Daniel,

OK! I gave it a few tries, but I've not been able to get tox up and running
(pip install -e . consistently fails with message "error: None"). So this
pull request has only been tested by running a restore with prefetch:

https://github.com/wal-e/wal-e/pull/144


I'll attach logs from that restore to the PR for reference. The commit
message at least offers a guess as to where prefetch might be tested.

-HJB


On Wed, Oct 22, 2014 at 1:27 PM, Daniel Farina <[email protected]> wrote:

> On Wed, Oct 22, 2014 at 1:08 PM, Hunter Blanks <[email protected]>
> wrote:
> > Daniel,
> >
> > Thanks for writing. Setting standby_mode = on and then triggering end of
> > recovery with trigger_file does work, regardless of whether prefetch is
> > enabled. More logs follow.
> >
> > I share your sympathies on the limitations of "exit-code-as-interface" we
> > get from PostgreSQL's restore_command. Of course, since we've daemonized
> > wal-prefetch, it doesn't actually matter if we exit non-zero? Couldn't we
> > just raise an exception if do_lzop_get() returns false?
>
> Yeah. That fix I think is quite tractable in a number of ways,
> including that one. Care to whip up a test (if not a huge pain) and
> patch?  If you don't have time to do it soon, I will, since there is a
> release pending.
>
> The other matter of dealing with crashes (unfortunately by experiment
> in my first design, syncing the WAL prefetched in this way is a
> meaningful bottleneck in practical scenarios) is a bit more
> troublesome.  I have my design above (boot-time based) which is not
> even quite 100% in events like messing with mounts.  And I'm not keen
> on something heavyhanded that punishes the common case for a
> vanishingly small (or, zero in the case of standby_mode=on) to solve
> the problem "completely".
>
> I think I may leave the crash recovery case as defect for now.
>

-- 
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