' % (
+                    self.name, pid, self.command, self.persist,
+                    self.callback.__name__)
 
 class ProcessManager:
 
     def __init__(self):
+        self.services = dict()
         self.namemap = dict()
         self.pidmap = dict()
+        log.debug(u'ProcessManager()')
+
+    def register(self, name, command, callback=None, persist=False,
+                max_errors=3, *args, **kwargs):
+        log.debug(u'ProcessManager.register(%s, %s, %s, %s, %s, %s, %s)',
+                  name, command, callback, persist, max_errors, args, kwargs)
+        assert not self.has(name)
+        pi = ProcessInfo(name, command, callback, persist, max_errors,
+                        args, kwargs)
+        self.services[name] = pi
+        return pi
+
+    def unregister(self, name):
+        log.debug(u'ProcessManager.unregister(%s)', name)
+        pi = self.services[name]
+        del self.services[name]
+        return pi
+
+    def _call(self, pi):
+        pi.start()
+        self.namemap[pi.name] = self.pidmap[pi.process.pid] = pi
+
+    def once(self, name, command, callback=None, persist=False,
+                max_errors=3, *args, **kwargs):
+        log.debug(u'ProcessManager.once(%s, %s, %s, %s, %s, %s, %s)',
+                  name, command, callback, persist, max_errors, args, kwargs)
+        assert not self.has(name)
+        pi = ProcessInfo(name, command, callback, persist, max_errors,
+                         args, kwargs)
+        self._call(pi)
+        return pi
+
+    def start(self, name):
+        log.debug(u'ProcessManager.start(%s)', name)
+        assert name not in self.namemap
+        self._call(self.services[name])
+
+    def stop(self, name):
+        log.debug(u'ProcessManager.stop(%s)', name)
+        assert name in self.namemap
+        self.namemap[name].stop()
+
+    def restart(self, name):
+        log.debug(u'ProcessManager.restart(%s)', name)
+        # we have to check first in namemap in case is an unregistered
+        # process (added with once())
+        if name in self.namemap:
+            pi = self.namemap[name]
+            # the process will change its PID, so we delete it while we know it
+            del self.pidmap[pi.process.pid]
+            pi.stop()
+            pi.process.wait()
+            pi.restart()
+            # add the new PID
+            self.pidmap[pi.process.pid] = pi
+        else:
+            self.start(name)
 
-    def call(self, name, callback, persist, *args, **kw):
-        proc = subprocess.Popen(*args, **kw)
-        procinfo = ProcessInfo(name, proc, args, kw, callback, persist)
-        self.namemap[name] = self.pidmap[proc.pid] = procinfo
+    def kill(self, name, signum):
+        log.debug(u'ProcessManager.kill(%s, %s)', name, signum)
+        assert name in self.namemap
+        self.namemap[name].kill(name, stop)
 
-    def sigchild_handler(self, signum):
+    def sigchild_handler(self, signum, stack_frame=None):
+        log.debug(u'ProcessManager.sigchild_handler(%s)', signum)
         try:
             (pid, status) = os.waitpid(-1, os.WNOHANG)
         except OSError, e:
-            if e.errno is e.ECHILD:
+            log.debug(u'ProcessManager.sigchild_handler(): OSError')
+            if e.errno is errno.ECHILD:
+                log.debug(u'ProcessManager.sigchild_handler(): OSError ECHILD')
                 return
             raise
+        log.debug(u'ProcessManager.sigchild_handler: pid=%s, status=%s',
+                      pid, status)
         while pid:
             if pid in self.pidmap:
                 p = self.pidmap[pid]
-                del self.namemap[p.name]
-                del self.pidmap[pid]
+                p.process.returncode = status
                 if p.callback is not None:
-                    p.callback(p)
-                if p.persist:
-                    self.call(p.name, p.callback, True, *p.args, **p.kw)
+                    log.debug(u'ProcessManager.sigchild_handler: '
+                                  u'calling %s(%s)', p.callback.__name__, p)
+                    p.callback(self, p)
+                if (p._dont_run or not p.persist
+                                or p._error_count >= p.max_errors):
+                    log.debug(u"ProcessManager.sigchild_handler: can't "
+                            u'persist, dont_run=%s, persist=%s, error_cout=%s, '
+                            u'max_errors=%s', p._dont_run, p.persist,
+                            p._error_count, p.max_errors)
+                    del self.namemap[p.name]
+                    del self.pidmap[pid]
+                    p.clear()
+                else:
+                    log.debug(u'ProcessManager.sigchild_handler: persist')
+                    if p.process.returncode == 0:
+                        p._error_count = 0
+                        log.debug(u'ProcessManager.sigchild_handler: '
+                                u'return OK, resetting error_count')
+                    else:
+                        p._error_count += 1
+                        log.debug(u'ProcessManager.sigchild_handler: return'
+                                u'not 0, error_count + 1 = %s', p._error_count)
+                    del self.pidmap[pid]
+                    p.restart()
+                    self.pidmap[p.process.pid] = p
             try:
                 (pid, status) = os.waitpid(-1, os.WNOHANG)
             except OSError, e:
@@ -51,17 +188,53 @@ class ProcessManager:
                     return
                 raise
 
-    def __getitem__(self, name):
+    def get(self, name):
         if isinstance(name, basestring): # is a name
-            return self.namemap[name]
+            if name in self.namemap:
+                return self.namemap[name]
+            if name in self.services:
+                return self.services[name]
         else: # is a pid
-            return self.pidmap[name]
+            if name in self.pidmap:
+                return self.pidmap[name]
+        raise KeyError, name
+    # Syntax sugar for self[name]
+    __getitem__ = get
 
-    def __contains__(self, name):
+    def has(self, name):
         if isinstance(name, basestring): # is a name
-            return name in self.namemap
+            if name in self.namemap:
+                return True
+            if name in self.services:
+                return True
         else: # is a pid
-            return name in self.pidmap
+            if name in self.pidmap:
+                return True
+        return False
+    # Syntax sugar for name in self
+    __contains__ = has
+
+
+if __name__ == '__main__':
+    logging.basicConfig(
+        level   = logging.DEBUG,
+        format  = '%(asctime)s %(levelname)-8s %(message)s',
+        datefmt = '%H:%M:%S',
+    )
+
+
+# Globals
+manager = ProcessManager()
+register = manager.register
+unregister = manager.unregister
+once = manager.once
+start = manager.start
+stop = manager.stop
+restart = manager.restart
+kill = manager.kill
+get = manager.get
+has = manager.has
+sigchild_handler = manager.sigchild_handler
 
 
 if __name__ == '__main__':
@@ -70,29 +243,85 @@ if __name__ == '__main__':
     import time
 
     sig = None
+    count = 0
 
-    def sigchild_handler(signum, stacktrace):
+    def SIGCHLD_handler(signum, stacktrace):
         global sig
         sig = signum
         print 'SIGCHLD', signum
 
-    def test_notify(proc):
-        print 'test died:', proc, proc.name, proc.process.pid
+    def notify(pm, pi):
+        global count
+        if pi.name == 'test-service':
+            print 'test-service count =', count
+            count += 1
+            if count > 4:
+                print 'set test-service non-persistent, start test-service-2'
+                pi.persist = False
+                assert 'test-service-2' not in manager.namemap
+                pm.start('test-service-2')
+                assert 'test-service-2' in manager.namemap
+                assert get('test-service-2').running
+        print 'died:', pi.name, pi.command
 
-    procman = ProcessManager()
+    register('test-service', ('sleep', '2'), notify, True)
+    assert 'test-service' in manager.services
+    assert 'test-service' not in manager.namemap
+    assert not get('test-service').running
+    assert manager['test-service'] == get('test-service')
+    assert has('test-service')
+    assert 'test-service' in manager
 
-    signal.signal(signal.SIGCHLD, sigchild_handler)
+    register('test-service-2', ('sleep', '3'), notify, False)
+    assert 'test-service-2' in manager.services
+    assert 'test-service-2' not in manager.namemap
+    assert not get('test-service-2').running
 
-    procman.call('test', test_notify, True, ('sleep', '5'))
+    signal.signal(signal.SIGCHLD, SIGCHLD_handler)
 
-    while True:
-        time.sleep(1)
-        print "Esperando...",
-        if 'test' in procman:
-            print procman['test']
-        else:
-            print
+    once('test-once', ('sleep', '5'), notify)
+    assert 'test-once' not in manager.services
+    assert 'test-once' in manager.namemap
+    assert get('test-once').running
+    assert get('test-once').process.pid
+    pid = get('test-once').process.pid
+    restart('test-once')
+    assert pid != get('test-once').process.pid
+    assert pid != manager.pidmap[get('test-once').process.pid].process.pid
+
+    start('test-service')
+    assert 'test-service' in manager.namemap
+    assert get('test-service').running
+    assert get('test-service').process.pid
+    pid = get('test-service').process.pid
+    restart('test-service')
+    assert pid != get('test-service').process.pid
+    assert pid != manager.pidmap[get('test-service').process.pid].process.pid
+
+    print "Known processes:", manager.services.keys()
+    print "Waiting...", manager.namemap.keys()
+    print "------------------------------------------------------------------"
+    while manager.pidmap:
+        signal.pause()
         if sig == signal.SIGCHLD:
+            sigchild_handler(sig)
             sig = None
-            procman.sigchild_handler(sig)
+        print "Known processes:", manager.services.keys()
+        print "Waiting...", manager.namemap.keys()
+        print "------------------------------------------------------------------"
+    assert 'test-service' not in manager.namemap
+    assert 'test-service-2' not in manager.namemap
+    assert 'test-once' not in manager.services
+    assert 'test-once' not in manager.namemap
+
+    restart('test-service')
+    assert get('test-service').process.pid
+    assert manager.pidmap[get('test-service').process.pid].process.pid
+
+    once('test-wait', ('sleep', '2'))
+    print 'test-wait running?', get('test-wait').running
+    assert get('test-wait').running
+    print 'Waiting test-wait to return...'
+    ret = get('test-wait').process.wait()
+    print 'Done! returned:', ret