aioftps3

FTP in front of AWS S3, using asyncio, and aiohttp. Only a subset of the FTP protocol is supported, with implicit TLS and PASV mode; connections will fail otherwise.

Installation

pip install aioftps3

An SSL key and certificate must be present $HOME/ssl.key and $HOME/ssl.crt respectively. To create a self-signed certificate, you can use openssl.

openssl req -new -newkey rsa:2048 -days 3650 -nodes -x509 -subj /CN=selfsigned \
    -keyout $HOME/ssl.key \
    -out $HOME/ssl.crt

Running

python -m aioftps3.server_main

Configuration

Configuration is through environment variables

Advanced usage

The code in aioftps3.server_main satisfies a very particular use case, which may not be useful to most. However, the bulk of the code can be used for other cases: you will have to write your own aioftps3.server_main-equivalent, using the functions aioftps3.server.on_client_connect and aioftps3.server_socket.server. For example, you could

• Store credentials, appropriately hashed, differently, .e.g. in a database.
• Have the credentials hashed differently.
• Allow/deny PASV mode data connections based on some condition.

See the source of aioftps3.server_main for how these functions can be used.

Creating a password and salt

python ./create_password.py

Running tests

Certificates must be created, and Minio, which emulates S3 locally, must be started

./certificates-create.sh && ./minio-start.sh

and then to run the tests themselves.

./tests.sh

Features / Design / Limitations

• Can upload files bigger than 2G: uses multipart upload under the hood.

• Does not store uploading files in memory before uploading them to S3: i.e. it is effectively a streaming upload. However, it's not completely streaming: each part of multipart upload is stored in memory before it begins to transfer to S3, in order to be able to hash its content and determine its length.

• For uploading files, hashes are computed incrementally as data comes in in order to not block the event loop just before uploads to S3.

• As few dependencies as is reasonable: aiohttp and its dependencies. Boto 3 is not used.

• May not behave well if upload to the server is faster than its upload to S3.

• There is some locking to deal with the same files being operated on concurrently. However...

• .... it does nothing to deal with eventual consistency of S3, and so some operations may appear to not have an immediate effect.

Building and running locally

docker build -t ftps-s3 . && \
docker run --rm -p 8021-8042:8021-8042 \
  -e AWS_AUTH_MECHANISM=secret_access_key \
  -e AWS_ACCESS_KEY_ID=ommitted \
  -e AWS_SECRET_ACCESS_KEY=ommitted \
  -e AWS_S3_BUCKET_REGION=eu-west-1 \
  -e AWS_S3_BUCKET_HOST=s3-eu-west-1.amazonaws.com \
  -e AWS_S3_BUCKET_NAME=my-bucket-name \
  -e AWS_S3_BUCKET_DIR_SUFFIX=/ \
  -e FTP_USERS__1__LOGIN=user \
  -e FTP_USERS__1__PASSWORD_HASHED=ommitted \
  -e FTP_USERS__1__PASSWORD_SALT=ommitted \
  -e FTP_COMMAND_PORT=8021 \
  -e FTP_DATA_PORTS_FIRST=4001 \
  -e FTP_DATA_PORTS_COUNT=2 \
  -e FTP_DATA_CIDR_TO_DOMAINS__1__CIDR=0.0.0.0/0 \
  -e FTP_DATA_CIDR_TO_DOMAINS__1__DOMAIN=0.0.0.0 \
  -e HEALTHCHECK_PORT=8022
  ftps-s3

Building and pushing to Quay

docker build -t ftps-s3 . && \
docker tag ftps-s3:latest quay.io/uktrade/ftps-s3:latest && \
docker push quay.io/uktrade/ftps-s3:latest

Building and pushing healthcheck application to Quay

docker build -t ftps-s3-healthcheck . -f Dockerfile-healthcheck && \
docker tag ftps-s3-healthcheck:latest quay.io/uktrade/ftps-s3-healthcheck:latest && \
docker push quay.io/uktrade/ftps-s3-healthcheck:latest

Building and pushing Minio, used for testing, to Quay

docker build -t ftps-s3-minio . -f Dockerfile-minio && \
docker tag ftps-s3-minio:latest quay.io/uktrade/ftps-s3-minio:latest && \
docker push quay.io/uktrade/ftps-s3-minio:latest