On Tue, Dec 16, 2014 at 08:48:48AM -0600, Chris Adams wrote:

I have a oVirt setup that has three nodes, all running CentOS 7, with a hosted engine running CentOS 6. Two of the nodes (node8 and node9) are configured for hosted engine, and the third (node2) is just a "regular" node (as you might guess from the names, more nodes are coming as I migrate VMs to oVirt).

On one node, node8, vdsm periodically loses its connection to libvirt, which causes vdsm to restart. There doesn't appear to be any trigger that I can see (not time of day, load, etc. related). The engine VM is up and running on node8 (don't know if that has anything to do with it).

I get some entries in /var/log/messages repeated continuously; the "ovirt-ha-broker: sending ioctl 5401 to a partition" I mentioned before, and the following:

Dec 15 20:56:23 node8 journal: User record for user '107' was not found: No such file or directory Dec 15 20:56:23 node8 journal: Group record for user '107' was not found: No such file or directory

I don't think those have any relevance (don't know where they come from); filtering those out, I see:

Dec 15 20:56:33 node8 journal: End of file while reading data: Input/output error Dec 15 20:56:33 node8 journal: Tried to close invalid fd 0 Dec 15 20:56:38 node8 journal: vdsm root WARNING connection to libvirt broken. ecode: 1 edom: 7 Dec 15 20:56:38 node8 journal: vdsm root CRITICAL taking calling process down. Dec 15 20:56:38 node8 journal: vdsm vds ERROR libvirt error Dec 15 20:56:38 node8 journal: ovirt-ha-broker mgmt_bridge.MgmtBridge ERROR Failed to getVdsCapabilities: Error 16 from getVdsCapabilities: Unexpected exception Dec 15 20:56:45 node8 journal: End of file while reading data: Input/output error Dec 15 20:56:45 node8 vdsmd_init_common.sh: vdsm: Running run_final_hooks Dec 15 20:56:45 node8 systemd: Starting Virtual Desktop Server Manager... <and then all the normal-looking vdsm startup>

It is happening about once a day, but not at any regular interval or time (was 02:23 Sunday, then 20:56 Monday).

vdsm.log has this at that time:

Thread-601576::DEBUG::2014-12-15 20:56:38,715::BindingXMLRPC::1132::vds::(wrapper) client [127.0.0.1]::call getCapabilities with () {} Thread-601576::DEBUG::2014-12-15 20:56:38,718::utils::738::root::(execCmd) /sbin/ip route show to 0.0.0.0/0 table all (cwd None) Thread-601576::DEBUG::2014-12-15 20:56:38,746::utils::758::root::(execCmd) SUCCESS: <err> = ''; <rc> = 0 Thread-601576::WARNING::2014-12-15 20:56:38,754::libvirtconnection::135::root::(wrapper) connection to libvirt broken. ecode: 1 edom: 7 Thread-601576::CRITICAL::2014-12-15 20:56:38,754::libvirtconnection::137::root::(wrapper) taking calling process down. MainThread::DEBUG::2014-12-15 20:56:38,754::vdsm::58::vds::(sigtermHandler) Received signal 15 Thread-601576::DEBUG::2014-12-15 20:56:38,755::libvirtconnection::143::root::(wrapper) Unknown libvirterror: ecode: 1 edom: 7 level: 2 message: internal error: client socket is closed MainThread::DEBUG::2014-12-15 20:56:38,755::protocoldetector::135::vds.MultiProtocolAcceptor::(stop) Stopping Acceptor MainThread::INFO::2014-12-15 20:56:38,755::__init__::563::jsonrpc.JsonRpcServer::(stop) Stopping JsonRPC Server Detector thread::DEBUG::2014-12-15 20:56:38,756::protocoldetector::106::vds.MultiProtocolAcceptor::(_cleanup) Cleaning Acceptor MainThread::INFO::2014-12-15 20:56:38,757::vmchannels::188::vds::(stop) VM channels listener was stopped. MainThread::INFO::2014-12-15 20:56:38,758::momIF::91::MOM::(stop) Shutting down MOM MainThread::DEBUG::2014-12-15 20:56:38,759::task::595::Storage.TaskManager.Task::(_updateState) Task=`26c7680c-23e2-42bb-964c-272e778a168a`::moving from state init -> state preparing MainThread::INFO::2014-12-15 20:56:38,759::logUtils::44::dispatcher::(wrapper) Run and protect: prepareForShutdown(options=None) Thread-601576::ERROR::2014-12-15 20:56:38,755::BindingXMLRPC::1142::vds::(wrapper) libvirt error Traceback (most recent call last): File "/usr/share/vdsm/rpc/BindingXMLRPC.py", line 1135, in wrapper res = f(*args, **kwargs) File "/usr/share/vdsm/rpc/BindingXMLRPC.py", line 463, in getCapabilities ret = api.getCapabilities() File "/usr/share/vdsm/API.py", line 1245, in getCapabilities c = caps.get() File "/usr/share/vdsm/caps.py", line 615, in get caps.update(netinfo.get()) File "/usr/lib/python2.7/site-packages/vdsm/netinfo.py", line 812, in get nets = networks() File "/usr/lib/python2.7/site-packages/vdsm/netinfo.py", line 119, in networks allNets = ((net, net.name()) for net in conn.listAllNetworks(0)) File "/usr/lib/python2.7/site-packages/vdsm/libvirtconnection.py", line 129, in wrapper __connections.get(id(target)).pingLibvirt() File "/usr/lib64/python2.7/site-packages/libvirt.py", line 3642, in getLibVersion if ret == -1: raise libvirtError ('virConnectGetLibVersion() failed', conn=self) libvirtError: internal error: client socket is closed

Does anything suspecious show up on libvirtd.log? I heard a report about a libvirt segmentation fault on centos7 - could this be it, too? Does `ps -o stime `pgrep libvirtd`` match with reported libvirt "internal error"? Would you provide the precise version of libvirt and vdsm? Can you think of anything that makes node8 "special"? Dan.