Error trying to create 5000 actors

[danielchiovitti]

When I try to create 5000 actors I'm getting the error "Killed" in the proxy.
With the help of @sleipnir I found this error:
Out of memory: Killed process 21551 (beam.smp) total-vm:29151740kB, anon-rss:16088396kB, file-rss:0kB, shmem-rss:66344kB, UID:1000 pgtables:48292kB oom_score_adj:0

Before the error the proxy created around 1000 actors

[sleipnir]

Some likely causes for actors to be so expensive could be the ActorRegistry stack or even the actor state...

spawn/apps/actors/lib/actors.ex

Line 35 in 5c2319b

ActorRegistry.register(actors)

spawn/apps/actors/lib/actors/actor/entity_supervisor.ex

Line 32 in 5c2319b

def lookup_or_create_actor(actor_system, %Actor{} = actor) do

I don't remember creating many atoms in the code, so maybe they don't help with that. Anyway, we will have to inspect the processes to find out the cause of the bootleg.

@danielchiovitti Please when starting the proxy, before making any request, type the following command in iex:

:observer_cli.start()

[sleipnir]

It seems to me that the state of each actor is too big for a start without any request.
6mb is a lot, we will have to rethink this

Also as I suspected the atoms are under control. We will have to decrease the initial state of the actors and probably the Actors.Registry.ActorRegistry too

[danielchiovitti]

[sleipnir]

We have to lower the state of the actors and that will lower the pressure on the supervisor as well.

Just for the sake of knowledge I opened a call on Horde to investigate if it is possible to use the new Elixir feature for partitioned supervisors derekkraan/horde#255. This does not have a direct relationship with the bug, but it does have an indirect one as the supervisor can be overwhelmed with the current architecture.