Volumes created with ontap-nas may not be mountable right away when load-sharing mirrors are in use

[remys89]

Hi,

As discussed with the team on slack today, the following issue:

tl;dr: Create a post-creation check to verify if a volume exists before starting the container who uses it.

We have a single test filer for running PoCs. We run NetApp ONTAP 9, Docker 1.13 & The latest nDVP plugin. When we perform a "Docker volume create" towards the netapp storage driver, the volume is not directly available on the filer, possibly because of a configured advertisement delay on our configuration.

[root@vm~]# docker volume create -d netapp --name=netappslack2 netappslack2 netapp:latest netappslack2

showmount gives /netappdvp_netappslack2 (everyone)

[root@vm~]# mount .99:/netappdvp_netappslack2 test mount.nfs: mounting .99:/netappdvp_netappslack2 failed, reason given by server: No such file or directory

Yet, containers expect the volume to be there for writing data. Now, the container starts and exits directly using status 32, saying the "file or directory is not available".

Waiting a few minutes, around 4, we perform the docker run again using the created volume. Then, it works fine and mounts the volume.

[root@VM~]# docker run --rm -it -v netappslack2:/take2 alpine ash / #

[dutchiechris]

I would be curious to see the entries made to the nDVP log and the internal mgt log of the cluster to check and compare timings.

For the nDVP logs check here how to get them.

For the cluster logs check here on how to get them. The specific logfile to look at is /mroot/etc/log/mlog/mgwd.log on the node that owns the aggregate.

Once you have access to both logs then (1) create a volume and then (2) check both logs and reply to this issue with the details.

In my lab I did the above steps and from the cluster logs the vol create took only 1 second from entries with Calling doCreate to doCreate complete so I am curious what is reported on your cluster. I have seen connect failures when using iSCSI storage on the 1st connect. In my research I learned that on the 1st connect the device will be formatted, and for a very large device, or with a low max QoS IOP throttle, the formatting takes longer than Docker is willing to wait. With NFS however you don't have this format phase so your connect issue must be different.

I hope with the logs we can see what is going on. Thanks!

[remys89]

At the moment of mounting, this was the logging @ nDVP.log:

INFO[2017-06-12T13:27:01Z] Starting NetApp Docker Volume Plugin.         port= storageDriverName=ontap-nas volumeDriver=netapp volumePath="/var/lib/docker-volumes/netapp"
INFO[2017-06-12T13:35:40Z] Initialized logging.                          logFileLocation="/var/log/netappdvp/netapp.log" logLevel=info
INFO[2017-06-12T13:35:41Z] Initialized Ontap NAS storage driver.         driverVersion=17.04.0 extendedDriverVersion=native
INFO[2017-06-12T13:35:41Z] Starting NetApp Docker Volume Plugin.         port= storageDriverName=ontap-nas volumeDriver=netapp volumePath="/var/lib/docker-volumes/netapp"
ERRO[2017-06-12T13:36:50Z] Problem mounting volume: netappdvp_netapptest mountpoint: /var/lib/docker-volumes/netapp/netappdvp_netapptest error: exit status 32'

Ill come back to you with the storage logging.

[remys89]

@dutchiechris uploaded to NetApp FTP, Scott Stanton will move it if i am correct.

[innergy]

As we discussed on Slack, this appears to be caused by the use of load-sharing mirrors, which means that new volumes appear in the namespace based on a snapshot schedule. This will only be an issue with the 'ontap-nas' driver.

There are multiple potential solutions that would get us closer to the desired behavior: the volume is ready to be mounted as soon as the create operation returns. We'll investigate.

Thanks for the report!