Hi, I have configured two machines for testing gfs filesystems. They are attached to a iscsi device and centos versions are: CentOS release 5.4 (Final) Linux node1.fib.upc.es 2.6.18-164.el5 #1 SMP Thu Sep 3 03:33:56 EDT 2009 i686 i686 i386 GNU/Linux The problem is if I try to mount a gfs partition it hangs. [root at node2 ~]# cman_tool status Version: 6.2.0 Config Version: 29 Cluster Name: gfs-test Cluster Id: 25790 Cluster Member: Yes Cluster Generation: 4156 Membership state: Cluster-Member Nodes: 2 Expected votes: 2 Quorum device votes: 2 Total votes: 4 Quorum: 3 Active subsystems: 9 Flags: Ports Bound: 0 Node name: node2.fib.upc.es Node ID: 2 Multicast addresses: 239.192.100.35 Node addresses: 147.83.41.130 [root at node2 ~]# cman_tool nodes Node Sts Inc Joined Name 0 M 0 2010-03-24 14:46:22 /dev/web/web 1 M 4156 2010-03-24 17:08:36 node1.fib.upc.es 2 M 4132 2010-03-24 14:46:09 node2.fib.upc.es [root at node2 ~]# group_tool hangs... [root at node1 ~]# mount -t gfs /dev/home2/home2 /home2 hangs... If I cancel the command I can return to the terminal and I don't see anything in log files. The resource /dev/home2/home2 is accessible by the two nodes and if I try to mount /home2 with lock_nolock there is no problem. cluster.conf is: <?xml version="1.0"?> <cluster alias="gfs-test" config_version="29" name="gfs-test"> <quorumd device="/dev/web/web" interval="1" min_score="1" tko="10" votes="2"> <heuristic interval="10" program="/bin/ping -t1 -c1 147.83.41.1" score="1"/> </quorumd> <fence_daemon post_fail_delay="0" post_join_delay="3"/> <clusternodes> <clusternode name="node1.fib.upc.es" nodeid="1" votes="1"> <fence> <method name="1"> <device name="gfs-test" nodename="node1.fib.upc.es"/> </method> </fence> </clusternode> <clusternode name="node2.fib.upc.es" nodeid="2" votes="1"> <fence> <method name="1"> <device name="gfs-test" nodename="node2.fib.upc.es"/> </method> </fence> </clusternode> </clusternodes> <cman/> <fencedevices> <fencedevice agent="fence_manual" name="gfs-test"/> </fencedevices> <rm> <resources> <clusterfs device="/dev/home2/home2" force_unmount="0" fsid="1605" fstype="gfs" mountpoint="/home2" name="home alumnes" options=""/> </resources> </rm> </cluster> Any help will be welcomed. Thanks, Sandra
From: sandra-llistes <sandra-llistes at fib.upc.edu>> The problem is if I try to mount a gfs > partition it hangs.No gfs expert here but, did you try mount '-vv' option? If it does not help, maybe an strace would... JD
Hi, I'm not a GFS expert either :-) In fact, I erased all the configuration of the cluster and try to do it again from scratch with luci/ricci configuration tools (perhaps I did something wrong last time) mount -vv gives the following information: [root at node1 gfs]# mount -t gfs -vv /dev/home2/home2 /home2 /sbin/mount.gfs: mount /dev/mapper/home2-home2 /home2 /sbin/mount.gfs: parse_opts: opts = "rw" /sbin/mount.gfs: clear flag 1 for "rw", flags = 0 /sbin/mount.gfs: parse_opts: flags = 0 /sbin/mount.gfs: parse_opts: extra = "" /sbin/mount.gfs: parse_opts: hostdata = "" /sbin/mount.gfs: parse_opts: lockproto = "" /sbin/mount.gfs: parse_opts: locktable = "" /sbin/mount.gfs: message to gfs_controld: asking to join mountgroup: /sbin/mount.gfs: write "join /home2 gfs lock_dlm gfs-test:gfs-data rw /dev/mapper/home2-home2" ... And hangs at that point (On the other node it happens the same) I tried it turning off the local firewalls on the nodes and they reached each other without problem with pings. Also, there are no more firewalls between them. The new configuration is more simple: [root at node1 gfs]# more /etc/cluster/cluster.conf <?xml version="1.0"?> <cluster alias="gfs-test" config_version="6" name="gfs-test"> <fence_daemon clean_start="0" post_fail_delay="0" post_join_delay="3"/> <clusternodes> <clusternode name="node1.fib.upc.es" nodeid="1" votes="1"> <fence> <method name="1"> <device name="test" nodename="node1.fib.upc.es"/> </method> </fence> </clusternode> <clusternode name="node2.fib.upc.es" nodeid="2" votes="1"> <fence> <method name="1"> <device name="test" nodename="node2.fib.upc.es"/> <device name="test" nodename="node2.fib.upc.es"/> </method> </fence> </clusternode> </clusternodes> <cman expected_votes="1" two_node="1"/> <fencedevices> <fencedevice agent="fence_manual" name="test"/> </fencedevices> <rm> <failoverdomains/> <resources> <clusterfs device="/dev/home2/home2" force_unmount="0" fsid="3280" fstype="gfs" mountpoint="/home2" name="home" self_fence="0"/> </resources> </rm> </cluster> Finally, I reformatted /dev/home2/home2 with the following command that gave no errors but it doesn't affect the final result: gfs_mkfs -O -j 3 -p lock_dlm -t gfs-test:gfs-data /dev/home2/home2 Thanks, Sandra PD: I append an strace but I can't see anything useful. -------------- next part -------------- An embedded and charset-unspecified text was scrubbed... Name: strace.txt URL: <lists.centos.org/pipermail/centos/attachments/20100326/81071eca/attachment-0002.txt>
It seems that all the problem was caused by selinux. I booted all the nodes with the kernel option selinux=0 and now I can mount GFS partitions without problems. Thanks, Sandra sandra-llistes wrote:> Hi, > > I'm not a GFS expert either :-) > In fact, I erased all the configuration of the cluster and try to do it > again from scratch with luci/ricci configuration tools (perhaps I did > something wrong last time) > > mount -vv gives the following information: > [root at node1 gfs]# mount -t gfs -vv /dev/home2/home2 /home2 > /sbin/mount.gfs: mount /dev/mapper/home2-home2 /home2 > /sbin/mount.gfs: parse_opts: opts = "rw" > /sbin/mount.gfs: clear flag 1 for "rw", flags = 0 > /sbin/mount.gfs: parse_opts: flags = 0 > /sbin/mount.gfs: parse_opts: extra = "" > /sbin/mount.gfs: parse_opts: hostdata = "" > /sbin/mount.gfs: parse_opts: lockproto = "" > /sbin/mount.gfs: parse_opts: locktable = "" > /sbin/mount.gfs: message to gfs_controld: asking to join mountgroup: > /sbin/mount.gfs: write "join /home2 gfs lock_dlm gfs-test:gfs-data rw > /dev/mapper/home2-home2" > ... > > And hangs at that point (On the other node it happens the same) > > I tried it turning off the local firewalls on the nodes and they reached > each other without problem with pings. Also, there are no more firewalls > between them. > > The new configuration is more simple: > [root at node1 gfs]# more /etc/cluster/cluster.conf > <?xml version="1.0"?> > <cluster alias="gfs-test" config_version="6" name="gfs-test"> > <fence_daemon clean_start="0" post_fail_delay="0" > post_join_delay="3"/> > <clusternodes> > <clusternode name="node1.fib.upc.es" nodeid="1" votes="1"> > <fence> > <method name="1"> > <device name="test" > nodename="node1.fib.upc.es"/> > </method> > </fence> > </clusternode> > <clusternode name="node2.fib.upc.es" nodeid="2" votes="1"> > <fence> > <method name="1"> > <device name="test" > nodename="node2.fib.upc.es"/> > <device name="test" > nodename="node2.fib.upc.es"/> > </method> > </fence> > </clusternode> > </clusternodes> > <cman expected_votes="1" two_node="1"/> > <fencedevices> > <fencedevice agent="fence_manual" name="test"/> > </fencedevices> > <rm> > <failoverdomains/> > <resources> > <clusterfs device="/dev/home2/home2" > force_unmount="0" fsid="3280" fstype="gfs" mountpoint="/home2" > name="home" self_fence="0"/> > </resources> > </rm> > </cluster> > > Finally, I reformatted /dev/home2/home2 with the following command that > gave no errors but it doesn't affect the final result: > gfs_mkfs -O -j 3 -p lock_dlm -t gfs-test:gfs-data /dev/home2/home2 > > Thanks, > > Sandra > > PD: I append an strace but I can't see anything useful.