TroubleShooting

Working with Server Logs

The CloudStack Management Server logs all web site, middle tier, and database activities for diagnostics purposes in /var/log/cloudstack/management/. The CloudStack logs a variety of error messages. We recommend this command to find the problematic output in the Management Server log:.

Note

When copying and pasting a command, be sure the command has pasted as a single line before executing. Some document viewers may introduce unwanted line breaks in copied text.

  1. grep -i -E 'exception|unable|fail|invalid|leak|warn|error' /var/log/cloudstack/management/management-server.log

The CloudStack processes requests with a Job ID. If you find an error in the logs and you are interested in debugging the issue you can grep for this job ID in the management server log. For example, suppose that you find the following ERROR message:

  1. 2010-10-04 13:49:32,595 ERROR [cloud.vm.UserVmManagerImpl] (Job-Executor-11:job-1076) Unable to find any host for [User|i-8-42-VM-untagged]

Note that the job ID is 1076. You can track back the events relating to job 1076 with the following grep:

  1. grep "job-1076)" management-server.log

The CloudStack Agent Server logs its activities in /var/log/cloudstack/agent/.

Data Loss on Exported Primary Storage

Symptom

Loss of existing data on primary storage which has been exposed as a Linux NFS server export on an iSCSI volume.

Cause

It is possible that a client from outside the intended pool has mounted the storage. When this occurs, the LVM is wiped and all data in the volume is lost

Solution

When setting up LUN exports, restrict the range of IP addresses that are allowed access by specifying a subnet mask. For example:

  1. echo “/export 192.168.1.0/24(rw,async,no_root_squash,no_subtree_check)” > /etc/exports

Adjust the above command to suit your deployment needs.

More Information

See the export procedure in the “Secondary Storage” section of the CloudStack Installation Guide

Recovering a Lost Virtual Router

Symptom

A virtual router is running, but the host is disconnected. A virtual router no longer functions as expected.

Cause

The Virtual router is lost or down.

Solution

If you are sure that a virtual router is down forever, or no longer functions as expected, destroy it. You must create one afresh while keeping the backup router up and running (it is assumed this is in a redundant router setup):

  • Force stop the router. Use the stopRouter API with forced=true parameter to do so.
  • Before you continue with destroying this router, ensure that the backup router is running. Otherwise the network connection will be lost.
  • Destroy the router by using the destroyRouter API.

Recreate the missing router by using the restartNetwork API with cleanup=false parameter. For more information about redundant router setup, see Creating a New Network Offering.

For more information about the API syntax, see the API Reference at http://cloudstack.apache.org/docs/api/.

Maintenance mode not working on vCenter

Symptom

Host was placed in maintenance mode, but still appears live in vCenter.

Cause

The CloudStack administrator UI was used to place the host in scheduled maintenance mode. This mode is separate from vCenter’s maintenance mode.

Solution

Use vCenter to place the host in maintenance mode.

Unable to deploy VMs from uploaded vSphere template

Symptom

When attempting to create a VM, the VM will not deploy.

Cause

If the template was created by uploading an OVA file that was created using vSphere Client, it is possible the OVA contained an ISO image. If it does, the deployment of VMs from the template will fail.

Solution

Remove the ISO and re-upload the template.

Unable to power on virtual machine on VMware

Symptom

Virtual machine does not power on. You might see errors like:

  • Unable to open Swap File
  • Unable to access a file since it is locked
  • Unable to access Virtual machine configuration

Cause

A known issue on VMware machines. ESX hosts lock certain critical virtual machine files and file systems to prevent concurrent changes. Sometimes the files are not unlocked when the virtual machine is powered off. When a virtual machine attempts to power on, it can not access these critical files, and the virtual machine is unable to power on.

Solution

See the following:

VMware Knowledge Base Article

Load balancer rules fail after changing network offering

Symptom

After changing the network offering on a network, load balancer rules stop working.

Cause

Load balancing rules were created while using a network service offering that includes an external load balancer device such as NetScaler, and later the network service offering changed to one that uses the CloudStack virtual router.

Solution

Create a firewall rule on the virtual router for each of your existing load balancing rules so that they continue to function.

Troubleshooting Internet Traffic

Below are a few troubleshooting steps to check whats going wrong with your network…

Trouble Shooting Steps

  1. The switches have to be configured correctly to pass VLAN traffic. You can verify if VLAN traffic is working by bringing up a tagged interface on the hosts and pinging between them as below…

    On host1 (kvm1)

    1. kvm1 ~$ vconfig add eth0 64
    2. kvm1 ~$ ifconfig eth0.64 1.2.3.4 netmask 255.255.255.0 up
    3. kvm1 ~$ ping 1.2.3.5

    On host2 (kvm2)

    1. kvm2 ~$ vconfig add eth0 64
    2. kvm2 ~$ ifconfig eth0.64 1.2.3.5 netmask 255.255.255.0 up
    3. kvm2 ~$ ping 1.2.3.4

    If the pings dont work, run tcpdump(8) all over the place to check who is gobbling up the packets. Ultimately, if the switches are not configured correctly, CloudStack networking wont work so fix the physical networking issues before you proceed to the next steps

  2. Ensure Traffic Labels are set for the Zone.

    Traffic labels need to be set for all hypervisors including XenServer, KVM and VMware types. You can configure traffic labels when you creating a new zone from the Add Zone Wizard.

    ../_images/networking-zone-traffic-labels.png

    On an existing zone, you can modify the traffic labels by going to Infrastructure, Zones, Physical Network tab.

    ../_images/networking-infra-traffic-labels.png

    List labels using CloudMonkey

    1. acs-manager ~$ cloudmonkey list traffictypes physicalnetworkid=41cb7ff6-8eb2-4630-b577-1da25e0e1145
    2. count = 4
    3. traffictype:
    4. id = cd0915fe-a660-4a82-9df7-34aebf90003e
    5. kvmnetworklabel = cloudbr0
    6. physicalnetworkid = 41cb7ff6-8eb2-4630-b577-1da25e0e1145
    7. traffictype = Guest
    8. xennetworklabel = MGMT
    9. ========================================================
    10. id = f5524b8f-6605-41e4-a982-81a356b2a196
    11. kvmnetworklabel = cloudbr0
    12. physicalnetworkid = 41cb7ff6-8eb2-4630-b577-1da25e0e1145
    13. traffictype = Management
    14. xennetworklabel = MGMT
    15. ========================================================
    16. id = 266bad0e-7b68-4242-b3ad-f59739346cfd
    17. kvmnetworklabel = cloudbr0
    18. physicalnetworkid = 41cb7ff6-8eb2-4630-b577-1da25e0e1145
    19. traffictype = Public
    20. xennetworklabel = MGMT
    21. ========================================================
    22. id = a2baad4f-7ce7-45a8-9caf-a0b9240adf04
    23. kvmnetworklabel = cloudbr0
    24. physicalnetworkid = 41cb7ff6-8eb2-4630-b577-1da25e0e1145
    25. traffictype = Storage
    26. xennetworklabel = MGMT
    27. =========================================================
  3. KVM traffic labels require to be named as “cloudbr0”, “cloudbr2”, “cloudbrN” etc and the corresponding bridge must exist on the KVM hosts. If you create labels/bridges with any other names, CloudStack (atleast earlier versions did) seems to ignore them. CloudStack does not create the physical bridges on the KVM hosts, you need to create them before before adding the host to Cloudstack.

    1. kvm1 ~$ ifconfig cloudbr0
    2. cloudbr0 Link encap:Ethernet HWaddr 00:0C:29:EF:7D:78
    3. inet addr:192.168.44.22 Bcast:192.168.44.255 Mask:255.255.255.0
    4. inet6 addr: fe80::20c:29ff:feef:7d78/64 Scope:Link
    5. UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
    6. RX packets:92435 errors:0 dropped:0 overruns:0 frame:0
    7. TX packets:50596 errors:0 dropped:0 overruns:0 carrier:0
    8. collisions:0 txqueuelen:0
    9. RX bytes:94985932 (90.5 MiB) TX bytes:61635793 (58.7 MiB)
  4. The Virtual Router, SSVM, CPVM public interface would be bridged to a physical interface on the host. In the example below, cloudbr0 is the public interface and CloudStack has correctly created the virtual interfaces bridge. This virtual interface to physical interface mapping is done automatically by CloudStack using the traffic label settings for the Zone. If you have provided correct settings and still dont have a working working Internet, check the switching layer before you debug any further. You can verify traffic using tcpdump on the virtual, physical and bridge interfaces.

    1. kvm-host1 ~$ brctl show
    2. bridge name bridge id STP enabled interfaces
    3. breth0-64 8000.000c29ef7d78 no eth0.64
    4. vnet2
    5. cloud0 8000.fe00a9fe0219 no vnet0
    6. cloudbr0 8000.000c29ef7d78 no eth0
    7. vnet1
    8. vnet3
    9. virbr0 8000.5254008e321a yes virbr0-nic
    1. xenserver1 ~$ brctl show
    2. bridge name bridge id STP enabled interfaces
    3. xapi0 0000.e2b76d0a1149 no vif1.0
    4. xenbr0 0000.000c299b54dc no eth0
    5. xapi1
    6. vif1.1
    7. vif1.2
  5. Pre-create labels on the XenServer Hosts. Similar to KVM bridge setup, traffic labels must also be pre-created on the XenServer hosts before adding them to CloudStack.

    1. xenserver1 ~$ xe network-list
    2. uuid ( RO) : aaa-bbb-ccc-ddd
    3. name-label ( RW): MGMT
    4. name-description ( RW):
    5. bridge ( RO): xenbr0
  6. The Internet would be accessible from both the SSVM and CPVM instances by default. Their public IPs will also be directly pingable from the Internet. Please note that these test would work only if your switches and traffic labels are configured correctly for your environment. If your SSVM/CPVM cant reach the Internet, its very unlikely that the Virtual Router (VR) can also the reach the Internet suggesting that its either a switching issue or incorrectly assigned traffic labels. Fix the SSVM/CPVM issues before you debug VR issues.

    1. root@s-1-VM:~# ping -c 3 google.com
    2. PING google.com (74.125.236.164): 56 data bytes
    3. 64 bytes from 74.125.236.164: icmp_seq=0 ttl=55 time=26.932 ms
    4. 64 bytes from 74.125.236.164: icmp_seq=1 ttl=55 time=29.156 ms
    5. 64 bytes from 74.125.236.164: icmp_seq=2 ttl=55 time=25.000 ms
    6. --- google.com ping statistics ---
    7. 3 packets transmitted, 3 packets received, 0% packet loss
    8. round-trip min/avg/max/stddev = 25.000/27.029/29.156/1.698 ms
    1. root@v-2-VM:~# ping -c 3 google.com
    2. PING google.com (74.125.236.164): 56 data bytes
    3. 64 bytes from 74.125.236.164: icmp_seq=0 ttl=55 time=32.125 ms
    4. 64 bytes from 74.125.236.164: icmp_seq=1 ttl=55 time=26.324 ms
    5. 64 bytes from 74.125.236.164: icmp_seq=2 ttl=55 time=37.001 ms
    6. --- google.com ping statistics ---
    7. 3 packets transmitted, 3 packets received, 0% packet loss
    8. round-trip min/avg/max/stddev = 26.324/31.817/37.001/4.364 ms
  7. The Virtual Router (VR) should also be able to reach the Internet without having any Egress rules. The Egress rules only control forwarded traffic and not traffic that originates on the VR itself.

    1. root@r-4-VM:~# ping -c 3 google.com
    2. PING google.com (74.125.236.164): 56 data bytes
    3. 64 bytes from 74.125.236.164: icmp_seq=0 ttl=55 time=28.098 ms
    4. 64 bytes from 74.125.236.164: icmp_seq=1 ttl=55 time=34.785 ms
    5. 64 bytes from 74.125.236.164: icmp_seq=2 ttl=55 time=69.179 ms
    6. --- google.com ping statistics ---
    7. 3 packets transmitted, 3 packets received, 0% packet loss
    8. round-trip min/avg/max/stddev = 28.098/44.021/69.179/17.998 ms
  8. However, the Virtual Router’s (VR) Source NAT Public IP address WONT be reachable until appropriate Ingress rules are in place. You can add Ingress rules under Network, Guest Network, IP Address, Firewall setting page.

    ../_images/networking-ingress-rule.png

  9. The VM Instances by default wont be able to access the Internet. Add Egress rules to permit traffic.

    ../_images/networking-egress-rule.png

  10. Some users have reported that flushing IPTables rules (or changing routes) on the SSVM, CPVM or the Virtual Router makes the Internet work. This is not expected behaviour and suggests that your networking settings are incorrect. No IPtables/route changes are required on the SSVM, CPVM or the VR. Go back and double check all your settings.

In a vast majority of the cases, the problem has turned out to be at the switching layer where the L3 switches were configured incorrectly.

This section was contibuted by Shanker Balan and was originally published on Shapeblue’s blog