Quoting "Peter Ross" <peter.r...@bogen.in-berlin.de>:
Quoting "Scott Sipe" <csco...@gmail.com>:On Wed, Jul 6, 2011 at 4:21 AM, Peter Ross <peter.r...@bogen.in-berlin.de>wrote:Quoting "Peter Ross" <peter.r...@bogen.in-berlin.de**>: Quoting "Peter Ross" <peter.r...@bogen.in-berlin.de**>:Quoting "Jeremy Chadwick" <free...@jdc.parodius.com>:On Wed, Jul 06, 2011 at 01:54:12PM +1000, Peter Ross wrote:Quoting "Jeremy Chadwick" <free...@jdc.parodius.com>: On Wed, Jul 06, 2011 at 01:07:53PM +1000, Peter Ross wrote:Quoting "Jeremy Chadwick" <free...@jdc.parodius.com>: On Wed, Jul 06, 2011 at 12:23:39PM +1000, Peter Ross wrote:Quoting "Jeremy Chadwick" <free...@jdc.parodius.com>: On Tue, Jul 05, 2011 at 01:03:20PM -0400, Scott Sipe wrote:I'm running virtualbox 3.2.12_1 if that has anything to do with it. sysctl vfs.zfs.arc_max: 6200000000 While I'm trying to scp, kstat.zfs.misc.arcstats.size is hovering right around that value, sometimes above, sometimes below (that's as it should be, right?). I don't think that it dies when crossing over arc_max. I can run the same scp 10 times and it might fail 1-3 times, with no correlation to the arcstats.size being above/below arc_max that I can see. Scott On Jul 5, 2011, at 3:00 AM, Peter Ross wrote: Hi all,just as an addition: an upgrade to last Friday's FreeBSD-Stable and to VirtualBox 4.0.8 does not fix the problem. I will experiment a bit more tomorrow after hours and grabsome statistics.Regards Peter Quoting "Peter Ross" <peter.r...@bogen.in-berlin.de**>: Hi all,I noticed a similar problem last week. It is also very similar to one reported last year: http://lists.freebsd.org/**pipermail/freebsd-stable/2010-** September/058708.html<http://lists.freebsd.org/pipermail/freebsd-stable/2010-September/058708.html> My server is a Dell T410 server with the same bge card (the same pciconf -lvc output as described by Mahlon: http://lists.freebsd.org/**pipermail/freebsd-stable/2010-** September/058711.html<http://lists.freebsd.org/pipermail/freebsd-stable/2010-September/058711.html> Yours, Scott, is a em(4).. Another similarity: In all cases we are using VirtualBox. I just want to mention it, in case it matters. I am still running VirtualBox 3.2. Most of the time kstat.zfs.misc.arcstats.size was reaching vfs.zfs.arc_max then, but I could catch one or two cases then the value was still below. I added vfs.zfs.prefetch_disable=1 to sysctl.conf but itdoes not help.BTW: It looks as ARC only gives back the memory when I destroy the ZFS (a cloned snapshot containing virtual machines). Even if nothing happens for hours the buffer isn't released.. My machine was still running 8.2-PRERELEASE so I am upgrading. I am happy to give information gathered on old/new kernel if it helps. Regards Peter Quoting "Scott Sipe" <csco...@gmail.com>:On Jul 2, 2011, at 12:54 AM, jhell wrote: On Fri, Jul 01, 2011 at 03:22:32PM -0700, Jeremy Chadwickwrote:On Fri, Jul 01, 2011 at 03:13:17PM -0400, Scott Sipe wrote:I'm running 8.2-RELEASE and am having new problems with scp. When scping files to a ZFS directory on the FreeBSD server -- most notably large files -- the transfer frequently dies after just a few seconds. In my last test, I tried to scp an 800mb file to the FreeBSD system and the transfer died after 200mb. It completely copied the next 4 times I tried, and then died again on the next attempt. On the client side: "Connection to home closed by remote host. lost connection" In /var/log/auth.log: Jul 1 14:54:42 freebsd sshd[18955]: fatal: Write failed: Cannot allocate memory I've never seen this before and have used scp before to transfer large files without problems. This computer has been used in production for months and has a current uptime of 36 days. I have not been able to notice any problems copying files to the server via samba or netatalk, orany problems inapache.Uname: FreeBSD xeon 8.2-RELEASE FreeBSD 8.2-RELEASE #0: Sat Feb 19 01:02:54 EST 2011 root@xeon:/usr/obj/usr/src/**sys/GENERIC amd64 I've attached my dmesg and output of vmstat -z. I have not restarted the sshd daemon or rebooted the computer. Am glad to provide any other information or test anything else. {snip vmstat -z and dmesg}You didn't provide details about your networking setup (rc.conf, ifconfig -a, etc.). netstat -m would be useful too. Next, please see this thread circa September 2010, titled "Network memory allocation failures": http://lists.freebsd.org/**pipermail/freebsd-stable/2010-** September/thread.html#58708<http://lists.freebsd.org/pipermail/freebsd-stable/2010-September/thread.html#58708> The user in that thread is using rsync, which relies onscp by default.I believe this problem is similar, if not identical, to yours.Please also provide your output of ( /usr/bin/limits -a )for the serverend and the client.I am not quite sure I agree with the need for ifconfig -a but some information about the networking driver your using for the interface would be helpful, uptime of the boxes. And configurationof the pool.e.g. ( zpool status -a ;zfs get all <poolname> ) You should probablyprop this information up somewhere so you can reference byURL wheneverneeded.rsync(1) does not rely on scp(1) whatsoever but rsync(1)can be made touse ssh(1) instead of rsh(1) and I believe that is what Jeremy isstating here but correct me if I am wrong. It does use ssh(1) by default. Its a possiblity as well that if using tmpfs(5) or mdmfs(8) for /tmp type filesystems that rsync(1) may be just filling up yourtemp ram areaand causing the connection abort which would beexpected. ( df -h ) would help here.Hello, I'm not using tmpfs/mdmfs at all. The clients yesterday were 3 different OSX computers (over gigabit). The FreeBSD server has 12gb of ram and no bce adapter. For what it's worth, the server is backed up remotely every night with rsync (remote FreeBSD uses rsync to pull) to an offsite (slow cable connection) FreeBSD computer, and I have not seen any errors in the nightly rsync. Sorry for the omission of networking info, here's the output of the requested commands and some that popped up in the other thread: http://www.cap-press.com/misc/ In rc.conf: ifconfig_em1="inet 10.1.1.1 netmask 255.255.0.0" ScottJust to make it crystal clear to everyone: There is no correlation between this problem and use of ZFS. People are attempting to correlate "cannot allocate memory" messages with "anything on the system that uses memory". The VM is much more complex than that. Given the nature of this problem, it's much more likely the issue is "somewhere" within a networking layer within FreeBSD, whether it be driver-level or some sort of intermediary layer. Two people who have this issue in this thread are both using VirtualBox. Can one, or both, of you remove VirtualBox from the configuration entirely (kernel, etc. -- not sure what is required) and then see if the issue goes away?On the machine in question I only can do it after hours so I will do it tonight. I was _successfully_ sending the file over the loopback interface using cat /zpool/temp/zimbra_oldroot.vdi | ssh localhost "cat > /dev/null" I did it, btw, with the IPv6 localhost address first (accidently), and then using IPv4. Both worked. It always fails if I am sending it through the bce(4) interface, even if my target is the VirtualBox bridged to the bce card (so it does not "leave" the computer physically). Below the uname -a, ifconfig -a, netstat -rn, pciconf -lv and kldstat output. I have another box where I do not see that problem. It copies files happily over the net using ssh. It is an an older HP ML 150 with 3GB RAM only but with a bge(4) driver instead. It runs the same last week's RELENG_8. I installed VirtualBox and enabled vboxnet (so it loads the kernel modules). But I do not run VirtualBox on it (because it hasn't enough RAM). Regards Peter DellT410one# uname -a FreeBSD DellT410one.vv.fda 8.2-STABLE FreeBSD 8.2-STABLE #1: Thu Jun 30 17:07:18 EST 2011 r...@dellt410one.vv.fda:/usr/**obj/usr/src/sys/GENERIC amd64 DellT410one# ifconfig -a bce0: flags=8943<UP,BROADCAST,**RUNNING,PROMISC,SIMPLEX,** MULTICAST> metric 0 mtu 1500 options=c01bb<RXCSUM,TXCSUM,** VLAN_MTU,VLAN_HWTAGGING,JUMBO_**MTU,VLAN_HWCSUM,TSO4,VLAN_** HWTSO,LINKSTATE> ether 84:2b:2b:68:64:e4 inet 192.168.50.220 netmask 0xffffff00 broadcast 192.168.50.255 inet 192.168.50.221 netmask 0xffffff00 broadcast 192.168.50.255 inet 192.168.50.223 netmask 0xffffff00 broadcast 192.168.50.255 inet 192.168.50.224 netmask 0xffffff00 broadcast 192.168.50.255 inet 192.168.50.225 netmask 0xffffff00 broadcast 192.168.50.255 inet 192.168.50.226 netmask 0xffffff00 broadcast 192.168.50.255 inet 192.168.50.227 netmask 0xffffff00 broadcast 192.168.50.255 inet 192.168.50.219 netmask 0xffffff00 broadcast 192.168.50.255 media: Ethernet autoselect (1000baseT <full-duplex>) status: active bce1: flags=8802<BROADCAST,SIMPLEX,**MULTICAST> metric 0 mtu 1500 options=c01bb<RXCSUM,TXCSUM,** VLAN_MTU,VLAN_HWTAGGING,JUMBO_**MTU,VLAN_HWCSUM,TSO4,VLAN_** HWTSO,LINKSTATE> ether 84:2b:2b:68:64:e5 media: Ethernet autoselect lo0: flags=8049<UP,LOOPBACK,**RUNNING,MULTICAST> metric 0 mtu 16384 options=3<RXCSUM,TXCSUM> inet6 fe80::1%lo0 prefixlen 64 scopeid 0xb inet6 ::1 prefixlen 128 inet 127.0.0.1 netmask 0xff000000 nd6 options=3<PERFORMNUD,ACCEPT_**RTADV> vboxnet0: flags=8802<BROADCAST,SIMPLEX,**MULTICAST> metric 0 mtu 1500 ether 0a:00:27:00:00:00 DellT410one# netstat -rn Routing tables Internet: Destination Gateway Flags Refs Use Netif Expire default 192.168.50.201 UGS 0 52195 bce0 127.0.0.1 link#11 UH 0 6 lo0 192.168.50.0/24 link#1 U 0 1118212 bce0 192.168.50.219 link#1 UHS 0 9670 lo0 192.168.50.220 link#1 UHS 0 8347 lo0 192.168.50.221 link#1 UHS 0 103024 lo0 192.168.50.223 link#1 UHS 0 43614 lo0 192.168.50.224 link#1 UHS 0 8358 lo0 192.168.50.225 link#1 UHS 0 8438 lo0 192.168.50.226 link#1 UHS 0 8338 lo0 192.168.50.227 link#1 UHS 0 8333 lo0 192.168.165.0/24 192.168.50.200 UGS 0 3311 bce0 192.168.166.0/24 192.168.50.200 UGS 0 699 bce0 192.168.167.0/24 192.168.50.200 UGS 0 3012 bce0 192.168.168.0/24 192.168.50.200 UGS 0 552 bce0 Internet6: Destination Gateway Flags Netif Expire ::1 ::1 UH lo0 fe80::%lo0/64 link#11 U lo0 fe80::1%lo0 link#11 UHS lo0 ff01::%lo0/32 fe80::1%lo0 U lo0 ff02::%lo0/32 fe80::1%lo0 U lo0 DellT410one# kldstat Id Refs Address Size Name 1 19 0xffffffff80100000 dbf5d0 kernel 2 3 0xffffffff80ec0000 4c358 vboxdrv.ko 3 1 0xffffffff81012000 131998 zfs.ko 4 1 0xffffffff81144000 1ff1 opensolaris.ko 5 2 0xffffffff81146000 2940 vboxnetflt.ko 6 2 0xffffffff81149000 8e38 netgraph.ko 7 1 0xffffffff81152000 153c ng_ether.ko 8 1 0xffffffff81154000 e70 vboxnetadp.ko DellT410one# pciconf -lv .. bce0@pci0:1:0:0: class=0x020000 card=0x028d1028 chip=0x163b14e4 rev=0x20 hdr=0x00 vendor = 'Broadcom Corporation' class = network subclass = ethernet bce1@pci0:1:0:1: class=0x020000 card=0x028d1028 chip=0x163b14e4 rev=0x20 hdr=0x00 vendor = 'Broadcom Corporation' class = network subclass = ethernetCould you please provide "pciconf -lvcb" output instead, specific to the bce chips? Thanks.Her it is: bce0@pci0:1:0:0: class=0x020000 card=0x028d1028 chip=0x163b14e4 rev=0x20 hdr=0x00 vendor = 'Broadcom Corporation' class = network subclass = ethernet bar [10] = type Memory, range 64, base 0xda000000, size 33554432, enabled cap 01[48] = powerspec 3 supports D0 D3 current D0 cap 03[50] = VPD cap 05[58] = MSI supports 16 messages, 64 bit enabled with 1 message cap 11[a0] = MSI-X supports 9 messages in map 0x10 cap 10[ac] = PCI-Express 2 endpoint max data 256(512) link x4(x4) ecap 0003[100] = Serial 1 842b2bfffe6864e4 ecap 0001[110] = AER 1 0 fatal 0 non-fatal 1 corrected ecap 0004[150] = unknown 1 ecap 0002[160] = VC 1 max VC0Thanks Peter. Adding Yong-Hyeon and David to the discussion, since they've both worked on the bce(4) driver in recent months (most of the changes made recently are only in HEAD), and also adding Jack Vogel of Intel who maintains em(4). Brief history for the devs: The issue is described "Network memory allocation failures" and was reported last year, but two users recently (Scott and Peter) have reported the issue again: http://lists.freebsd.org/**pipermail/freebsd-stable/2010-** September/thread.html#58708<http://lists.freebsd.org/pipermail/freebsd-stable/2010-September/thread.html#58708> And was mentioned again by Scott here, which also contains some technical details: http://lists.freebsd.org/**pipermail/freebsd-stable/2011-** July/063172.html<http://lists.freebsd.org/pipermail/freebsd-stable/2011-July/063172.html> What's interesting is that Scott's issue is identical in form but he's using em(4), which isn't known to behave like this. Both individuals are using VirtualBox, though we're not sure at this point if that is the piece which is causing the anomaly. Relevant details of Scott's system (em-based): http://www.cap-press.com/misc/ Relevant details of Peter's system (bce-based): http://lists.freebsd.org/**pipermail/freebsd-stable/2011-** July/063221.html<http://lists.freebsd.org/pipermail/freebsd-stable/2011-July/063221.html> http://lists.freebsd.org/**pipermail/freebsd-stable/2011-** July/063223.html<http://lists.freebsd.org/pipermail/freebsd-stable/2011-July/063223.html> I think the biggest complexity right now is figuring out how/why scp fails intermittently in this nature. The errno probably "trickles down" to userland from the kernel, but the condition regarding why it happens is unknown.BTW: I also saw 2 of the errors coming from a BIND9 running in a jail on that box. DellT410one# fgrep -i allocate /jails/bind/20110315/var/log/**messages Apr 13 05:17:41 bind named[23534]: internal_send: 192.168.50.145#65176: Cannot allocate memory Jun 21 23:30:44 bind named[39864]: internal_send: 192.168.50.251#36155: Cannot allocate memory Jun 24 15:28:00 bind named[39864]: internal_send: 192.168.50.251#28651: Cannot allocate memory Jun 28 12:57:52 bind named[2462]: internal_send: 192.168.165.154#1201: Cannot allocate memory My initial guess: it happens sooner or later somehow - whether it is a lot of traffic in one go (ssh/scp copies of virtual disks) or a lot of traffic over a longer period (a nameserver gets asked again and again).Scott, are you also using jails? If both of you are: is there any possibility you can remove use of those? I'm not sure how VirtualBox fits into the picture (jails + VirtualBox that is), but I can imagine jails having different environmental constraints that might cause this. Basically the troubleshooting process here is to remove pieces of the puzzle until you figure out which piece is causing the issue. I don't want to get the NIC driver devs all spun up for something that, for example, might be an issue with the jail implementation.I understand this. As said, I do some afterhours debugging tonight. The scp/ssh problems are happening _outside_ the jails. The bind runs _inside_ the jail. I wanted to use the _host_ system to send VirtualBox virtual disks and filesystems used by jails to archive them and/or having them available on other FreeBSD systems (as a cold standby solution).I just switched off the VirtualBox (without removing the kernel modules). The copy succeeds now. Well, it could be a VirtualBox related problem, or is the server just relieved to have 2GB more memory at hands now? Do you have a quick idea to "emulate" the 2GB memory load usually delivered by VirtualBox?Well, managed that (using lookbusy) Interestingly I could copy a large file (30GB) without problems, as soon asI switched off the VirtualBox. As said, the kernel modules weren't unloaded,they are still there. The copy crashes seconds after I started the VirtualBox. According to vmstat and top I had more free memory (ca. 1.5GB) as I had without VirtualBox and lookbusy (ca. 350MB). So, it looks (to me, at least) as I have a VirtualBox related problem, somehow. Any ideas? I am happy to play a bit more to get it sorted although it has some limits (it is running the company mailserver, after all) Regards PeterThis is it -- I'm seeing the exact same thing. Scp dies reliably with VirtualBox running. Quit VirtualBox and I was able to scp about 30 large files with no errors. Once I started VirtualBox an in-progress scp died within seconds. Ditto that the Kernel modules merely being loaded don't seem to make a difference, it's VirtualBox actually running. virtualbox-ose-3.2.12_1Hi, I wonder whether anyone has new ideas.I am puzzled that it happens when VirtualBoxes are running, while the load or unload of the VirtualBox kernel modules doesn't seem to have an effect.Should I describe the case at the -emulation mailing list to get some ideas from the engineers working on VirtualBox?
I finally did that, and, especially thanks to Adam Vande More, have a solution http://lists.freebsd.org/pipermail/freebsd-emulation/2011-July/008977.html I put net.graph.maxdata=65536 in my /boot/loader.conf.Before I could see that the write error increased the "NetGraph data items" failures (seen via vmstat -z).
In one way it makes sense that the _start_ of the VirtualBox makes the difference. It is a busy company mailserver with SMTP and HTTP access and a lot of traffic going through - it all has to go through the netgraph system.Is this assumption false/true? Is the tuning a fix or just a workaround? Should it be noted somewhere, e.g. as a package message when installing VirtualBox?
I am not the only one bitten by it, and I would like to avoid disappointment for others in the future, or just remarks as "FreeBSD virtualization sucks" based on this problem. (E.g. Marlon was running out of time and had to head down the Citrix path).
Regards Peter _______________________________________________ freebsd-stable@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-stable To unsubscribe, send any mail to "freebsd-stable-unsubscr...@freebsd.org"
