Dfdaemon crashloopbackoff when /var/run/dragonfly is mounted on hostPath and dfdaemon crashes unexpectedly

[e-ngo]

Bug report:

When upgrading to dragonfly's helm version 1.3.3+, the uds is now mounted on hostpath (with no way to opt out). Upon non graceful restarts such as process OOMKill, the dfdaemon will be stuck in crashloopbackoff state as it is unable to bind to the UDS.

Expected behavior:

Restarts should verify that the UDS is freed. If not, it should explicitly unbind it, before going through its normal binding process.

How to reproduce it:

Upgrade to helm 1.3.3. Mount dfdaemon's /var/run/dragonfly/ socket-dir as hostpath as per the helm above.

Run kill -9 PID of the dfdaemon client

Observe the log line:

2025-01-23T18:56:45.912674754+00:00 ERROR  dfdaemon download grpc server failed: Address already in use (os error 98)
    at dragonfly-client/src/bin/dfdaemon/main.rs:360

and crashloopbackoffs

Environment:

• Dragonfly version:
• OS:
• Kernel (e.g. uname -a):
• Others:
• • helm: 1.3.5

[e-ngo]

Should be doable via https://github.com/dragonflyoss/client/blob/main/dragonfly-client/src/grpc/dfdaemon_download.rs#L128-L171. Happy to put in a PR