[erlang-bugs 8] Re: possible supervisor bug in r14b02

Ingela Anderton Andin <>
Mon Mar 28 10:24:59 CEST 2011


Hi!

We will take a look at your patch it sounds like it is the right thing 
to do.
Temporary processes should not be restarted so you should not have
to save their  init-arguments, although until the last release they 
where saved
so it was possible to restart them! Especially for simple_one_for_one 
supervisors
that may have lots of temporary processes memory consumption can
go sky high if you save them.

Regards  Ingela Erlang OTP team - Ericsson AB

Filipe David Manana wrote:
> Hi,
>
> In R14B02 I noticed that for a child with a "temporary" restart_type,
> we discard its A component of the MFA tuple when adding the childspec
> to the list of the supervisor's children [1].
>
> When the child terminates, its spec is never removed from the list of
> the supervisor's children specs.
> Then if we call supervisor:restart_child/2 after the child terminates,
> the handle_call clause for restart_child gets the childspec with an
> MFA  that is {M, F, undefined} [2]. At that point do_start_child will
> call apply(M, F, undefined) [3] which will cause the supervisor to
> reply with an error, instead of returning {ok, Pid} as in previous
> releases. An example for the returned error:
>
>  {error,{'EXIT',{badarg,[{erlang,apply,[gen_server,start_link,undefined]},
>                               {supervisor,do_start_child,2},
>                               {supervisor,handle_call,3},
>                               {gen_server,handle_msg,5},
>                               {proc_lib,init_p_do_apply,3}]}}}
>
> The patch at [4] fixes the issue for me. The particular code that is
> no longer working in R14B02 but worked on all previous releases, is
> from Apache CouchDB, see [5]
>
> This issue was introuced by OTP-9064 (reading from the R14B02 release notes).
> Was this intended behaviour? It doesn't make much sense for me to keep
> a temporary childspec in the supervisor once the child terminates, so
> I believe deleting it from the state is the right thing to do.
>
> [1] - https://github.com/erlang/otp/blob/dev/lib/stdlib/src/supervisor.erl#L787
> [2] - https://github.com/erlang/otp/blob/dev/lib/stdlib/src/supervisor.erl#L314
> [3] - https://github.com/erlang/otp/blob/dev/lib/stdlib/src/supervisor.erl#L246
> [4] - https://github.com/fdmanana/otp/commit/2697042aa9ebab2fcd208c93b7f454b25bc580d4
> [5] - https://github.com/apache/couchdb/blob/trunk/src/couchdb/couch_replicator.erl#L119
>
>   



More information about the erlang-bugs mailing list