Re: inbox.comp.version-control.git just stopped working

[Sergey Organov]

Hi Eric,

Eric Wong  writes:

> Sergey Organov  wrote:
>> Eric Wong  writes:
>> > I'm not familiar with Gnus or Emacs at all, but is there a way
>> > you can show me a protocol dump?
>> 
>> Please see attachment.
>
> (any reason this wasn't sent to the public list? others may be
>  able to help if my 2nd attempt to fix did not work)

The only reason is the presence of rather long attachment.

> Anyways, I noticed you're pipelining a lot of XOVER commands and
> maybe
> https://public-inbox.org/meta/20180307190810.20784-...@80x24.org/raw
> helps (deployed to news.public-inbox.org)

Apparently it does, as it now works without any intervention from my
part.

Thanks a lot!

-- Sergey

Re: inbox.comp.version-control.git just stopped working

[Eric Wong]

Sergey Organov  wrote:
> Eric Wong  writes:
> > I'm not familiar with Gnus or Emacs at all, but is there a way
> > you can show me a protocol dump?
> 
> Please see attachment.

(any reason this wasn't sent to the public list? others may be
 able to help if my 2nd attempt to fix did not work)

Anyways, I noticed you're pipelining a lot of XOVER commands and
maybe
https://public-inbox.org/meta/20180307190810.20784-...@80x24.org/raw
helps (deployed to news.public-inbox.org)

Otherwise, I also looked at the emacs source ( ./lisp/gnus/nntp.el )
and found:

  (defvoo nntp-maximum-request 400
"*The maximum number of the requests sent to the NNTP server at one time.
  If Emacs hangs up while retrieving headers, set the variable to a
  lower value.")

Maybe you can set nntp-maximum-request to a smaller number
(maybe 5) and see if it helps.
--
unsubscribe: meta+unsubscr...@public-inbox.org
archive: https://public-inbox.org/meta/

Re: inbox.comp.version-control.git just stopped working

[Sergey Organov]

Quickly: still doesn't work :-(

Eric Wong  writes:
> Sergey Organov  wrote:
>> Hi,
>> 
>> I'm using inbox.comp.version-control.git for a few days now, and it
>> worked fine till now. I'm using Gnus from within Emacs to access it
>> via NNTP.
>> 
>> Just a few minutes ago, however, attempts to enter the
>> inbox.comp.version-control.git from Gnus started to give "Server
>> closed connection" error, even though some other groups on
>> public-inbox still work fine.
>
> Uh oh, did you notice any delay in how long it took for you to see
> this message from when Gnus started connecting?

It's about 4 seconds after Gnus successfully opens connection.

>
> I'm not familiar with Gnus or Emacs at all, but is there a way you can
> show me a protocol dump?

I dunno how to do it right now, but I'll try to figure out. Maybe using
tcpdump is the simplest...

>
>> The only difference I see is that inbox.comp.version-control.git has
>> much more messages than most of the other groups on the
>> public-inbox... wait...
>> 
>> Here are 3 fattest groups, and none of them I can visit from Gnus:
>> 
>> 438320: inbox.comp.lang.ruby.core
>> 438320: inbox.comp.lang.ruby.dev
>
> Eeek, I noticed 438320 was introduced via ruby-talk; so maybe I have a
> bug in that causing messages to leak into the other Ruby lists I
> follow :x

Trying to open those groups didn't work somewhat differently, but now it
does, I think due to lesser number of messages:

85962: inbox.comp.lang.ruby.core
85958: inbox.comp.lang.ruby.dev


>
>> 341199: inbox.comp.version-control.git
>
> OK, 341199 looks reasonable.
>
>> To check if it's Gnus problem, I've just visited:
>> 
>> 1523921: gmane.linux.debian.devel.bugs.general
>> 
>> and it works fine.
>> 
>> Any idea what's wrong between Gnus and public-inbox for large groups? 
>
> Not sure, I haven't made any changes to the code running on
> "public-inbox.org" in a few weeks, now.

I think it's there for a long time, and only showed itself once a
group gets big enough.

> I suspect there's a fairness problem in the NNTP server in how it
> handles XOVER and that might cause timeouts in clients.
>
> Fwiw, I might not be online much the next few days.

No problem, thanks for looking at it!

Eric Wong  writes:

> Eric Wong  wrote:
>> Sergey Organov  wrote:
>> > Any idea what's wrong between Gnus and public-inbox for large groups? 
>> 
>> Not sure, I haven't made any changes to the code running on
>> "public-inbox.org" in a few weeks, now.  I suspect there's a
>> fairness problem in the NNTP server in how it handles XOVER
>> and that might cause timeouts in clients.
>
> I just deployed this RFC on news.public-inbox.org and hope it
> improves things:
>
> https://public-inbox.org/meta/20180307095222.18699-...@80x24.org/raw
>
>> Fwiw, I might not be online much the next few days.
>
> Lets hope we get to the bottom of this quickly :>

That doesn't seem to be it, as it's still doesn't work, the same way as
before. I'll get back with protocol dump collected...

Thanks!

-- Sergey

Re: inbox.comp.version-control.git just stopped working

[Eric Wong]

Eric Wong  wrote:
> Sergey Organov  wrote:
> > Any idea what's wrong between Gnus and public-inbox for large groups? 
> 
> Not sure, I haven't made any changes to the code running on
> "public-inbox.org" in a few weeks, now.  I suspect there's a
> fairness problem in the NNTP server in how it handles XOVER
> and that might cause timeouts in clients.

I just deployed this RFC on news.public-inbox.org and hope it
improves things:

https://public-inbox.org/meta/20180307095222.18699-...@80x24.org/raw

> Fwiw, I might not be online much the next few days.

Lets hope we get to the bottom of this quickly :>
--
unsubscribe: meta+unsubscr...@public-inbox.org
archive: https://public-inbox.org/meta/

inbox.comp.version-control.git just stopped working

[Sergey Organov]

Hi,

I'm using inbox.comp.version-control.git for a few days now, and it
worked fine till now. I'm using Gnus from within Emacs to access it via
NNTP.

Just a few minutes ago, however, attempts to enter the
inbox.comp.version-control.git from Gnus started to give "Server closed
connection" error, even though some other groups on public-inbox still
work fine.

The only difference I see is that inbox.comp.version-control.git has
much more messages than most of the other groups on the
public-inbox... wait...

Here are 3 fattest groups, and none of them I can visit from Gnus:

438320: inbox.comp.lang.ruby.core
438320: inbox.comp.lang.ruby.dev
341199: inbox.comp.version-control.git

To check if it's Gnus problem, I've just visited:

1523921: gmane.linux.debian.devel.bugs.general

and it works fine.

Any idea what's wrong between Gnus and public-inbox for large groups? 

-- Sergey