Some release-related watchdog enabled by default?

Tristan Sloughter t@REDACTED
Sun Aug 2 16:23:29 CEST 2020

I've never heard of this but I will take the opportunity to suggest against using the daemon/run_erl way of running a release.

In the RC releases of rebar3 and relx `start` has been renamed to `daemon` to hopefully make it clearer that this is not the recommended command -- I feel being named `start` made it seem to users that it is the expected way to run a release.

See how it does using `foreground` instead. And yea, no need to use heart if you use a process manager like systemd to run the release with `foreground` , or are using a container platform.


On Sun, Aug 2, 2020, at 04:50, Olivier Boudeville wrote:
> Hi,
> Apparently, when running a release (at least if it was generated with 
> latest stable rebar3/relx and Erlang/OTP 23.0) as a daemon (with 
> run_erl), and even if not using specifically 'heart', a local UNIX 
> process keeps on trying (once per second!) to connect to the 
> corresponding VM.
> In my case, the release changes its cookie at start-up (based on an 
> application-level configuration file) and this may explain why, in the 
> erlang.log.* file, a line such as:
> (XXX@REDACTED)1> =ERROR REPORT==== 2-Aug-2020::12:17:46.042486 ===
> ** Connection attempt from node '24FIMJ0T6I9HU@REDACTED' rejected. Invalid 
> challenge reply. **
> is output each and every second (such a watchdog being thus unable to 
> connect, as it is presumably using the original cookie)
> Here are my questions:
> - what Erlang/OTP/relx/rebar3 component is responsible for this, and is 
> it really useful - especially if no specific heart behaviour is defined?
> - how can this be fixed? (preferably disabling these connection attempts 
> for good; otherwise telling such component which cookie to use in order 
> to poll properly)
> - is it something new? With earlier releases (pre-23.0?) of the same 
> codebase, such a behaviour was not noticed [1]
> If ever it mattered, this release relies on a non-standard EPMD port and 
> is run through authbind as a low-privileged user.
> Thanks in advance for any hint!
> Best regards,
> Olivier.
> PS: this question had already been posted a while ago in 
> yet did not 
> receive an answer.
> [1] It may have happened roughly when releases switched (for some 
> reason, and unless I am mistaken) from being executed with 'start' to, 
> now, either 'foreground' or 'daemon'.
> -- 
> Olivier Boudeville

More information about the erlang-questions mailing list