tart exec fails after Linux VM reboot

[smcloughlinTT]

Version: 2.32.0

tart exec works the first time it's run on a VM but fails after restarting the VM.

tart clone ghcr.io/cirruslabs/ubuntu:latest test-vm
tart set test-vm --disk-size 20
tart run test-vm --no-graphics & disown
tart exec test-vm "ls"  # This should ls the home directory of the VM
tart stop test-vm
tart run test-vm --no-graphics & disown

# This will hang for ~10s and then fail with "Failed to connect to the VM using its control socket: The operation couldn't be completed. (GRPC.GRPCConnectionPoolError error 1.), is the Tart Guest Agent running?"
tart exec test-vm "ls"

This appears to be a macOS-side issue where stale control sockets are not cleaned up. I've been able to successfully work around this with this script:

#!/bin/bash
# Wrapper around 'tart run' that cleans up stale control sockets first.
# Usage: tart-run <vm-name> [extra tart run args...]

VM_NAME="$1"
if [ -z "$VM_NAME" ]; then
  echo "Usage: tart-run <vm-name> [args...]" >&2
  exit 1
fi

SOCK="$HOME/.tart/vms/$VM_NAME/control.sock"

if [ -S "$SOCK" ]; then
  echo "Removing stale control socket for $VM_NAME..."
  rm -f "$SOCK"
fi

shift
exec tart run "$VM_NAME" "$@"

[edigaryev]

I've had to slightly modify your shell sequence (because e.g. tart exec requires a VM name argument before the command starts) and wrapped it in the following Bash script:

#!/usr/bin/env bash

# Set shell options to enable fail-fast behavior
#
# * -e: fail the script when an error occurs or command fails
# * -u: fail the script when attempting to reference unset parameters
# * -o pipefail: by default an exit status of a pipeline is that of its
#                last command, this fails the pipe early if an error in
#                any of its commands occurs
#
set -euo pipefail

VM_NAME="test-vm"

# Create a VM
tart pull ghcr.io/cirruslabs/ubuntu:latest
tart clone ghcr.io/cirruslabs/ubuntu:latest "${VM_NAME}"
tart set "${VM_NAME}" --disk-size 20
tart run "${VM_NAME}" --no-graphics & disown

# Wait a bit and execute some command
sleep 10
tart exec "${VM_NAME}" "id" 

# Re-create the VM
tart stop "${VM_NAME}"
tart run "${VM_NAME}" --no-graphics & disown

# Wait a bit and execute some command
sleep 10
tart exec "${VM_NAME}" "id"

# Cleanup
tart stop "${VM_NAME}"

However, I can't seem to reproduce what you're observing:

% ./reproduce.sh 
pulling ghcr.io/cirruslabs/ubuntu:latest...
pulling manifest...
ghcr.io/cirruslabs/ubuntu@sha256:c8ea4bd63d32c8204f0592be5ce33c082cefdc9c64dc97a71d84428d80b3c89d image is already cached and linked!
uid=1000(admin) gid=109(admin) groups=109(admin)
Stopping VM...
uid=1000(admin) gid=109(admin) groups=109(admin)
Stopping VM...

tart run itself automatically cleans up the control socket on start: https://github.com/cirruslabs/tart/blob/094f85004618e10017ad8ae8a0bdb7829f6087a3/Sources/tart/ControlSocket.swift#L20-L22

1. Could you please try reproducing this again using the script above?

2. Which macOS version are you running on the host?

[smcloughlinTT]

Sorry about the incorrect ITR; I've updated the original bug.

1. I used your reproduce.sh but I still hit the same failure.

pulling ghcr.io/cirruslabs/ubuntu:latest...
pulling manifest...
pulling disk (2.8 GB compressed)...
100%
pulling NVRAM...
uid=1000(admin) gid=109(admin) groups=109(admin)
Stopping VM...
Failed to connect to the VM using its control socket: The operation couldn’t be completed. (GRPC.GRPCConnectionPoolError error 1.), is the Tart Guest Agent running?

2. Sequoia 15.7.3

[edigaryev]

Please see #1224 (comment).