examples/python/comm-failure/comm-failure.py

   1 # Copyright (c) 2010-2023. The SimGrid Team. All rights reserved.
   2 #
   3 # This program is free software; you can redistribute it and/or modify it
   4 # under the terms of the license (GNU LGPL) which comes with this package.
   5
   6 import sys
   7
   8 from simgrid import Engine, Actor, Comm, NetZone, Link, LinkInRoute, Mailbox, this_actor, NetworkFailureException
   9
  10
  11 def sender(mailbox1_name: str, mailbox2_name: str) -> None:
  12     mailbox1: Mailbox = Mailbox.by_name(mailbox1_name)
  13     mailbox2: Mailbox = Mailbox.by_name(mailbox2_name)
  14
  15     this_actor.info(f"Initiating asynchronous send to {mailbox1.name}")
  16     comm1: Comm = mailbox1.put_async(666, 5)
  17
  18     this_actor.info(f"Initiating asynchronous send to {mailbox2.name}")
  19     comm2: Comm = mailbox2.put_async(666, 2)
  20
  21     this_actor.info("Calling wait_any..")
  22     pending_comms = [comm1, comm2]
  23     try:
  24         index = Comm.wait_any([comm1, comm2])
  25         this_actor.info(f"Wait any returned index {index} (comm to {pending_comms[index].mailbox.name})")
  26     except NetworkFailureException:
  27         this_actor.info("Sender has experienced a network failure exception, so it knows that something went wrong")
  28         this_actor.info("Now it needs to figure out which of the two comms failed by looking at their state:")
  29
  30     this_actor.info(f"  Comm to {comm1.mailbox.name} has state: {comm1.state_str}")
  31     this_actor.info(f"  Comm to {comm2.mailbox.name} has state: {comm2.state_str}")
  32
  33     try:
  34         comm1.wait()
  35     except NetworkFailureException as err:
  36         this_actor.info(f"Waiting on a FAILED comm raises an exception: '{err}'")
  37
  38     this_actor.info("Wait for remaining comm, just to be nice")
  39     pending_comms.pop(0)
  40     try:
  41         Comm.wait_any(pending_comms)
  42     except Exception as e:
  43         this_actor.warning(str(e))
  44
  45
  46 def receiver(mailbox_name: str) -> None:
  47     mailbox: Mailbox = Mailbox.by_name(mailbox_name)
  48     this_actor.info(f"Receiver posting a receive ({mailbox_name})...")
  49     try:
  50         mailbox.get()
  51         this_actor.info(f"Receiver has received successfully ({mailbox_name})!")
  52     except NetworkFailureException:
  53         this_actor.info(f"Receiver has experience a network failure exception ({mailbox_name})")
  54
  55
  56 def link_killer(link_name: str) -> None:
  57     link_to_kill = Link.by_name(link_name)
  58     this_actor.sleep_for(10.0)
  59     this_actor.info(f"Turning off link '{link_to_kill.name}'")
  60     link_to_kill.turn_off()
  61
  62 def main():
  63     e = Engine(sys.argv)
  64     zone: NetZone = NetZone.create_full_zone("AS0")
  65     host1 = zone.create_host("Host1", "1f")
  66     host2 = zone.create_host("Host2", "1f")
  67     host3 = zone.create_host("Host3", "1f")
  68
  69     link_to_2 = zone.create_link("link_to_2", "1bps").seal()
  70     link_to_3 = zone.create_link("link_to_3", "1bps").seal()
  71
  72     zone.add_route(host1, host2, [link_to_2])
  73     zone.add_route(host1, host3, [link_to_3])
  74     zone.seal()
  75
  76     Actor.create("Sender", host1, sender, "mailbox2", "mailbox3")
  77     Actor.create("Receiver-1", host2, receiver, "mailbox2").daemonize()
  78     Actor.create("Receiver-2", host3, receiver, "mailbox3").daemonize()
  79     Actor.create("LinkKiller", host2, link_killer, "link_to_2").daemonize()
  80
  81     e.run()
  82
  83
  84 if __name__ == "__main__":
  85     main()