Re: postgresql systemd service fails to start only on boot but notmanually

## Doron Behar (doron.behar@gmail.com):

> My server fails to start PostgreSQL only on boot, if I restart it
> manually afterwards it doesn't have any problem starting. Here is the
> log extracted from the journal:
> 
> ```
> 2018-09-21 20:46:40.028 CEST [306] LOG:  listening on IPv4 address "127.0.0.1", port 5432
> 2018-09-21 20:46:40.036 CEST [306] LOG:  listening on Unix socket "/run/postgresql/.s.PGSQL.5432"
> 2018-09-21 20:46:40.233 CEST [337] LOG:  database system was shut down at 2018-09-21 20:46:21 CEST
> 2018-09-21 20:48:10.441 CEST [352] WARNING:  worker took too long to start; canceled
> 2018-09-21 20:49:10.469 CEST [352] WARNING:  worker took too long to start; canceled

This would indicate that your machine is overloaded during start -
perhaps there's just too much being started at the same time?
ObRant: that's what happens if people take "system startup duration"
as a benchmark and optimize for that - sure, running one clumsy shell
script after another isn't effective usage of today's systems,
but starting eight dozens programs all at once may have other
side effects. Really, with the hardware taking small ages to find
it's own arse before even loading the boot loader, those few seconds
weren't worth optimizing - and if people reboot their computers so
often that startup time takes a measurable toll on their productive
day, perhaps they should rather spend their time thinking about their
usage pattern than "optimizing" the startup process.

So, now that I've got that off my chest... your machine propably tries to
do too much at the same time when booting: the worker processes take
longer than 90 seconds to start. Slow CPU or storage maybe?

> 2018-09-21 20:49:10.478 CEST [306] LOG:  database system is ready to accept connections
> 2018-09-21 20:49:10.486 CEST [306] LOG:  received fast shutdown request

And in the mean time, systemd has lost it's patience, declares the
start as failed and terminates the process group. (The default systemd
timeout is 90 seconds, at least in some releases of systemd, so
this fits quite nicely).

You could try to work around this by increasing TimeoutStartSec
in postgresql's systemd unit (or even globally), which perhaps
only hides the problem until the next service suddenly doesn't
start anymore.
You could move postgresql to the end of the boot order by
adding "After=..." to the Unit section of the systemd service
file, the value behind "After=" being all the other services in
the same target, which should reduce parallelism and improve
PostgreSQL's startup behaviour.
A more advanced variant of that would be to create a new
systemd target, make that start "After" multiuser.target
or even graphical.target (depending on your setup), make sure
it "Requires" the current default systemd target and make
postgresql the only additional service in that target.
(This would be the cleanest solution, but you should get some
grasp of systemd and how your specific distribution uses it
before meddling with the default targets; I don't know every
distribution/version variant of systemd integration, so I
can't give that specific instructions here).
Or you figure out what the heck your machine is running
during startup any why it is that slow, and try to fix that.

Regards,
Christoph

-- 
Spare Space