I realize it's not iron-clad without understanding what's going on in the
code, but...my tests today with another system and S3 bucket confirm the
same behavior. Concurrency does not affect it. Here's what I did:

- created a new S3 bucket in Oregon
- attempted the same backup as normal; it failed with the "[Errno
104] Connection reset by peer" error.
- added the "-p 1" flag to the backup command; it failed in the same way
- changed WALE_S3_PREFIX to point at another new S3 bucket, this time in
"US Standard"
- the backup worked (with and without the -p flag).

Even if this discussion thread is findable on Google, I bet a lot of people
are hitting this - who would think the choice of bucket region would
matter? I was ready to accept that I couldn't use WAL-E at all until I
heard this tip (which would be a shame, since it's a far better way to do
replication than the out-of-the-box suggestions from Postgres).

Dave


On Thu, Mar 13, 2014 at 11:24 PM, Daniel Farina <[email protected]> wrote:

>
> On Mar 13, 2014 11:18 PM, "David Weaver" <[email protected]> wrote:
> >
> > I was told, on a tip from a fellow engineer, to switch my S3 bucket from
> the "Oregon" region (which I had been using) to the "US Standard" one. Sure
> enough, that worked, and I no longer see the "[Errno 104] Connection reset
> by peer" errors from my WAL-E operations.
> >
> > I don't think this is a CPU issue (as a medium-sized instance should be
> plenty powerful for something like this), but one related to the S3 region.
> At least in my case, this distinction was essential to my being able to use
> WAL-E at all - perhaps it's worth at least a mention in the documentation?
>
> Fascinating. If you have copious free time I'd like to know if reducing
> concurrency helps.
>
> As for putting it in the documentation: so far this seems too micro to
> itemize given I don't have a way to organize such copious detail without
> diluting top-line information.  The causes and fixes aren't multiply
> confirmed nor iron clad either.
>
> Fixing up the back trace to instead return a nice error with a HINT field
> may be good, provided there was formulaic information for the hint.
>
> I think this mailing list archive will suffice for someone searching for
> the error.
>

-- 
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