Strange memory leak

Andrew MacLennan apm at connect.com.au
Tue Dec 21 06:39:59 UTC 2004


A little over two months ago I upgraded our news architecture to newer
hardware (2 x Sun V240s, 2Gb RAM, Solaris 8) and INN 2.4.1

After cutting them across to live operation we noticed a memory leak,
chewing up memory through the kernel and not releasing it.  A reboot
once a week was the only way to free it up again (and hardly an elegant
solution)

Patch clusters were checked, then a process of elimination began with
INN functions.  When I switched "noreader" to true in inn.conf the
memory problems ceased.

Relevant config from inn.conf (with NNRP enabled):
---
#Reading                                                                                                              
allownewnews:           true
articlemmap:            false
clienttimeout:          600
initialtimeout:         10
msgidcachesize:         10000
nnrpdcheckart:          true
noreader:               false
readerswhenstopped:     false
readertrack:            false
nfsreader:              false
nfsreaderdelay:         60
tradindexedmmap:        true
nnrpdloadlimit:         16
---

And from readers.conf:
---
auth "localhost" {
    hosts: "localhost, 127.0.0.1, stdin"
    default: "<localhost>"
}
 
access "localhost" {
    users: "<localhost>"
    newsgroups: "*"
    access: RPA
}

access "connect-readers" {
    users:              "connect-reader"
    newsgroups:         "*,!control*"
}
                                                                                                             auth "connect-office" {
    hosts:             
"*.off.connect.com.au,210.8.131.*,210.9.8.*,*.bur.connect.com.au"
    default:            "connect-reader"
}

auth "connect-dialup" {
    hosts:             
"*.connect.net.au,*.interconnect.com.au,*.enternet.com.au,*.roaming.connect.com.au"
    default:            "connect-reader"
}
---

Followed by a collection of more auth groups that permit different host
definitions (by IP or subnet) that default to "connect-reader" for their
access.

The memory usage continued to grow even when there was minimal usage of
NNRP on the host in question, and it wasn't until it was disabled
completely that the situation disappeared. Is there something I missed? 
Does shunting all of the auth groups through the one access group cause
any problems?  Has anyone else seen similar problems?

Thanks,
Andrew
-- 
Andrew MacLennan     apm at connect.com.au     +61-3-8686-2305
Systems Administrator
AAPT Limited                         http://www.aapt.com.au
180-188 Burnley St   Richmond  VIC  3121  Australia




More information about the inn-workers mailing list