Date: Mon, 27 Jun 2022 13:03:51 +0000
To: blallo@autistici.org
From: Simon Ser <contact@emersion.fr>
Cc: ~emersion/hut-dev@lists.sr.ht
Reply-To: Simon Ser <contact@emersion.fr>
Subject: Re: [PATCH] Make timeout configurable
Message-ID: <NWPVz4QyKgPHOFwL0brUZ-8vFsObnQKDMplMRCbvezIB1NXFeYvElZpkRmV30y-UurEMrA6lBLjXX6wDew_POxGKluAuBTVU8lwHh4k9vOQ=@emersion.fr>
In-Reply-To: <20220626081614.q6basbarqr2qpjxf@rita>
References: <20220516150540.42563-1-blallo@autistici.org> <3JNzmIMfnnIA2cTXpbkaMFRkxng1BQv1Luatof43V25F4MpgSFS_k5E9zFm7WPdpnR0DasSRPKgS3dAr1w5iodrZTsc8UvEifatPy6ImF6Y=@emersion.fr> <20220626081614.q6basbarqr2qpjxf@rita>
Feedback-ID: 1358184:user:proton
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable

On Sunday, June 26th, 2022 at 10:16, <blallo@autistici.org> wrote:

> Simon Ser:
>
> > On Monday, May 16th, 2022 at 17:02, Blallo blallo@autistici.org wrote:
> >
> > > The http client used for interacting with the hut instance had a
> > > hardcoded 30 seconds timeout. While this might be ok in most cases, I
> > > happened to hit one of the corner cases when trying to push a tarball=
 to
> > > site.ht. This commit adds a global flag to set the value, defaulting =
to
> > > the old one.
> >
> > Hm. If possible, I'd prefer to do the right thing by default and set a
> > longer timeout for queries involving large file uploads, without
> > requiring the user to manually pass a --timeout option.
>
> (Sorry for the late reply). I understand that it would be a great
> improvement from the user experience perspective, but it is a bit
> cumbersome to implement. I have two approaches in mind:
> - Let the client compute the size of the objects it wants to push to
> the remote

This input file may come from stdin, in which case it's not possible to
compute the size.

> - Let the client use retries with "exponential" timeout increase
> (like, first timeout 30s, then 60s, then a last try with 120s?)
>
> Both approaches have drawbacks: in the first case either we have to
> specify a body size (that we have to compute) for each call or consume
> the io.Reader passed to the client before submitting the true
> request; in the second case, the worst case has the user waiting for
> 30+60+120 =3D 210 seconds.
>
> I prefer the first approach, but I want your opinion before trying to
> do that.

Can't we just use a fixed timeout of e.g. 5min, just for this command?