Malte Schröder wrote:
On Thu, 23 Oct 2008 01:10:58 +0200
Henrik Nordstrom <henrik@xxxxxxxxxxxxxxxxxxx> wrote:
On ons, 2008-10-22 at 11:31 +0200, Malte Schröder wrote:
Not normal.
Squid version?
3.0.STABLE10
And how did you measure these? You are not counting TIME_WAIT sockets
are you?
by getting cache_object://localhost/filedescriptors and checking
sockets in ESTABLISHED state.
I attached four files with the output of the filedescriptor page.
"start" is the state directly after the launch of squid, 1 is the state
after loading slashdot.org, 2 after reloading slashdot and 3 after
an additional reload. I increased pconn_timeout to 5 minutes so the
effect becomes more visible. The parent is a squid 2.6.STABLE20.
Okay, this is getting into stuff for squid-dev. cc'd.
If you are able build your own squid this patch will at least lead an
indication of whats going on with the idle FD.
Amos
--
Please use Squid 2.7.STABLE4 or 3.0.STABLE9
=== modified file 'src/pconn.cc'
--- src/pconn.cc 2008-07-12 15:11:10 +0000
+++ src/pconn.cc 2008-10-23 10:34:18 +0000
@@ -235,7 +235,7 @@
IdleConnList *list;
const char *aKey;
- LOCAL_ARRAY(char, desc, FD_DESC_SZ);
+ char desc[FD_DESC_SZ];
if (fdUsageHigh())
{
@@ -262,7 +262,7 @@
list->push(fd);
assert(!comm_has_incomplete_write(fd));
- snprintf(desc, FD_DESC_SZ, "%s idle connection", host);
+ snprintf(desc, FD_DESC_SZ, "idle connection for %s", aKey);
fd_note(fd, desc);
debugs(48, 3, "PconnPool::push: pushed FD " << fd << " for " << aKey);
}
