remote_console cause atom leaks on target node

[elmirk]

Hi!
May be this is not an issue, but...
I use .../bin/myapp foreground to start release (target node),
then I connect to release with ../bin/myapp remote_console
and every time when I do connecting to running release with remote console there is atoms leak on target node.

As I understand there is two reasons:
1 - because used random id's for remote console
2 - because nodetool use
append_node_suffix(Name, Suffix) ->
case re:split(Name, "@", [{return, list}, unicode]) of
[Node, Host] ->
list_to_atom(lists:concat([Node, Suffix, os:getpid(), "@", Host]));
[Node] ->
list_to_atom(lists:concat([Node, Suffix, os:getpid()]))
end.

When I modify start script to not use random id but use predefined (for my case only one remote_console could be used at the moment) and also remove os:getpid() from nodetool escript - now there is no growing atoms number in target node.

Are there another way to do this with some configuration?

thanks!

[tsloughter]

Not currently. But a way to set the name of the node with an environment variable makes sense. Where it uses the random name if one isn't set.

[elmirk]

ok. I see.

[zsoci]

It happened at our site. I would keep the fix name and use the ENV-var if there is one. It is a bit frightening to have a tool that in long run fills up the atom table.

[tsloughter]

@zsoci in case you didn't see it #868 will fix this. But also with OTP 23.1 and above the issue will not exist because it does not use nodetool for communicating to the running node anymore.

[zsoci]

Thanks @tsloughter . We can live with #868 as it limins the new atoms to #ofnodes*1000. Moving to 23 is not simple as the issue is happening on our prod riak cluster (up and running for years) running on R16. Well we will try to use a generated script to see.