[Magic CLI]: Command “magic init my-project” runs indefinitely on Ubuntu 24.04.1 LTS

[ZoaiCapital]

Issue description

The magic shell command runs indefinitely without completing or showing any output. Similarly, when trying to initialize a project using magic init my-project, the same behavior occurs. Both commands hang indefinitely without providing any feedback or errors.

Expected behavior:
The commands should complete successfully, either initializing a project or setting up the environment.

Steps to reproduce

Steps to Reproduce:

• Install Magic CLI using the command:

curl -ssL https://magic.modular.com/4519bd52-ec0f-43e6-a8d1-865720aafd1e | bash

• Verify installation using:

magic --version

The command confirms the CLI is installed successfully.

• Run the following command to initialize a new project:

magic init my-project

Result: The command hangs indefinitely.

Version Info

- Provide magic CLI version by pasting the output of `magic --version`
magic 0.5.0 - (based on pixi 0.37.0)

- What OS did you install the magic CLI on?

Ubuntu 24.04.1 LTS

- Include your CPU architecture e.g. x86-64 or arm64.
x86_64
AMD Ryzen 9 7950X 16-Core Processor

[nguiard]

Hi, happening to me too:

• magic 0.5.1 - (based on pixi 0.37.0)
• Ubuntu 24.04.1 LTS
• x86-64 - Intel Xeon W10885M

Edit: not happening (working fine) on another machine having Ubuntu 22.04 LTS, x86-64, magic 0.5.1

[zbowling]

Can you run with -vv in the args. so magic init -vv my-project and paste the output?

[ZoaiCapital]

I ran the command and let it execute for 2 minutes before interrupting it. The issue persists, as the execution hangs during this time.

[zbowling]

can you try with --vvvv ? Increase the spam level further.

[zbowling]

If you are on macos can you run sample magic --wait -f sample.txt on another shell before running magic and then post the spample.txt?

[ZoaiCapital]

I still have the same issue.

On my mac it works well, the problem occurs on Ubuntu 24.04.1 LTS

[jakewins]

Have the same issue on Arch so I don't think it's an ubuntu problem.

Running strace it looks like it locks up at 100% CPU right after it closes some socket.. so I guess whatever magic does after it's done with that socket is the problem, maybe?

$  strace magic -vvv init example

<snip lots of syscalls>

stat("/home/jake/.modular/webUserId", {st_mode=S_IFREG|0644, st_size=37, ...}) = 0
open("/home/jake/.modular/webUserId", O_RDONLY|O_LARGEFILE|O_CLOEXEC) = 9
fcntl(9, F_SETFD, FD_CLOEXEC)           = 0
fstat(9, {st_mode=S_IFREG|0644, st_size=37, ...}) = 0
read(9, "a9f21ffc-5228-4110-9dfa-c5301f82"..., 37) = 37
read(9, "", 32)                         = 0
close(9)                                = 0

socket(AF_NETLINK, SOCK_RAW|SOCK_CLOEXEC, NETLINK_ROUTE) = 9

<snip: Lots of sentto/recvfrom calls>

recvfrom(9, [{nlmsg_len=20, nlmsg_type=NLMSG_DONE, nlmsg_flags=NLM_F_MULTI, nlmsg_seq=2, nlmsg_pid=2145380}, 0], 8192, MSG_DONTWAIT, NULL, NULL) = 20
close(9)                                  = 0

<process stalls here, no more syscalls after that close()>

$ uname -a
Linux jake-p14 6.10.13-3-MANJARO #1 SMP PREEMPT_DYNAMIC Tue Oct  8 03:24:49 UTC 2024 x86_64 GNU/Linux

[zbowling]

Thanks! I pushed a pre-release build of magic 0.6.0 that may (or may not) resolve the deadlocking issue a few people have been hitting. We still haven't been able to reproduce the issue some folks are seeing internally but this build has mitigations that might help from the traces we received and has an new ENV that may help if it still doesn't (based on hunch on where the issue might be coming from with doing socket establishment in certain network environments with a telemetry server even if telemetry isn't being sent).

You can install this (assuming magic isn't deadlocking for you) with magic self-update --prerelease.

If it is I made a script to directly reinstall this pre-release: curl -ssL https://gist.githubusercontent.com/zbowling/4153d8db6dc98ff4ecd73fbe65f4df92/raw/55d9a64273c68b38d6b7d9a19911b27ab9d4d46f/fetch.sh | bash

If it still locks up for you with this build you can try doing export MAGIC_TELEMETRY=false in your shell before invoking any commands and seeing if that resolves it for you.

[nguiard]

Hi @zbowling , thanks! So, for information, in my case:

• doing just magic self-update --prerelease deadlocked
• before trying your script, I figure I'd try MAGIC_TELEMETRY=false magic self-update --prerelease. This deadlocked too
• your script worked for upgrading
• trying magic init my-project after upgrading still deadlocked
• doing MAGIC_TELEMETRY=false magic init my-project, however, worked without a problem

[jakewins]

Can confirm the same, export MAGIC_TELEMETRY=false after running your bash script made things work.

Flipping that env var back to false makes it deadlock again.

Thanks for looking into this!

[zbowling]

Awesome. Thanks for letting me know. Still trying to get a repro internally but the fact that new env override works narrows it down a bit. It’s something about specific network environments I can’t replicate and probably an updated crash reporting, telemetry, and SSL library not all playing well together.

If you don’t want to keep setting the env you can also now do ‘magic telemetry --disable’ and it should persist and avoid the issue on that build.

[ZoaiCapital]

Thanks for the quick update and the patch!

The pre-release build of magic 0.6.0 resolved the deadlocking issue for me. I used the provided curl command to install the pre-release and followed the instructions:

export MAGIC_TELEMETRY=false
magic init my-project

Everything is working smoothly now. Thanks for your help!