• Troubleshooting Guide
  • UI
  • Manager and Engines
  • CSI driver
  • Flexvolume Driver
• Common issues
  • Volume can be attached/detached from UI, but Kubernetes Pod/StatefulSet etc cannot use it
    • Using with Flexvolume Plugin

Troubleshooting Guide

For a more in-depth troubleshooting flow please see https://github.com/longhorn/longhorn/wiki/Troubleshooting

There are a few components in Longhorn: Manager, Engine, Driver and UI. By default, all of those components run as pods in the longhorn-system namespace in the Kubernetes cluster.

Most of the logs are included in the Support Bundle. You can click the Generate Support Bundle link at the bottom of the UI to download a zip file that contains Longhorn-related configuration and logs. See Support Bundle for detail.

One exception is the dmesg, which needs to be retrieved from each node by the user.

UI

Make use of the Longhorn UI is a good start for the troubleshooting. For example, if Kubernetes cannot mount one volume correctly, after stop the workload, try to attach and mount that volume manually on one node and access the content to check if volume is intact.

Also, the event logs in the UI dashboard provides some information of probably issues. Check for the event logs in Warning level.

Manager and Engines

You can get the logs from the Longhorn Manager and Engines to help with troubleshooting. The most useful logs are the ones from longhorn-manager-xxx, and the logs inside Longhorn instance managers, e.g. instance-manager-xxxx, instance-manager-e-xxxx and instance-manager-r-xxxx.

Since normally there are multiple Longhorn Managers running at the same time, we recommend using kubetail, which is a great tool to keep track of the logs of multiple pods. To track the manager logs in real time, you can use:

kubetail longhorn-manager -n longhorn-system

CSI driver

For the CSI driver, check the logs for csi-attacher-0 and csi-provisioner-0, as well as containers in longhorn-csi-plugin-xxx.

Flexvolume Driver

The FlexVolume driver is deprecated as of Longhorn v0.8.0 and should no longer be used.

First check where the driver has been installed on the node. Check the log of longhorn-driver-deployer-xxxx for that information.

Then check the kubelet logs. The FlexVolume driver itself doesn’t run inside the container. It would run along with the kubelet process.

If kubelet is running natively on the node, you can use the following command to get the logs:

journalctl -u kubelet

Or if kubelet is running as a container (e.g. in RKE), use the following command instead:

docker logs kubelet

For even more detailed logs of Longhorn FlexVolume, run the following command on the node or inside the container (if kubelet is running as a container, e.g. in RKE):

touch /var/log/longhorn_driver.log

Common issues

Volume can be attached/detached from UI, but Kubernetes Pod/StatefulSet etc cannot use it

Using with Flexvolume Plugin

Check if the volume plugin directory has been set correctly. This is automatically detected unless the user explicitly sets it.

By default, Kubernetes uses /usr/libexec/kubernetes/kubelet-plugins/volume/exec/, as stated in the official document.

Some vendors choose to change the directory for various reasons. For example, GKE uses /home/kubernetes/flexvolume instead.

The correct directory can be found by running ps aux|grep kubelet on the host and check the --volume-plugin-dir parameter. If there is none, the default /usr/libexec/kubernetes/kubelet-plugins/volume/exec/ will be used.

Profiling

Engine, replica, and sync agent runtime

You can enable the pprof server dynamically to perform runtime profiling. To enable profiling, you can:

1. Shell into the instance manager pod.
2. Identify the runtime process and its port using ps:

   $ ps aux | more
   
   USER         PID %CPU %MEM    VSZ   RSS TTY      STAT START   TIME COMMAND
   ...
   root        1996  0.0  0.6 1990080 20996 ?       Sl   Jul25   0:05 /host/var/lib/longhorn/engine-binaries/longhornio-longhorn-engine-v1.10.0/longhorn --volume-name     vol replica /host/var/lib/longhorn/replicas/vol-3004fc59 --size 1073741824 --disableRevCounter --replica-instance-name vol-r-ec7e35e4 --snapshot-max-count 250     --snapshot-max-size 0 --sync-agent-port-count 7 --listen 0.0.0.0:10000
   root        2004  0.0  0.6 1695152 22708 ?       Sl   Jul25   0:09 /host/var/lib/longhorn/engine-binaries/longhornio-longhorn-engine-v1.10.0/longhorn --volume-name     vol sync-agent --listen 0.0.0.0:10002 --replica 0.0.0.0:10000 --listen-port-range 10003-10009 --replica-instance-name vol-r-ec7e35e4
   root        2031  0.0  0.6 1916348 23760 ?       Sl   Jul25   0:46 /engine-binaries/longhornio-longhorn-engine-v1.10.0/longhorn --engine-instance-name vol-e-0     controller vol --frontend tgt-blockdev --disableRevCounter --size 1073741824 --current-size 0 --engine-replica-timeout 8 --file-sync-http-client-timeout 30     --snapshot-max-count 250 --snapshot-max-size 0 --replica tcp://10.42.2.7:10000 --replica tcp://10.42.0.15:10000 --replica tcp://10.42.1.7:10000 --listen 0.0.0.0:10010
   

3. Enable the pprof server for the desired runtime (for example, sync-agent):

   In this example, the sync-agent process listens on port 10002.

   $ /host/var/lib/longhorn/engine-binaries/longhornio-longhorn-engine-v1.10.0/longhorn --url http://localhost:10002 profiler enable --port 36060
   $ /host/var/lib/longhorn/engine-binaries/longhornio-longhorn-engine-v1.10.0/longhorn --url http://localhost:10002 profiler show
   
   Profiler enabled at Addr: *:36060
   

   The pprof server is now accessible at http://localhost:36060 inside the instance manager pod.

4. Use the pprof interface for runtime inspection. For more details, refer to the official pprof documentation.
5. Disable the profiler after completing your analysis:

   $ /host/var/lib/longhorn/engine-binaries/longhornio-longhorn-engine-v1.10.0/longhorn --url http://localhost:10002 profiler disable
   
   Profiler is disabled!