rply causes a crash on read-only filesystems

[dsevero]

I currently have the following file setup on AWS Lambda:

challenge
├── handlers.py
├── hello.hy
├── __init__.hy
└── __init__.py

handlers.py

import hy
from hello import hello

hello.hy

(import [json [dumps]])

(defn hello [event context]
  (setv body {"message" "Go Serverless v1.0! Your function executed successfully!"
              "input" event})
  ({"statusCode" 200
    "body" (dumps body)}))

AWS lambda calls the hello function inside handlers.py. My objective was to use this file as a proxy for writing aws lambda functions with hylang. But I keep getting this error when executing it in production:

module initialization error
[Errno 30] Read-only file system: '/home/sbx_user1059'

(the number 1059 varies)

I think this is due to the AST compiling on importing hy in handlers.py. AWS lambda does not permit that directory to be written.

Any way that this could be circumvented?

[refi64]

You can disable bytecode compilation by setting the environment variable PYTHONDONTWRITEBYTECODE. The real bug here is that we're ignoring sys.dont_write_bytecode, which is set via python -B, which AWS probably uses.

[dsevero]

Still got the same error :/

[Kodiologist]

I don't think byte-compilation should be the issue, because Hy's call to write_code_as_pyc is wrapped in a try that catches IOError and OSError. The quoted error message doesn't include a stack trace, but the presence of an errno suggests it's an OSError.

By the way, @daniel-severo, this probably isn't your problem here, but it's a Bad Idea to have a Python module and a Hy module with the same name in the same directory, especially when that name is __init__.

[Kodiologist]

The quoted error message doesn't include a stack trace

The first thing to do here is probably to see how to convince AWS Lambda to give you the stack trace.

[dsevero]

@Kodiologist yeah, I'm gonna remove one of them. I'll try to investigate it further.

[vodik]

@kirbyfan64 not respecting sys.dont_write_bytecode is fixed in #1518

[dsevero]

Looks like the bytecode generation wasn't the problem. For some reason, after generating python code with hy2py, this import statement is the problem:

from hy.core.language import complement, distinct, drop_while, first, inc, iterate, last, reduce, take_while

Maybe something in the importer is trying to write data somewhere.

[Kodiologist]

Did you get a traceback? https://docs.aws.amazon.com/lambda/latest/dg/python-exceptions.html

[dsevero]

I was unable to get any more logging information from aws lambda.

Although, strangely enough, ti looks like it breaks BEFORE the actual import statement.

print("before everything")
from hy.core.language import complement, distinct, drop_while, first, inc, iterate, last, reduce, take_while

module initialization error
[Errno 30] Read-only file system: '/home/sbx_user1073'
before everything

[Kodiologist]

I was able to get a stack trace by putting the import hy inside the function called by AWS Lambda instead of the mainline code.

  "errorMessage": "[Errno 30] Read-only file system: '/home/sbx_user1065'",
  "errorType": "OSError",
  "stackTrace": [
    [
      "/var/task/handlers.py",
      2,
      "hello",
      "import hy"
    ],
    [
      "/var/task/hy/__init__.py",
      11,
      "<module>",
      "import hy.importer  # NOQA"
    ],
    [
      "/var/task/hy/importer.py",
      7,
      "<module>",
      "from hy.compiler import hy_compile, HyTypeError"
    ],
    [
      "/var/task/hy/compiler.py",
      11,
      "<module>",
      "from hy.lex.parser import mangle"
    ],
    [
      "/var/task/hy/lex/__init__.py",
      9,
      "<module>",
      "from hy.lex.parser import parser"
    ],
    [
      "/var/task/hy/lex/parser.py",
      374,
      "<module>",
      "parser = pg.build()"
    ],
    [
      "/var/task/rply/parsergenerator.py",
      194,
      "build",
      "os.makedirs(cache_dir, mode=0o0700)"
    ],
    [
      "/var/lang/lib/python3.6/os.py",
      210,
      "makedirs",
      "makedirs(head, mode, exist_ok)"
    ],
    [
      "/var/lang/lib/python3.6/os.py",
      210,
      "makedirs",
      "makedirs(head, mode, exist_ok)"
    ],
    [
      "/var/lang/lib/python3.6/os.py",
      220,
      "makedirs",
      "mkdir(name, mode)"
    ]
  ]
}

It seems that the guilty code is in rply.parsergenerator, which tries to write out a cache of the generated parser without catching any errors. I've opened an issue at alex/rply#74.

[mathvdh]

We can write to /tmp on AWS lambda ... We could have a way to configure where is the cache located (independently of the fact that it should write on read-only fs ..)

[Kodiologist]

@daniel-severo Did upgrading rply solve your problem?

[dsevero]

Sort of. It enabled me to deploy my code to aws lambda as long as I don't use anything from the shadow library. For some reason it acusses that the module isn't available.

Since I use toolz most of the time, this wasn't such a big issue.