delay restart (retry)

docs/httpok.rst

@@ -118,6 +118,10 @@ Command-Line Syntax
    Disable "eager" monitoring:  do not check the URL or emit mail if no
    monitored process is in the RUNNING state.
 
+.. cmdoption:: -w <seconds>, --delay=<seconds>
+
+   Number of seconds delay before really make a restart.
+
 .. cmdoption:: <URL>
 
    The URL to which to issue a GET request.

superlance/httpok.py

@@ -26,7 +26,7 @@
 
 doc = """\
 httpok.py [-p processname] [-a] [-g] [-t timeout] [-c status_code] [-b inbody]
-          [-m mail_address] [-s sendmail] URL
+          [-m mail_address] [-s sendmail] [-w delay_seconds] URL
 
 Options:
 
@@ -81,6 +81,8 @@
 
 -E -- not "eager":  do not check URL / emit mail if no process we are
       monitoring is in the RUNNING state.
+
+-w -- specify number of seconds delay before really make a restart.
 
 URL -- The URL to which to issue a GET request.
 
@@ -114,7 +116,7 @@ def usage():
 class HTTPOk:
     connclass = None
     def __init__(self, rpc, programs, any, url, timeout, status, inbody,
-                 email, sendmail, coredir, gcore, eager, retry_time):
+                 email, sendmail, coredir, gcore, eager, delay, retry_time):
         self.rpc = rpc
         self.programs = programs
         self.any = any
@@ -128,6 +130,7 @@ def __init__(self, rpc, programs, any, url, timeout, status, inbody,
         self.coredir = coredir
         self.gcore = gcore
         self.eager = eager
+        self.delay = delay
         self.stdin = sys.stdin
         self.stdout = sys.stdout
         self.stderr = sys.stderr
@@ -272,6 +275,10 @@ def mail(self, email, subject, msg):
     def restart(self, spec, write):
         namespec = make_namespec(spec['group'], spec['name'])
         if spec['state'] is ProcessStates.RUNNING:
+            if self.delay and (spec['now']-spec['start'])<self.delay:
+                write('%s is in RUNNING state %d seconds (<%d), delay restarting' % \
+                    (namespec, spec['now']-spec['start'], self.delay))
+                return
             if self.coredir and self.gcore:
                 corename = os.path.join(self.coredir, namespec)
                 cmd = self.gcore + ' "%s" %s' % (corename, spec['pid'])
@@ -299,7 +306,7 @@ def restart(self, spec, write):
 
 def main(argv=sys.argv):
     import getopt
-    short_args="hp:at:c:b:s:m:g:d:eE"
+    short_args="hp:at:c:b:s:m:g:d:eEw:"
     long_args=[
         "help",
         "program=",
@@ -313,6 +320,7 @@ def main(argv=sys.argv):
         "coredir=",
         "eager",
         "not-eager",
+        "delay=",
         ]
     arguments = argv[1:]
     try:
@@ -334,6 +342,7 @@ def main(argv=sys.argv):
     email = None
     timeout = 10
     retry_time = 10
+    delay = None
     status = '200'
     inbody = None
 
@@ -375,6 +384,9 @@ def main(argv=sys.argv):
         if option in ('-E', '--not-eager'):
             eager = False
 
+        if option in ('-w', '--delay'):
+            delay = int(value)
+
     url = arguments[-1]
 
     try:
@@ -388,7 +400,7 @@ def main(argv=sys.argv):
         return
 
     prog = HTTPOk(rpc, programs, any, url, timeout, status, inbody, email,
-                  sendmail, coredir, gcore, eager, retry_time)
+                  sendmail, coredir, gcore, eager, delay, retry_time)
     prog.runforever()
 
 if __name__ == '__main__':

superlance/tests/httpok_test.py

@@ -63,7 +63,7 @@ def _makeOne(self, *opts):
         return self._getTargetClass()(*opts)
 
     def _makeOnePopulated(self, programs, any, response=None, exc=None,
-                          gcore=None, coredir=None, eager=True):
+                          gcore=None, coredir=None, eager=True, delay=None):
         if response is None:
             response = DummyResponse()
         rpc = DummyRPCServer()
@@ -76,9 +76,10 @@ def _makeOnePopulated(self, programs, any, response=None, exc=None,
         inbody = None
         gcore = gcore
         coredir = coredir
+        delay = delay
         prog = self._makeOne(rpc, programs, any, url, timeout, status,
                              inbody, email, sendmail, coredir, gcore, eager,
-                             retry_time)
+                             delay, retry_time)
         prog.stdin = StringIO()
         prog.stdout = StringIO()
         prog.stderr = StringIO()
@@ -312,5 +313,24 @@ def test_runforever_connrefused_error(self):
         self.assertEqual(mailed[1],
                     'Subject: httpok for http://foo/bar: bad status returned')
 
+    def test_runforever_not_eager_running_delay150(self):
+        programs = ['foo', 'bar']
+        any = None
+        prog = self._makeOnePopulated(programs, any, exc=True, eager=False, delay=150)
+        prog.stdin.write('eventname:TICK len:0\n')
+        prog.stdin.seek(0)
+        prog.runforever(test=True)
+        lines = [x for x in prog.stderr.getvalue().split('\n') if x]
+        self.assertEqual(lines[0],
+                         ("Restarting selected processes ['foo', 'bar']")
+                         )
+        self.assertEqual(lines[1], 'foo is in RUNNING state 100 seconds (<150), delay restarting')
+        self.assertEqual(lines[2], 'bar not in RUNNING state, NOT restarting')
+        mailed = prog.mailed.split('\n')
+        self.assertEqual(len(mailed), 9)
+        self.assertEqual(mailed[0], 'To: [email protected]')
+        self.assertEqual(mailed[1],
+                    'Subject: httpok for http://foo/bar: bad status returned')
+
 if __name__ == '__main__':
     unittest.main()