[Feature Request] Is it possible to run Mojo in a Docker container on M2 Max silicon?

[devforfu]

Review Mojo's priorities

• I have read the roadmap and priorities and I believe this request falls within the priorities.

What is your request?

I am trying to run Mojo locally on my M2 Max silicon desktop. I know that it is not supported natively, but I thought that running in Docker container would work. However, it looks like it is not the case. Running in a container derived from ubuntu:latest returns the error.

no packages found for the current target hardware - 
please run `modular host-info` and file a ticket at 
/~https://github.com/modularml/mojo with your hardware information

Here is my Dockerfile.

FROM ubuntu:latest

RUN apt-get update -y && \
  apt-get -y install curl gpg-agent && \
  apt-get install -y apt-transport-https && \
  keyring_location=/usr/share/keyrings/modular-installer-archive-keyring.gpg && \
  curl -1sLf 'https://dl.modular.com/bBNWiLZX5igwHXeu/installer/gpg.0E4925737A3895AD.key' |  gpg --dearmor >> ${keyring_location} && \
  curl -1sLf 'https://dl.modular.com/bBNWiLZX5igwHXeu/installer/config.deb.txt?distro=debian&codename=wheezy' > /etc/apt/sources.list.d/modular-installer.list && \
  apt-get update && \
  apt-get install -y modular

RUN modular auth <auth-token> && modular install mojo

Does it mean that running Mojo on M2 chips isn't possible at the moment? I don't have in-depth knowledge of how Docker containers work. So, I don't know if it is an expected behavior and this kind of virtualization is just not supported (should use a "real" Linux machine), or I am doing it wrong.

Thank you!

What is your motivation for this change?

It would be great to have an option of running Mojo in a Docker container for systems that don't support it natively.

Any other details?

No response

[devforfu]

There is a Dockerfile in the repo, but it also fails when I am trying to build a container. This time, however, with a different error.

=> ERROR [3/6] RUN curl -fsSL https://repo.anaconda.com/ [...]
------
[...]
9.642 qemu-x86_64: Could not open '/lib64/ld-linux-x86-64.so.2': No such file or directory

[devforfu]

Ok, it looks like that one is an issue of running Docker on an Apple's silicon. There is a solution that helped me to get to the next building stage. (Still running.) Here is the change to the Dockerfile.

FROM --platform=linux/amd64 ubuntu:20.04

[jackos]

@devforfu it will fail when you try to run mojo as x86 emulation doesn't currently work, there's an open feature request for it here: #510 and we're actively working on native support.

For now you'll have to ssh to an x86-64 machine and install it there.

[devforfu]

@jackos ah, I see, thank you!

[HananyTolba]

I have just made available a docker imager with mojo ready for use : docker pull hananytolba/mojo:0.2.1 for use docker run -ti hananytolba/mojo:0.2.1 bash and inupt mojo

[getorca]

same issue, using the dockerfile /~https://github.com/modularml/mojo/blob/main/examples/docker/Dockerfile.mojosdk and opening in a dev container in vs code sort of works. It can run the python cells but it fails on mojo cells.

eg from https://www.modular.com/blog/an-easy-introduction-to-mojo-for-python-programmers fails on the cell:

from Tensor import Tensor
from DType import DType
from Range import range
from SIMD import SIMD
from Math import sqrt
from Time import now

let n: Int = 10_000_000
var a = Tensor[DType.float64](n)
var b = Tensor[DType.float64](n)

for i in range(n):
    a[i] = anp[i].to_float64()
    b[i] = bnp[i].to_float64()

with:

error: Expression [7]:23:1: no viable expansions found
fn __lldb_expr__7(inout __mojo_repl_arg: __mojo_repl_context__):
^

Expression [7]:25:28:   call expansion failed - no concrete specializations
    __mojo_repl_expr_impl__(__mojo_repl_arg, __get_address_as_lvalue(__mojo_repl_arg.`secs`.load().address), __get_address_as_lvalue(__mojo_repl_arg.`python_numpy_dist`.load().address), __get_address_as_lvalue(__mojo_repl_arg.`___lldb_expr_failed`.load().address), __get_address_as_lvalue(__mojo_repl_arg.`python_naive_dist`.load().address), __get_address_as_lvalue(__mojo_repl_arg.`time`.load().address), __get_address_as_lvalue(__mojo_repl_arg.`print_formatter`.load().address), __get_address_as_lvalue(__mojo_repl_arg.`np`.load().address), __get_address_as_lvalue(__mojo_repl_arg.`n`.load().address), __get_address_as_lvalue(__mojo_repl_arg.`bnp`.load().address), __get_address_as_lvalue(__mojo_repl_arg.`blist`.load().address), __get_address_as_lvalue(__mojo_repl_arg.`anp`.load().address), __get_address_as_lvalue(__mojo_repl_arg.`alist`.load().address))
                           ^

Expression [7]:29:1:     no viable expansions found
def __mojo_repl_expr_impl__(inout __mojo_repl_arg: __mojo_repl_context__, inout `secs`: __mlir_type.`!kgen.declref<@"$python"::@"$object"::@PythonObject>`, inout `python_numpy_dist`: __mlir_type.`!kgen.declref<@"$python"::@"$object"::@PythonObject>`, inout `___lldb_expr_failed`: __mlir_type.`!kgen.declref<@"$builtin"::@"$bool"::@Bool>`, inout `python_naive_dist`: __mlir_type.`!kgen.declref<@"$python"::@"$object"::@PythonObject>`, inout `time`: __mlir_type.`!kgen.declref<@"$python"::@"$object"::@PythonObject>`, inout `print_formatter`: __mlir_type.`!kgen.declref<@"$python"::@"$object"::@PythonObject>`, inout `np`: __mlir_type.`!kgen.declref<@"$python"::@"$object"::@PythonObject>`, inout `n`: __mlir_type.`!kgen.declref<@"$python"::@"$object"::@PythonObject>`, inout `bnp`: __mlir_type.`!kgen.declref<@"$python"::@"$object"::@PythonObject>`, inout `blist`: __mlir_type.`!kgen.declref<@"$python"::@"$object"::@PythonObject>`, inout `anp`: __mlir_type.`!kgen.declref<@"$python"::@"$object"::@PythonObject>`, inout `alist`: __mlir_type.`!kgen.declref<@"$python"::@"$object"::@PythonObject>`) -> None:
^

Expression [7]:43:26:       call expansion failed - no concrete specializations
  __mojo_repl_expr_body__()
                         ^

Expression [7]:31:3:         no viable expansions found
  def __mojo_repl_expr_body__() -> None:
  ^

Expression [7]:35:5:           could not simplify operator get_alignof(!lit.unresolved, #kgen.target<triple = "x86_64-unknown-linux-gnu", cpu = "znver1", features = "+64bit,+adx,+aes,+avx,+avx2,+bmi,+bmi2,+clflushopt,+clzero,+cmov,+crc32,+cx16,+cx8,+f16c,+fma,+fsgsbase,+fxsr,+lzcnt,+mmx,+movbe,+mwaitx,+pclmul,+popcnt,+prfchw,+rdrnd,+rdseed,+sahf,+sha,+sse,+sse2,+sse3,+sse4.1,+sse4.2,+sse4a,+ssse3,+xsave,+xsavec,+xsaveopt,+xsaves", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128", simd_bit_width = 256>)
    var a = Tensor[DType.float64](n)
    ^

expression failed to parse (no further compiler diagnostics)

but other mojo cells work... so I'm not sure if this is a separate issue.

[benz0li]

FYI

• (CUDA-based) JupyterLab MAX/Mojo docker stack #2647
• MAX/Mojo Data Science Dev Containers #2648

plus

• [docs] Add section 'Dev Container' to the Mojo stdlib docs #2664