Daniel,

Thanks for writing.

Ah. I'll itemize the rationale:
>
> Postgres does a system() call to run the archive fetch command, and
> blocks until it is complete, then applies the fetched WAL.  If one
> only does parallelism, the result is that, say, 8 WAL segments will be
> downloaded in parallel and then Postgres will apply them...something
> that can take some time.  And in that time, no downloads are
> happening, which is a big loss.
>
>
Indeed. When I woke up this morning, it all made sense.


> I'm a bit reticent to do this in the release candidate part of the
> release cycle, but we can get on it pronto for 0.9dev and apply the
> patch first before more interesting changes go in. I'd recommend,
> then, using that.
>
>
That's fine by me. I've already got a package archive containing 0.8c1 and
its dependencies, so I don't think you should hold up the release while I
putter this out.


> Finally, I think I did things this way because one can do "apt-get
> install python-daemon" and get the dependency they need on Ubuntus.


That is a considerate choice, and indeed, when I first operationalized
WAL-E, I, too, reached for the Ubuntu system packages as a means for
installing WAL-E's deps. Sadly, some of the OpenStack dependencies didn't
line up with what was in Ubuntu 12.04 / 14.04. So it made more sense to
materialize all the dependencies into python packages (bdist_eggs. 2014 and
all, but really) and just follow a similar path to how we deploy other
Python environments.

-HJB

-- 
You received this message because you are subscribed to the Google Groups 
"wal-e" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
For more options, visit https://groups.google.com/d/optout.

Reply via email to