>
> - if (fr.isFinished()) {
> + if (fr.isFinished() || fr.hasData() || fr.failed != null) {
>
> Is it possible for fr.isFinished to be false when fr.hasData is true or
> fr.failed is non-null? (I'm not saying take it out, but is isFinished() buggy
> or am I missing something?)
It solved most of the stalled at 100% bugs for me, so I suspect the
finished has some bugs.
>
> If you think the code is working, you should post a jar somewhere and ask for
> wider testing on devl.
>
> Blocks seems to get reset to 0 / 0 after a while at 100%, any ideas why?
I didn't experienced it yet. Will test further.
>
> "ETA: -9s" is probably not what we want. :)
It happens when the start of the file progresses fast, then slows
down, eg if it's in the store. Well, it should stop at 0:)
>
> And I did get one page to stall. It was in the datastore and loaded instantly
> when I clicked on the url and pressed enter, but until that point it went
> nearly 3 minutes without progress, showing 0 / 1 blocks and updating only the
> time counter.
>
> The URL was this, but I doubt it's a problem with the page.
> http://127.0.0.1:8888/freenet:u...@mjbn3mnx5j5xoa-osoccjtwrvtwj5r52wn4xtkw64yq,NYAzL61qMPrAYC8B7Anpwx8h7CPYzBnpJIZNVgnR1rU,AQACAAE/Chaosradio/43/ChaosradioExpress/index.html
>
Will investigate it too. Don't think it is page specific though.
By the way, I noticed something interesting. At FProxyFetchTracker 's
run() method, if I'm not mistaking, it cancels all cancellable
fetches. If it is, then listening for an already removed Progress
object makes no event. Is this the way it should work?
> I had a different problem with another page, a file stalled close to the end,
> showing no new blocks, no failures, but when I reloaded it showed a MIME type
> warning page instantly. I'm not sure what's going on there, it might be some
> wierd interaction with the FProxy fetch code. Nothing obvious in the logs
> (NPEs etc).
>
> Arguably we should show the mime type warning as soon as we know the mime
> type, but it may be best to keep the current behaviour for now.
Need to think about this one.
>
> If a request is being answered from the datastore, is it possible that the
> events from that one request could overwhelm the events from the other
> requests, and thus that they might not show progress when there was progress?
> You might want to throttle events on the node side so we don't send more than
> 1 per 100ms from each request, or something?
>
No, it is fixed now. If an element already generated an event, it
won't generate another one till the client grabs the first one. Every
notification needs 10-30ms to deliver, and the connection sharing
checks 100ms and has a bucket of 10messages. So it is not possible to
lose an event.
_______________________________________________
Devl mailing list
[email protected]
http://emu.freenetproject.org/cgi-bin/mailman/listinfo/devl
