Process leak when using custom LaunchAgent with KeepAlive

[elfgzp]

Description

I encountered a process leak issue when running Colima with a custom LaunchAgent configuration. After investigating, I found the cause and wanted to share my findings in case it helps others.

Environment

• macOS: Darwin 25.2.0 (Apple Silicon)
• Colima version: (installed via Homebrew)
• VM Type: vz

The Problem

I had created a custom LaunchAgent plist at ~/Library/LaunchAgents/com.github.abiosoft.colima.plist with the following configuration:

<key>KeepAlive</key>
<true/>

When Colima failed to start (due to an unrelated disk attachment error), the KeepAlive setting caused launchd to continuously restart the process. However, each restart attempt spawned new limactl processes (especially limactl usernet) that were not properly cleaned up when the parent process exited.

Over time, this resulted in 3,800+ orphaned limactl processes accumulating on my system, which caused:

• Extremely high system load (Load Average: 200-600)
• kernel_task consuming significant CPU managing these processes
• Overall system sluggishness

Error Logs

From /tmp/colima.err, the startup was failing repeatedly:

time="2026-02-06T11:09:36+08:00" level=fatal msg="error starting vm: error at 'starting': exit status 1"
time="2026-02-06T11:09:47+08:00" level=fatal msg="error starting vm: error at 'starting': exit status 1`
... (repeated ~50 times in 10 minutes)

From ~/.colima/_lima/colima/ha.stderr.log:

{"level":"fatal","msg":"failed to run attach disk \"colima\", in use by instance \"colima\"","time":"2026-02-06T11:19:54+08:00"}

Resolution

I noticed that the official Homebrew service configuration uses a different KeepAlive setting:

<key>KeepAlive</key>
<dict>
    <key>SuccessfulExit</key>
    <true/>
</dict>

After updating my LaunchAgent to use SuccessfulExit instead of a plain <true/>, and cleaning up the orphaned processes, the issue was resolved.

I also found that using brew services start colima with the official Homebrew plist works well and avoids this problem entirely.

Investigation Notes

• The issue appears specific to using KeepAlive: true (boolean) versus KeepAlive.SuccessfulExit: true (dictionary)
• When startup fails repeatedly, the boolean form causes immediate restart without proper cleanup
• The SuccessfulExit form only restarts on clean exits, preventing the restart loop
• It might be helpful to document this behavior for users creating custom LaunchAgents

I was able to resolve this on my end, but wanted to report it in case it's helpful for the project or other users.

[elfgzp]

Additional Technical Details

My Custom LaunchAgent Configuration (problematic)

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
    <key>Label</key>
    <string>com.github.abiosoft.colima</string>
    <key>ProgramArguments</key>
    <array>
        <string>/bin/zsh</string>
        <string>-c</string>
        <string>export PATH="/opt/homebrew/bin:/opt/homebrew/sbin:$PATH"; /opt/homebrew/bin/colima start --foreground</string>
    </array>
    <key>RunAtLoad</key>
    <true/>
    <key>KeepAlive</key>
    <true/>
    <key>EnvironmentVariables</key>
    <dict>
        <key>PATH</key>
        <string>/opt/homebrew/bin:/opt/homebrew/sbin:/usr/local/bin:/usr/bin:/bin:/usr/sbin:/sbin</string>
    </dict>
    <key>StandardOutPath</key>
    <string>/tmp/colima.log</string>
    <key>StandardErrorPath</key>
    <string>/tmp/colima.err</string>
</dict>
</plist>

Process Count Statistics

At peak, I had:

• 3,825 limactl processes (mostly limactl usernet)
• 3,824 qemu-related processes
• Total system processes: 4,458 (vs ~650 after cleanup)

Each orphaned limactl usernet process was consuming 2-3.5% CPU continuously.

System Impact

Before cleanup:

• Load Avg: 205.10, 300.84, 625.85
• CPU usage: 21.59% user, 34.64% sys, 43.76% idle

After cleanup:

• Load Avg: 77.21, 233.34, 550.83 (still recovering)
• CPU usage: 31.69% user, 31.26% sys, 37.4% idle

Related Issues

• launchd auto start support missing? #74 - launchd auto start support missing
• Daemon fail in when launching colima start from launchd user agent #490 - Daemon fail when launching from launchd
• Colima daemon #262 - Colima daemon enhancement
• stopping colima with launchctl unload path/to/colima.launchd.plist doesn't stop colima #1346 - launchctl unload doesn't stop colima

These discuss launchd integration but I didn't see this specific process leak scenario mentioned.

Root Cause

The disk lock error prevented VM startup, but KeepAlive: true forced immediate restarts. Each restart spawned new limactl processes without proper cleanup, creating a leak. The Homebrew plist uses KeepAlive.SuccessfulExit which only restarts on clean exits.

[blackblather]

Thank you. This is an oddly specific issue to encounter, but I just had it as well and you're the only one I found that had it.
Your solution helped me.

In my case it kept giving me this error: "failed to run attach disk \"colima\", in use by instance \"colima\""
So in my case I just run:

colima delete --force
colima start --foreground

in my launch daemon.

Its not ideal to delete and start it, but so far it works. Maybe a future update will fix this issue the proper way.

Not sure how this will play out if/when I start using docker volumes. Will running colima delete also remove the volumes? Not sure. So far, so good.

Thanks