[Potlatch-dev] [OT?] problem with serving potlatch2.swf on XP

[Dave Stubbs]

On Mon, Oct 4, 2010 at 11:42 AM, Andy Allan <gravitystorm at gmail.com> wrote:
> On Sun, Oct 3, 2010 at 10:44 PM, Steve Bennett <stevagewp at gmail.com> wrote:
>> This probably isn't a potlatch-specific problem, but if anyone can
>> help, I'd be very grateful. After doing quite a few cycles of making a
>> change to potlatch.css, then reloading the css, Chrome eventually
>> stopped loading the .swf. I switched to Opera, which worked for a bit
>> until it crashed, and only half loads the .swf. Then FireFox
>> eventually did the same thing.
>>
>> What's happening seems to be the local web server (HFS, but tried
>> nginx and got the same result) serves up the first few icons and other
>> files fine, then runs out of ... something ... and starts returning
>> 503 (service unavailable). So half the icons end up display as red
>> crosses, the main map area remains blank, and the background sources
>> dropdown is unpopulated. Rebooting didn't help.
>>
>> (The setup I'm using, to be clear, is a local modified potlatch2.html
>> file that points to the live OSM server.)
>>
>> As I said, probably not a potlatch-specific problem (unless maybe some
>> kind of resource isn't being freed correctly), but it's causing me big
>> problems so if anyone has any ideas...
>
> I've seen this before on my own system. Simply put, flash makes
> another connection for everything it downloads. I solved it by upping
> the number of simultaneous clients permitted on the server, i.e.
>
> http://trac.openstreetmap.org/changeset/23044
>
> You'll probably need to do something similar, and/or investigate if
> your server has a total number of connections before bailing out. I'm
> slightly concerned that our current approach for icons (hundreds of
> individual files, hundreds of connections) might be, err, sub-optimal
> for production systems.
>

This will probably be browser specific -- flash uses the browser to
request further URLs. I think the debug player may do something
different, and probably broken.

Firefox and Chrome generally seem to reuse the connections, so on my
work PC wireshark shows about 3 or 4 connections over the start
period, with 2 TCP connections being used for most of the graphics on
a keep-alive basis. It probably would still be more efficient for us
to package them somehow though.

Note that the browser is also responsible for any gzip compression of
XML, and should handle this transparently to flash player, and in fact
does, at least on firefox, so all the bbox requests are compressed
already.

Dave