:: Re: [devuan-dev] Introducing apt-pa…
Página superior
Eliminar este mensaje
Responder a este mensaje
Autor: onefang
Fecha:  
A: devuan-dev
Asunto: Re: [devuan-dev] Introducing apt-panopticon, my Devuan mirror checker script.
On 2020-01-02 09:59:28, Denis Roio wrote:
> old-thread, I'm feeling like chipping in


I refresh it every now and then with an update. The next update isn't
quite ready yet.

> On Sat, 09 Nov 2019, onefang wrote:
>
> > C and Lua, and even C + Lua, are my current favourite programming
> > languages, along with some assembler when needed. I've been a
> > professional programmer since the late '70s, and I have used 100
> > programming languages in my career, so I consider myself to have
> > earned the right to be a language bigot. B-)
>
> I think you are just very experienced and then also wise. Being a
> programmer since the late '90s I just come to the same conclusion. C
> + Lua is a mindblowing combination, esp. considering the portability:
> people must keep in mind Lua is *embeddable* and very easy to do so.
>
> Recently I've been coding this https://dev.zenroom.org which is all
> C99+Lua and I am *very happy* with the result in terms of speed,
> maintainability and portability.
>
> Said that I admit that recent hypes are more interesting than before:
> Rust, Golang, Vlang, Nim etc. have much bigger value propositions than
> the previous generation Perl, Python and Ruby. I keep an eye since
> they are bytecode compilable or even in case of Vlang compilable to C
> (a very interesting approach).
>
> But I doubt that anything can resist the test of time like C and as
> well Lua because of the brilliant way it is designed and because of
> the enormous battle-testing given by its adoption in WoW and also less
> frivolous and more mission critical deployements.
>
> There are also new C+Lua products incredibly performant: nginx-lua /
> Openresty, the lua-lapis web framework (and moonscript lang wrapper)
> and last not least tarantool which is a bit more complex but
> incredibly handy and performant for large scale distributed
> deployements.


I was hoping that Openresty would be in the Devuan repos, but alas no. I
should play with it someday anyway. I also have my eye on lighttpd and
it's Lua support. Lua-lapis and tarantool I don't recall reading about.
I am on the lookout for a decent C + Lua based web framework. Though I
would prefer something that is in Devuan repos, or at least has a
developer maintained Debian repo I can add. Or I may end up writing my
own. lol

> At last, many thanks for apt-panopticon. Looking at it after some
> months now I think we could draw a line and let some mirrors go, those
> that are evidently offline as well badly performing.


I didn't want to start using it to make any decisions until I get a full
release out, and right now I think it's close to an alpha release. I'm
still tweaking things, and especially the historic data has some test
artefacts in it that are skewing the figures. I had recently fixed a bug
that was keeping the reported speeds low. The weekly statistics are also
still awaiting a whole week of clean data, I expect more 100%s to show in
a day or two.

My plan is -

Carefully go over the reported problems, to make sure they are not false
negatives. I'm part way through this now.

Make an alpha release.

Start posting weekly email reports. Likely I'll automate this.

Ask both this and the mirror mailing list to give it a good look at, see
if anyone can spot any problems with my code and the results. In
particular those mirror admins who have had persistent issues reported
get a chance to tell me why I'm reporting things wrong.

Write better documentation, fix up any problems, and make better graphs.

Make a beta release, and warn everyone that actual decisions will soon be
based on the results.

Talk with admins of mirrors that are still showing consistent problems
that I can't attribute to bugs in my code. See if we can resolve those
issues, or fix my bugs.

Make a full release, and as one of the official mirror herders, start
using the reported results to inform decisions about mirrors. Then I
think we can draw the line. And perhaps make having a clean record for a
week / month a test to pass before a mirror is added to the DNS-RR.

Add it to my Icinga tests, and have it send emails to this list, and
maybe to the admin concerned, when mirrors get ERRORS.

Starting on Monday I move my server to a new faster and bigger server, a
sort of Christmas gift from my server company, it's the same price. B-)

After I'm done with that move I'll shift my apt-panopticon testing to
every ten minutes instead of hourly as it is at the moment. Coz the new
server has more than enough disk space to spare. The test runs are
currently consistently under a minute, I might switch to testing each
minute later, depending how things go. Or lesser tests once a minute,
more involved tests more often. There's already code for doing that.

--
A big old stinking pile of genius that no one wants
coz there are too many silver coated monkeys in the world.