A try to make greater robustness when calling client ping

Samo Penic

2018-12-23 1bc9bba5a3bb201345fcf6b690b1c884f8ab8a7d

 tsclient.py

@@ -10,6 +10,21 @@
import sys
import socket
from threading import Thread, Event
import re



glob_ts_version='00000'

def getTrisurfVersion():
   p = subprocess.Popen('trisurf --version', shell=True, stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
   lines=p.stdout.readlines()
   version=re.findall(r'[0-9a-f]{7}(?:-dirty)?', lines[0].decode('ascii'))
   p.wait()
   if(len(version)):
      return version[0]
   else:
      return "unknown version"

def get_hostname():
   return socket.gethostname()
@@ -18,7 +33,8 @@
   return ((([ip for ip in socket.gethostbyname_ex(socket.gethostname())[2] if not ip.startswith("127.")] or [[(s.connect(("8.8.8.8", 53)), s.getsockname()[0], s.close()) for s in [socket.socket(socket.AF_INET, socket.SOCK_DGRAM)]][0][1]]) + ["no IP found"])[0])

def get_client_id(addr, my_ip, my_hostname, subrun):
   client_auth={'ip':my_ip,'hostname':my_hostname, 'subrun':subrun}
   global glob_ts_version
   client_auth={'ip':my_ip,'hostname':my_hostname, 'subrun':subrun, 'trisurf_version':glob_ts_version }
   response=requests.post(addr+"/api/register/", data=client_auth)
   if(response.status_code==200):
      client_data=json.loads(response.text)
@@ -38,7 +54,7 @@
      status=client_data['status']
      return (rid,tape,vtu,status)
   else:
      print(response.text)
      #print(response.text)
      if(response.status_code==400):
         raise ValueError
      else:
@@ -162,15 +178,19 @@
            print("[{}] Could not get CID.".format(self.id))
            self.sleep(10)
            continue
         #print("Got CID. getting RID.")
         client_ping_time_elapsed=0
         concurrent_runs=client_ping(self.conn_address,cid)
         print("[{}] Connected and got client ID {}.".format(self.id, cid))
         try:
            concurrent_runs=client_ping(self.conn_address,cid)
            client_ping_time_elapsed=0
         except:
            self.sleep(10)
            continue
         self.subrunsStartStop(concurrent_runs)
         while(not self.isStopped()): #successfully registered, now start pinging and searching for job
            try:
               (rid,tape,vtu,status)=get_run(self.conn_address,cid)
            except NameError:
               print("[{}] Could not get RID.".format(self.id))
               #print("[{}] Could not get RID.".format(self.id))
               self.sleep(10)
               client_ping_time_elapsed+=10
               if(client_ping_time_elapsed>=self.max_client_ping_time_elapsed):
@@ -236,7 +256,7 @@
                     try:
                        ping_run(self.conn_address, cid, rid)
                     except:
                        print("[{}] Could not ping.".format(self.id))
                        print("[{}] Could not prolong a lease on the run.".format(self.id))
                        self.p.terminate()
                        self.p=None
                        removeDir(self.workingdir.fullpath())
@@ -259,7 +279,15 @@
                  self.sleep(self.update_seconds-1)
                  client_ping_time_elapsed+=self.update_seconds
                  if(client_ping_time_elapsed>self.max_client_ping_time_elapsed-self.update_seconds/2):
                     concurrent_runs=client_ping(self.conn_address,cid)
                     try:
                        concurrent_runs=client_ping(self.conn_address,cid)
                     except:
                        print("[{}] Could not client ping.".format(self.sid))
                        self.p.terminate()
                        self.p=None
                        removeDir(self.workingdir.fullpath())
                        self.workingdir=None
                        break
                     self.subrunsStartStop(concurrent_runs)
                     client_ping_time_elapsed=0

@@ -338,7 +366,8 @@
#--- END SIGINT and SIGTERM----

if __name__ == '__main__':

   #global glob_ts_version
   glob_ts_version=getTrisurfVersion()
   signal.signal(signal.SIGINT, signal_handler)
   signal.signal(signal.SIGTERM, signal_handler)