Bug: Should soft fail and continue on fallocate error on AWS Lambda

[heri16]

Goals

Trying to run realm-js on nodejs 8.10 AWS Lambda.

Expected Results

Should be able to open a realm path on /tmp

Actual Results

{
    "errorMessage": "posix_fallocate() failed: Operation not permitted",
    "errorType": "Error",
    "stackTrace": [
        "Function.open (/var/task/node_modules/realm/lib/extensions.js:72:47)",
        "exports.handler (/var/task/api/realm.js:45:25)"
    ]
}

Steps to Reproduce

1. Create a serverless framework project.
2. npm install realm.
3. Deploy the function below: sls deploy -f realm
4. Try invoking the function below: sls invoke -f realm --data "{'test': 'ok'}"

Code Sample

'use strict'

// See: https://realm.io/docs/javascript/latest/#getting-started

const process = require('process')

const debug = require('debug')('realm')
const Realm = require('realm')

const CarSchema = {
  name: 'Car',
  properties: {
    make: 'string',
    model: 'string',
    miles: { type: 'int', default: 0 },
  },
}

// Define new ES6 class
class Person {
  get fullName() {
    return `${this.firstName} ${this.lastName}`
  }

  static get schema() {
    return {
      name: 'Person',
      properties: {
        name: 'string',
        birthday: 'date',
        cars: 'Car[]',
        picture: 'data?', // optional property
      },
    }
  }
}

exports.handler = async (event, context) => {
  console.log('Received event', JSON.stringify(event, 3)) // eslint-disable-line no-console
  console.log('Received context', JSON.stringify(context, 3)) // eslint-disable-line no-console

  let realm
  try {
    process.chdir('/tmp')
    realm = await Realm.open({
      path: '/tmp/cars.realm',
      inMemory: true,
      schema: [CarSchema, Person],
    })
  } catch (err) {
    console.error(err) // eslint-disable-line no-console
    throw err
  }

  // Create Realm objects and write to local storage
  realm.write(() => {
    const myCar = realm.create('Car', {
      make: 'Honda',
      model: 'Civic',
      miles: 1000,
    })
    myCar.miles += 20 // Update a property value
  })

  // Query Realm for all cars with a high mileage
  const cars = realm.objects('Car').filtered('miles > 1000')

  // Will return a Results object with our 1 car
  debug('cars.length:', cars.length) // => 1

  // Add another car
  realm.write(() => {
    realm.create('Car', {
      make: 'Ford',
      model: 'Focus',
      miles: 2000,
    })
  })

  // Query results are updated in realtime
  debug('cars.length:', cars.length) // => 2
  console.log('cars.length:', cars.length) // eslint-disable-line no-console

  return {
    statusCode: 200,
    headers: {
      'Content-Type': 'application/json',
    },
    body: JSON.stringify(cars),
  }
}## Version of Realm and Tooling

• Realm JS SDK Version: ? Latest
• Node or React Native: ? Nodejs v8.10
• Client OS & Version: AWS Lambda Runtime
• Which debugger for React Native: None

[kneth]

Sorry for the late response. The only thing I can suggest is that you check the permissions of /tmp.

[heri16]

I am sorry, this is not a user-help issue. There is no benefit from checking the permissions of /tmp when the PaaS provider is the one that controls it.

Realm-core needs to be fixed to dynamically account for environments that do not support sparse files.

[kneth]

@heri16 Thanks for the hint on sparse files.

@finnschiermer Can you enlighten us on sparse files and mmap()?

[kneth]

@heri16 Do you know which file system /tmp is? You can see it from the output of mount?

[jeddWebill]

Hi Is there any update on this?

[heri16]

@kneth it it the ephemeral filesystem provided on AWS Lambda.

[heri16]

Instead of doing of programming for every type of POSIX filesystem out there, should do defensive programming instead.

[kneth]

@heri16 @jeddWebill @emanuelez

We have a fix coming up in #2178. I plan to release it as v2.22.0-beta.1 but I will notify you when happens.

[kneth]

@heri16 @jaggs6 I have release a fix for AWS Lambda as v2.22.0-beta.1. I kindly ask you to try it out and tell me if it works out for you. If it does, I will release v2.22.0 Wednesday or Thursday next week.

[kneth]

v2.22.0-beta.1 doesn't fix it. But I don't think in it in our code. Running fallocate -l 1M foo.img give the same error.

[mithunph]

Hi @kneth @bmunkholm ,
Is the issue fixed now? I am still facing the same issue

[kneth]

@mithunph Unfortunately not. AWS support hasn't been able to explain why it doesn't work.

[mithunph]

Hi @kneth ,
Will it work atleast under EC2 instance instead of AWS Lambda(serverless)?

[bmunkholm]

@mithunph Absolutely. Our whole Cloud is running there :-)

[mithunph]

Thanks @bmunkholm @kneth

[kneth]

We have a proper fix in #4023