Bootstrap --recover-from errors out with url scheme error

[apatrida]

Bug Report

Description

When running talosctl ... bootstrap --recover-from=/some/local/file we end up with error

Failed to run pre stage: error recovering from the snapshot: URL scheme must be http, https, unix, or unixs

Logs

summary of the only two log entries related:

 172.16.73.112: user: warning: [2025-06-26T14:46:07.677662372Z]: [talos] recovering etcd from snapshot: hash bbcbd3b7, revision 124687081, total keys 5517, total size 93896704
172.16.73.112: user: warning: [2025-06-26T14:46:07.678452372Z]: [talos] service[etcd](Failed): Failed to run pre stage: error recovering from the snapshot: URL scheme must be http, https, unix, or unixs:

from command

talosctl --talosconfig ~/.talos/tc-reset upgrade --nodes 172.16.73.112 --image ghcr.io/siderolabs/installer:v1.10.1 --wait --debug
WARNING: 172.16.73.112: server version 1.9.3 is older than client version 1.10.4
◱ watching nodes: [172.16.73.112]
    * 172.16.73.112: waiting for actor ID

will attach support.zip in comments.

Environment

• Talos version: [talosctl version --nodes <problematic nodes>]

Client:
	Tag:         v1.10.4
	SHA:         3c119bf8
	Built:
	Go version:  go1.24.4
	OS/Arch:     darwin/arm64
Server:
	NODE:        172.16.73.112
	Tag:         v1.9.3
	SHA:         d40df438
	Built:
	Go version:  go1.23.5
	OS/Arch:     linux/amd64
	Enabled:     RBAC

• Kubernetes version: [kubectl version]

• cannot report, this error is blocking k8s from even starting, as etcd isn't up yet.

• Platform:

Talos on metal

[apatrida]

tc-kc-200-support.zip

[apatrida]

persists in versions 1.10.1, and 1.10.4

[apatrida]

other note from etc.d client:

default:
    return nil, fmt.Errorf("URL scheme must be http, https, unix, or unixs: %s", in)

the error should report a URL at the end but doesn't, so Talos is passing in something empty when it constructs this call.

we have 2 ip's advertised per node, wondering if that is part of the problem. Changing our advertised to narrow to 1 each to see.

[apatrida]

Confirmed, if you advertise more than 1 addres per node, you cannot use restore / recover from a snapshot. you have to limit to 1, then do it, and add back the others after. This means anyone in this case can get into terminal conditions with etc.d that they cannot recover if they do not know this work around.

[smira]

Thanks for reporting this, this is indeed a bug