From owner-freebsd-net@FreeBSD.ORG Mon Apr 11 06:54:33 2011 Return-Path: Delivered-To: freebsd-net@freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2001:4f8:fff6::34]) by hub.freebsd.org (Postfix) with ESMTP id 5B161106564A; Mon, 11 Apr 2011 06:54:33 +0000 (UTC) (envelope-from przemyslaw@frasunek.com) Received: from lagoon.freebsd.lublin.pl (lagoon.freebsd.lublin.pl [IPv6:2a02:2928:a::3]) by mx1.freebsd.org (Postfix) with ESMTP id D86A58FC1D; Mon, 11 Apr 2011 06:54:32 +0000 (UTC) Received: from [IPv6:2a02:2928:a:ffff:965:9c23:fece:cec1] (unknown [IPv6:2a02:2928:a:ffff:965:9c23:fece:cec1]) by lagoon.freebsd.lublin.pl (Postfix) with ESMTPSA id EE91C23944A; Mon, 11 Apr 2011 08:54:31 +0200 (CEST) Message-ID: <4DA2A5AA.4060802@frasunek.com> Date: Mon, 11 Apr 2011 08:54:34 +0200 From: Przemyslaw Frasunek Organization: frasunek.com User-Agent: Mozilla/5.0 (Windows; U; Windows NT 6.1; pl; rv:1.9.2.15) Gecko/20110303 Thunderbird/3.1.9 MIME-Version: 1.0 To: Gleb Smirnoff , Eugene Grosbein , freebsd-net@freebsd.org References: <4D947756.6050808@freebsd.lublin.pl> <4D9F6C71.1040209@frasunek.com> <4DA171BA.9000507@frasunek.com> <4DA1E39C.9090300@rdtc.ru> <4DA23090.8060206@frasunek.com> <20110411054932.GU84445@FreeBSD.org> In-Reply-To: <20110411054932.GU84445@FreeBSD.org> X-Enigmail-Version: 1.1.1 Content-Type: text/plain; charset=KOI8-R Content-Transfer-Encoding: 7bit Cc: Subject: Re: mpd5/Netgraph issues after upgrading to 7.4 X-BeenThere: freebsd-net@freebsd.org X-Mailman-Version: 2.1.5 Precedence: list List-Id: Networking and TCP/IP with FreeBSD List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Mon, 11 Apr 2011 06:54:33 -0000 > IMO, any kind of memory allocation code (malloc, uma, netgraph item > allocator) never return EPERM, they return ENOMEM or ENOBUFS. > > So, there is a bug somewhere else. I think so, but for me it still looks like resource shortage. As I wrote before, when EPERM starts appearing, I'm unable to run "ngctl list". I observed that on mpd boxes I have very large number of failed allocations from zone "128 Bucket". On other boxes with the same release (7.4) and the same amount of physical memory (2 GB), but without mpd5, this value is almost zero. Also other zones has non-zero number of failures: [root@kalina-gw ~]# vmstat -z | grep -v ' 0$' ITEM SIZE LIMIT USED FREE REQUESTS FAILURES UMA Slabs: 64, 0, 3172, 427, 9555, 21 64 Bucket: 268, 0, 199, 39, 467, 45 128 Bucket: 524, 0, 558, 121, 13258, 64298286 64: 64, 0, 18525, 1889, 328077543, 8 128: 128, 0, 16822, 1538, 1718123, 31 mbuf: 256, 0, 3, 7130, 282690299, 65 udp_inpcb: 180, 25608, 348, 554, 5052711, 3 (above is on mpd box after 8 hours of uptime) [root@pkp-gw ~]# vmstat -z | grep -v ' 0$' ITEM SIZE LIMIT USED FREE REQUESTS FAILURES 64 Bucket: 268, 0, 193, 87, 514, 53 128 Bucket: 524, 0, 7621, 2, 3003617, 1225 pfsrctrpl: 124, 10013, 0, 1426, 14172484, 2979678 (and this is on non-mpd box after 355 days of uptime) I'm going to try to track down origin of EPERMs by adding some debug printfs in netgraph code.