On Saturday 24 February 2001 23:42, Ian wrote:
Hi Ian:
> > Weren't we going to release 0.3.7.1 - I have just got back from a
> conference and so I haven't been keeping up. Didn't we agree that there
> was a serious bug in 0.3.7 which was seriously damaging the network?
>
That's my story and I'm sticking to it.
Here's where I think things stand:
---
0) ThreadPool stall / freeze bug -- aka the bug that damages the network
This has been patched in the head branch. The change was minor and
completely understood. Just for paranoia's sake I added some
System.err.println's to my local tree so that I can monitor how the
ThreadPool is working. I have been keeping an eye on it for the last couple
of days. It looks good.
I am confident that this bug is definitively fixed.
After running for 24 hours my node (running w/o fproxy, see below) has 90
processes serving 44 inbound connections. This seems about right.
---
1) Thread leak concern.
There was some concern about the number of processes fred spawns.
Questions were raised about whether ThreadPool is leaking threads.
*ThreadPool is not leaking threads*.
Chris submitted a tiny patch to allow fproxy to be run in a
separate JVM (this behavior is busted in the head branch).
By running fproxy in its own JVM I was able to track down a really nasty
one thread per MSK read thread leak in the MapFile / SimplifiedClient.
I submitted a patch which while by no means elegant is probably acceptable.
This patch has not been reviewed / rejected or applied.
---
2) Anonymity filter triggered thread leak in fproxy.
I reported another thread leak bug to this list, which happen when
the anonymity filter throws exceptions.
I don't have a fix for this bug. But I think that it can be ignored for the
0.3.7.1 release. Until this is fixed, people who want to run the anonymity
filter can run fproxy in a separate JVM and periodically restart it.
The way fproxy handles (or doesn't handle) aborting requests is suspect and
bears further scrutiny. However, I do not want to touch the SimplifiedClient
tar baby...
---
3) Bugs in MapHandler.lookup
I traced the bug that you reported on February 13 to a race condition/ bad
error condition handling in MapHandler.lookup. I fixed this problem in my
local tree and mailed a patch to the list almost two weeks ago.
This patch has not been reviewed / rejected or applied.
---
Here's a list of things that I would like to see happen before the 0.3.7.1
release. I don't mean to come over the transom, but I have almost a half
a man week into debugging / patching at this point, and I would like to
see my efforts improve the quality of the code base.
Action Items before 0.3.7.1 release:
0) review / apply Chris' fix to ProxyServer
I think you have to comment out the existing:
listenPort = params.getint("listenPort", defaultListenPort);
line also.
1) review / apply my thread leak fix to MapFile
2) review / apply my earlier bugfix to MapHandler
Regards,
--gj
--
Web page inside Freenet:
freenet:MSK at SSK@enI8YFo3gj8UVh-Au0HpKMftf6QQAgE/homepage//
_______________________________________________
Devl mailing list
Devl at freenetproject.org
http://www.uprizer.com/mailman/listinfo/devl
