This is the multi-page printable view of this section. Click here to print.
Advanced
1 - Enhanced supply chain security with gomodjail
gomodjail is an experimental library sandbox for Go modules.
gomodjail imposes syscall restrictions on a specific set of Go modules, so as to mitigate their potential vulnerabilities and supply chain attack vectors. A restricted module is hindered to access files and execute commands.
gomodjail can be enabled for nerdctl by using the nerdctl.gomodjail binary.
lima nerdctl.gomodjail ...
For the gomodjail policy applied to nerdctl.gomodjail, see https://github.com/containerd/nerdctl/blob/main/go.mod.
2 - Accelerating rootless networking with bypass4netns
bypass4netns is an experimental accelerator for rootless networking.
On macOS hosts, it is highly recommended to use the vzNAT networking in conjunction to reduce the overhead of Lima’s user-mode networking:
limactl start --network vzNAT
To enable bypass4netns, the daemon process (bypass4netnsd) has to be installed in the VM as follows:
lima containerd-rootless-setuptool.sh install-bypass4netnsd
Then run a container with an annotation nerdctl/bypass4netns=true:
# 192.168.64.1 is the IP address of the "bridge100" interface on the macOS host
lima nerdctl run --annotation nerdctl/bypass4netns=true alpine \
sh -euc 'apk add iperf3 && iperf3 -c 192.168.64.1'
Benchmark result:
| Mode | Throughput |
|---|---|
| Rootless without bypass4netns | 2.30 Gbits/sec |
| Rootless with bypass4netns | 86.0 Gbits/sec |
| Rootful | 90.3 Gbits/sec |
Benchmarking environment
- Lima version: 2.0.0-alpha.2
- nerdctl 2.1.6
- containerd 2.1.4
- bypass4netns 0.4.2
- Container: Alpine Linux 3.22.2
- iperf 3.19.1-r0 (apk)
- Guest: Ubuntu 25.04
- Host: macOS 26.0.1
- iperf 3.19.1 (Homebrew)
- Hardware: MacBook Pro 2024 (M4 Max, 128 GiB)
3 - Accelerating start-up time with eStargz
eStargz is an OCI-compatible container image format that reduces start-up latency using lazy-pulling technique.
The support for eStargz is available by default in Lima.
The timings below were measured on an Apple M5 Max (macOS, VZ-backend Lima, default template) pulling the native arm64 images. Numbers are a median of three cold runs (image removed with nerdctl rmi between each run).
Without eStargz:
$ time lima nerdctl run ghcr.io/stargz-containers/python:3.13-org python3 -c 'print("hi")'
hi
real 0m14.031s
user 0m0.017s
sys 0m0.018s
With eStargz:
$ time lima nerdctl --snapshotter=stargz run ghcr.io/stargz-containers/python:3.13-esgz python3 -c 'print("hi")'
hi
real 0m3.275s
user 0m0.017s
sys 0m0.016s
Examples of eStargz images can be found at https://github.com/containerd/stargz-snapshotter/blob/main/docs/pre-converted-images.md.
See also: