Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[stoatwblr]

I discovered and setup categories in the last day or so and have immediately run into this nasty problem.

Amule is now chewing 99% CPU most of the time and is unresponsive for hours at a time.

Restarting and removing the category definitions from amule.conf  returns to more normal behaviour, so it seems like it's specific to the category stuff.

A bit more experimenting seem to show that there's a "knee" point of about 8 categories when things break.

I have about 1500 downloads queued, MOST are stopped - only about 100 active sessions and about 2000 files shared out.

Turning up debugging shows nothing.

Any ideas?

[stoatwblr]

Same problem occurs with SVN.

[Stu Redman]

You might be running out of filehandles with 1500 downloads. That's really overdoing things a bit.

[stoatwblr]

As I said, only about 100 active, the rest are stopped and queued.

I have 32k filehandles set in ulimits, no filehandles errors manifesting.

I'm using Amule, as amuled is known to break with this many entries set.

This _only_ sems to  manifest under the following conditions:

More than 8 categories set (I set one category per TV series, trying to get them tossed into individual directories)

AND

Downloads active in more than 4 categories.

There's definitely some sort of problem in the category code and it's not confined to 2.2.4

 I've left the machine churning for 48 hours to see i it was some form of indexing/hashing issue but it's still just as bad at the end of that period and during this time no download/uploads have been allowed (ports are open, but EC, etc not responsive)

Why:

Most of this stuff is "rare" - inasmuch as 0-3 nodes sharing. It usually takes 3  months to get a complete movie or tv show season down  - mostly it's ancient SF stuff. The high number of shares is because I have the disk space and I figure it's better to seed the completed rare stuff back out for a couple of months after downloading when possible.

[Stu Redman]

As I said, only about 100 active, the rest are stopped and queued.

It still takes a filehandle for each download, no matter if stopped or active. Just for your information (I'm not saying that's related to your problem).

I'm using Amule, as amuled is known to break with this many entries set.

Definitely. 

More than 8 categories set (I set one category per TV series, trying to get them tossed into individual directories)
AND
Downloads active in more than 4 categories.

Thank you for the report. I'll look into the code if I find someting suspicious with the cats, but not before my vacation probably.

Does the view have any influence? Like when you have the networks tab open instead of the transfers tab? Or one specific cat instead of all downloads ? Have you tried disabling progress bars (which can be a CPU hog with many big files) ?

And (most important) - have you tried the SVN version ?

[stoatwblr]

I'm in no hurry on fixing this, but it'd be nice to have working

I'm currently using SVN 9565 with no change in the behaviour.

Views don't seem to make any difference.

I haven't tried switching off progress bars.

Last attempt:

I had everything stopped, tossed all the files into categories (~28) and then started going into the categories to start off downloads.  Attempting to start in category #5 got everything grinding to a halt.

Previously

With downloads active, tossing files into categories created on the fly the grind started at 7-8 created tabs.

[stoatwblr]

Update: switching off progress bars makes no difference.

[stoatwblr]

This appears fixed in SVN 9607 or earlier.

[Stu Redman]

I'm quite sure there was nothing changed regarding cats since 9565.

[stoatwblr]

Perhaps recent fixes revolving around crashes with more than 600 items queued have solved this too - or reduced the problem significantly.

[bobmarleyfan]

I had the same in emule on Windows when I used it.