launchpad-dev team mailing list archive

[Robert Collins <robertc@xxxxxxxxxxxxxxxxx>]

On Sat, Feb 18, 2012 at 7:10 AM, Julian Edwards
<julian.edwards@xxxxxxxxxxxxx> wrote:
> On Friday 17 February 2012 14:49:11 Jonathan Lange wrote:
>> On Thu, Feb 16, 2012 at 11:34 PM, Robert Collins
>>
>> <robertc@xxxxxxxxxxxxxxxxx> wrote:
>> > I think we should disable longpoll; its not finished, noone is working
>> > on it, and it causes semi-regular confusion for folk in the LP team
>> > when they cannot use code.launchpad.net.
>> >
>> > Any objections? [I'm not proposing removing the code yet].
>>
>> I have a mild objection. As a frequent user of the website, I would
>> really like to see this feature completed.
>
> Hear hear.
>
> We spent so much time and effort getting it this far that my team and I would
> be *very* frustrated if it were turned off.

Me too - I'm also very frustrated that its stalled, that we switched
your squad to maintenance without finishing the project; and that
finish-this-on-maintenance didn't happen. There are good reasons for
all those things, but we're still where we are today.

> Personally, I don't care about people who have 20+ tabs open on Launchpad, I
> think they're freaks and edge cases.  The majority of users will not be doing
> that and will get extreme satisfaction from the working feature.
>
> Let's get the wildcard domain thing sorted out which will alleviate the
> initial concerns and for those using Firefox with its global connection limit
> - tough.

The wildcard approach would indeed support up to 23 *total* tabs in
the browser : but note that on the 24th tab, the site will halt. This
cap is shared amongst all pages, so twitter, facebook, identi,ca,
gmail, google plus - each of those counts towards the limit; you don't
actually need /all that/ many things active to run into a limit.

I think the wildcard approach + docs is a reasonable first step, but I
think we need to make sure we can tell how many concurrent polls users
have (because that will cost *us* as well) - one way would be to
capture the LP username in the long poll request and log it : even if
its not trusted data (e.g. not secure) we'd be able to get indicative
statistics ('total concurrency was no higher than N for X users').

-Rob