Error encountered when using CAPD on btrfs: Error response from daemon: Duplicate mount point: /dev/mapper

[cannonpalms]

What steps did you take and what happened?

I'm having difficulties running CAPD/the Tilt local development environment. When I deploy a workload cluster, the DockerCluster controller immediately errors with the following message:

failed to create load balancer: error creating container \"capi-quickstart-lb\": Error response from daemon: Duplicate mount point: /dev/mapper

I first reported this issue in a Slack thread where @killianmuldoon kindly helped to identify that the problem is reproducible with the btrfs docker storage driver but not with the overlay2 docker storage driver.

This suggests to me that the issue is unlikely to be within CAPD itself, but more likely a lower layer like kind or docker. However, this can serve as a helpful tracking issue for the issue that bubbles to the surface when attempting to use CAPD.

What did you expect to happen?

Expected a workload cluster named capi-quickstart-cluster to be created.

Cluster API version

v1.3.5

Kubernetes version

1.26

Anything else you would like to add?

No response

Label(s) to be applied

/kind bug
One or more /area label. See https://github.com/kubernetes-sigs/cluster-api/labels?q=area for the list of labels.

[fabriziopandini]

/triage accepted

I have recently aligned the flags used to create containers between kind and CAPD (#8157), and this included also a change for btrfs
@cannonpalms could you check with latest rc release?

[cannonpalms]

Hi @fabriziopandini, the same issue occurs on v1.4.0-rc.1. ISTM (and my opinion is an uneducated one) that both CAPD AND kind are adding the /dev/mapper bind/mount point now. Perhaps CAPD can remove its workaround for btrfs support and defer to kind?

I will test a build that does this and report back.

[cannonpalms]

Simply removing CAPD's explicit bind/mount worked just fine... Please see the linked PR and tell me where I'm going wrong in my thinking here. I don't think we need to bind/mount anymore now that kind does it for us.

[fabriziopandini]

commented on the PR

[ipetrov117]

Hit the same issue when deploying a product using CAPD on an OS with the BTRFS filesystem. From the information here and in the PR I figure that just removing the second bind/mount in the code is enough. Is that the case?

[killianmuldoon]

From the information here and in the PR I figure that just removing the second bind/mount in the code is enough. Is that the case?

As far as I found in my local testing that fixed the issue. If @cannonpalms is happy with it, it would be good if you could pick up the PR.

[ipetrov117]

From the information here and in the PR I figure that just removing the second bind/mount in the code is enough. Is that the case?

As far as I found in my local testing that fixed the issue. If @cannonpalms is happy with it, it would be good if you could pick up the PR.

Alright, I will take a stab at the PR when I have some free time in the upcoming days.