bug#23064: herd has incomplete status reporting, not so helpful

[Danny Milosavljevic <dannym@scratchpost.org>]

> Sounds like postgresql died and shepherd did not notice?  Or maybe it
> keeps trying to respawn it?  What did /var/log/shepherd.log say?

2016-03-19 10:34:48 Service postgres has been started.
2016-03-19 10:34:49 Respawning postgres.
2016-03-19 10:34:49 Service postgres has been started.
2016-03-19 10:34:50 Respawning postgres.
2016-03-19 10:34:50 Service postgres has been started.
2016-03-19 10:34:51 Respawning postgres.
2016-03-19 10:34:51 Service postgres has been started.
2016-03-19 10:34:52 Respawning postgres.
2016-03-19 10:34:52 Service postgres has been started.
2016-03-19 10:34:53 Respawning postgres.
2016-03-19 10:34:53 Service postgres has been started.
2016-03-19 10:34:54 Respawning postgres.
2016-03-19 10:34:54 Service postgres has been started.

> Currently the Shepherd does not do log stdout/stderr of its child
> processes, which sucks somewhat.

Yeah, that's not good.

> However, daemons can usually be told to write to syslog, which is more
> appropriate than writing things to stdout/stderr anyway.  

It can always be unable to open syslog for some reason. 

If shepherd can't/doesn't redirect stderr on its own, it would be nice to have init write it somewhere and everyone else inherit it as default. Just throwing it away is not nice. 

But I agree, the chance of being able to write it to syslog is high. Btw: How does guixsd know to start the syslog service before the postgres service?

>What’s the right command-line/configuration option to have postgresql use syslog?

The option is

   log_destination = 'syslog'

in postgresql.conf

which is generated in gnu/services/databases.scm (%default-postgres-config).

Does shepherd back off from respawing it eventually (if it respawns too fast) or will it log the same messages into syslog once every 0.1 s until my disk is full? :->