summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorantrik <antrik@users.sf.net>2011-04-09 23:36:08 +0200
committerantrik <antrik@users.sf.net>2011-04-10 01:02:51 +0200
commitd612b8e3f329ff22fec1b81f4c237faa6773dab6 (patch)
treec78674ddcbc07702f62ccbfabb60816f3ecb4c20
parentbf7c9659d7f1be99e1692c609160212e5a94fbc8 (diff)
open_issues/pfinet: Properly describe the thread explosion problem
-rw-r--r--open_issues/pfinet.mdwn17
1 files changed, 13 insertions, 4 deletions
diff --git a/open_issues/pfinet.mdwn b/open_issues/pfinet.mdwn
index 8782fe08..7aadd736 100644
--- a/open_issues/pfinet.mdwn
+++ b/open_issues/pfinet.mdwn
@@ -10,9 +10,18 @@ License|/fdl]]."]]"""]]
[[!tag open_issue_hurd]]
-IRC, #hurd
+In certain situations, pfinet spawns more and more threads,
+apparently without any bounds.
-pfinet explosion
+The thread creation happens in bursts rather than continuously.
+According to a backtrace in GDB,
+all the threads are functional and waiting for client requests.
+(The bursts are getting smaller as the number of threads rises,
+but probably only because the enormous number of existing threads
+slows down processing in general.)
- <antrik> I reproduce it with freeciv client connected to a remote X server
- <antrik> it suffices to run freeciv-gtk2, and clicking "new game"
+This can be triggered quite reliably by X clients running on the Hurd system,
+connected to an X server on another machine over TCP,
+and transferring fairly large amounts of data.
+The easiest way to reproduce it I found is launching freeciv-gtk2,
+pressing the "new game" button, and then simply waiting for a while.