Op vr 07-11-2003, om 16:25 schreef Jonas Sonntag:
> 2. Fileinfo tab is not updated correctly:
> 
> 2.1 available sources are not displayed correctly: a source is not added 
> to the available sources count in certain states (e.g. when the source 
> is in state 'Push sent' if i remeber it right)

As far as I understood, this also happens with the GTK1 frontend.


> 4. maybe this is known allready: i can overwrite max_ultrapeers = 4 with 
> up_connections when in ultrapeer mode

max-ultrapeers is a leaf setting only. Frontends should probably not
display this setting when in non leaf mode.

> 5. not a bug, but i noticed that _a lot_ of clients are removed from 
> parq again and again because they are re-requesting files too fast. it 
> looks like we are removing 90% of all parq slots because the clients 
> don't behave like they should. these are not only the usual suspects 
> lime/bear but also gtkg clients. now i'm sure this is meant as a good 
> thing in theory, but it makes it practically impossible for most of the 
> clients to get 10 files from my share. they get 2 upload slots, the 
> other 8 are queued and taken to the end of the queue over and over again 
> so they will never get the requested files. maybe gtkg is a bit too 
> conservative here?

All clients which do not respect the Retry-After header (the delay after
which a client may retry its download attempt) will most probably be
banned. Unless the retry rate is high (20 minutes or more).
Of course this can happen with GTK-G clients too: That is when a user
manually forces a retry.
GTKG is rather polite (I think)
1st attempt: You get queued
2nd (too early attempt): You get a warning, and not the full PARQ
headers are sent, however you are still queued
3rd (too early attempt): Now you are removed from the PARQ (so you are
back to the end of the queue)
4th (too early attempt): Then you have done it, you are banned for a
while. All download requests are simply ignored. And it is treated as a
hammering attempt.

As you see, you can always retry 1 time to early, which has no
consequences but a warning only.


> 6. again parq: i noticed that even when there are slots available, all 
> download requests are queued first. sometimes they get something like 
> 'try again in 0s' reply. fact is i have never seen all of my 10 upload 
> slots occupied even when there are 20+ clients queued. i haven't thought 
> of a theory yet, but it definitly looks a bit strange.

Depends on how you look at it. The logic in PARQ actually allows this
behaviour (on purpose). For example, when the retry rate is 20 minutes,
and no-one else before you wants that file anymore, you can assume you
actually have a spare slot for 20 minutes until that client retries
again. Unless they implement PARQ, because as soon as a position is
available a signal is sent to the client that it may attempt to download
immediatly.


> 8. and finally a 'bug' that is more like user stupidity ;P .. but maybe 
> could be worked around: my temp directory is a subdirectory within one 
> of my shared directories. when i reshare and there are active downloads 
> gtkg will loop with 'file xxx has changed while calculating sha1' as 
> long as the download is active. maybe it could notice that this is a 
> partial file and skip it? it should be shared anyway by pfs i think?

Well it won't share the file :), ok, so the warnings are anoying.
However it detects that it is a PFS file and will only share it as such.

-- 
Jeroen Asselman <[EMAIL PROTECTED]>

Attachment: signature.asc
Description: Dit berichtdeel is digitaal ondertekend

Reply via email to