Обсуждение: Dynamic background workers & docs question

<div dir="ltr">I was looking at <a
href="http://www.postgresql.org/docs/devel/static/bgworker.html">http://www.postgresql.org/docs/devel/static/bgworker.html</a>
witha client today.<br /><br />It says:<br />"Unlike RegisterBackgroundWorker, which can only be called from within the
postmaster,RegisterDynamicBackgroundWorkermust be called from a regular backend."<br /><br />Is that the correct
restriction?In particular, don't we allow calling RegisterDynamicBackgroundWorker from another background worker? (In
thelauncher/worker kind of scenario, like AutoVacuum).<br /><br />Also:<br />"Background workers are expected to be
continuouslyrunning; if they exit cleanly, postgres will restart them immediately. "<br /><br />This doesn't apply to
dynamicones, which we might want to clarify. Do we have a "term" for non-dynamic background workers? "static
workers"?<br/><br /><br /><br />--<br /> Magnus Hagander<br /> Me: <a
href="http://www.hagander.net/">http://www.hagander.net/</a><br/>  Work: <a
href="http://www.redpill-linpro.com/">http://www.redpill-linpro.com/</a><br/></div> 

On Mon, Mar 24, 2014 at 5:54 PM, Magnus Hagander <magnus@hagander.net> wrote:
> I was looking at http://www.postgresql.org/docs/devel/static/bgworker.html
> with a client today.
>
> It says:
> "Unlike RegisterBackgroundWorker, which can only be called from within the
> postmaster,RegisterDynamicBackgroundWorker must be called from a regular
> backend."
>
> Is that the correct restriction? In particular, don't we allow calling
> RegisterDynamicBackgroundWorker from another background worker? (In the
> launcher/worker kind of scenario, like AutoVacuum).
Yes, you can start a dynamic background worker from another background
worker, have a look for example at contrib/worker_spi. Perhaps the
correct wording would be "RegisterDynamicBackgroundWorker must be
called from a regular backend or another background worker".

> Also:
> "Background workers are expected to be continuously running; if they exit
> cleanly, postgres will restart them immediately. "
>
> This doesn't apply to dynamic ones, which we might want to clarify. Do we
> have a "term" for non-dynamic background workers? "static workers"?
In the code or the documentation, there is no explicit
differentiation, bgworkers are either called plainly "bgworker", or
"dynamic bgworker". Perhaps the solution here is simply to say
"background workers started by the postmaster are expected blabla".
-- 
Michael

On Mon, Mar 24, 2014 at 11:03 PM, Magnus Hagander <magnus@hagander.net> wrote:
> That's what I thought. Can a dynamic background worker start *another*
> dynamic background worker, or can they only be started from "first level"
> background workers?
I have never really tried by myself, but I don't see any reason why it
wouldn't work as it is only a matter of doing what is for example in
worker_spi_launch:worker_spi.c. Btw, a bgworker could also behave like
a "regular backend" as mentioned in the docs, so a regular backend is
just a subclass of a bgworker :)

>> > Also:
>> > "Background workers are expected to be continuously running; if they
>> > exit
>> > cleanly, postgres will restart them immediately. "
>> >
>> > This doesn't apply to dynamic ones, which we might want to clarify. Do
>> > we
>> > have a "term" for non-dynamic background workers? "static workers"?
>> In the code or the documentation, there is no explicit
>> differentiation, bgworkers are either called plainly "bgworker", or
>> "dynamic bgworker". Perhaps the solution here is simply to say
>> "background workers started by the postmaster are expected blabla".
> That, or we need to invite a term for it?
Hm... Seems like an overkill. The main difference between a
non-dynamic and dynamic bgworker is the way they are registered.
"Static" bgworkers use RegisterBackgroundWorker that can only be
called in _PG_init when a module is loaded with
shared_preload_libraries. Dynamic bgworkers use
RegisterDynamicbackgroundWorker. And this differentiation is clearly
done in the 2nd paragraph.

So perhaps the solution here is simply to write "Background workers
registered with RegisterBackgroundWorker are expected...".

I am not a native English speaker, but "static" sounds like a daemon
process that has to restart, and a bgworker could perform a one-time
task as well.
-- 
Michael