Skip to content
Build the astropy, sep, numpy, scipy packages and strip them down to run in Lambda
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
LICENSE.txt
README.md
build.sh
process.py
requirements.txt

README.md

astropy-sep-lambda

Build the astropy, sep, numpy, and scipy packages and strip them down to run in Lambda

This repo contains a build.sh script that's intended to be run in an Amazon Linux docker container, and build astropy, sep, numpy, and scipy for use in AWS Lambda. For more info about how the script works, and how to use it, see the blog post by ryansb on deploying sklearn to Lambda.

This repository is a fork of the excellent work by ryanb in https://github.com/ryansb/sklearn-build-lambda .

To build the zipfile, pull the Amazon Linux image and run the build script in it.

$ docker pull amazonlinux:2017.09
$ docker run -v $(pwd):/outputs -it amazonlinux:2017.09 /bin/bash /outputs/build.sh

That will make a file called venv.zip in the local directory that's around 50MB.

Once you run this, you'll have a zipfile containing astropy, sep, numpy, scipy and their dependencies. This repository also contains a file called process.py which imports these packages and detects sources with SEP package (based on Source Extractor).

import os
import subprocess
import uuid

libdir = os.path.join(os.getcwd(), 'lib')

import warnings
from astropy.utils.data import CacheMissingWarning
warnings.simplefilter('ignore', CacheMissingWarning)

import boto3
import glob
import numpy as np
import sep

from astropy import wcs
from astropy import units as u
import astropy.io.fits as fits
import astropy.io.ascii as ascii
from astropy.table import Table, Column
from astropy.stats import sigma_clipped_stats, gaussian_fwhm_to_sigma, median_absolute_deviation

GAUSS_3_7x7 = np.array(
[[ 0.004963,  0.021388,  0.051328,  0.068707,  0.051328,  0.021388,  0.004963],
 [ 0.021388,  0.092163,  0.221178,  0.296069,  0.221178,  0.092163,  0.021388],
 [ 0.051328,  0.221178,  0.530797,  0.710525,  0.530797,  0.221178,  0.051328],
 [ 0.068707,  0.296069,  0.710525,  0.951108,  0.710525,  0.296069,  0.068707],
 [ 0.051328,  0.221178,  0.530797,  0.710525,  0.530797,  0.221178,  0.051328],
 [ 0.021388,  0.092163,  0.221178,  0.296069,  0.221178,  0.092163,  0.021388],
[ 0.004963, 0.021388, 0.051328, 0.068707, 0.051328, 0.021388, 0.004963]])

def detect_with_sep(event, detect_thresh=2., npixels=8, grow_seg=5,
                          gauss_fwhm=2., gsize=3, im_wcs=None, root='mycat'):

...
...
...

Extra Packages

To add extra packages to the build, add them to the requirements.txt file alongside the build.sh in this repo. All packages listed there will be installed in addition to those already described in build.sh

Testing locally

Testing Lambda locally is a pain, but thanks to the efforts of the Lambci folks in https://github.com/lambci/docker-lambda, we can test a function locally as follows:

First, build the Lambda function locally with the command from above:

$ docker run -v $(pwd):/outputs -it amazonlinux:2017.09 /bin/bash /outputs/build.sh

This should leave you with a venv.zip file. Unzip this with:

$ unzip venv.zip -d venv

Next enter the venv directory and try running the command, passing in your AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY as environment variables:

$ cd venv
$ docker run --rm -e AWS_ACCESS_KEY_ID='XXXXXXXXX' -e AWS_SECRET_ACCESS_KEY='XXXXXXXXX' -v "$PWD":/var/task lambci/lambda:python2.7 process.handler '{"s3_output_bucket": "dsmo-lambda-test-outputs", "fits_s3_key":"hst/public/icsc/icsca0voq/icsca0voq_drz.fits", "fits_s3_bucket":"stpubdata"}'

Sizing and Future Work

In its current form, this set of packages weighs in at 50MB and could probably be reduced further by:

  1. Pre-compiling all .pyc files and deleting their source
  2. Removing test files
  3. Removing documentation

According to this article the size limit for a zipped Lambda package (the venv.zip file) is 50MB, however, reading around it seems like Lambda is tolerant of significantly larger packages when the zipped package is posted to S3.

You can’t perform that action at this time.