Snippet that uploads a local repo directory to an s3-backed one

sync-repo-to-s3.bash

# Don't forget the trailing slash
REPO_BUCKET="s3://bucketname/"

sync-repo-to-s3() {
  # Of course, this isn't as simple as just syncing the local repo to S3.
  REPO_NAME=$(basename "$1")
  SCRATCH_DIR="/tmp/daqri-repo-uploader.$$"
  # We assume you've already run `dpkg-scanpackages` in the local copy of the repo

  # Construct an s3 path that the s3 transport is happy with
  # the s3 apt transport is expecting a path like:
  #
  # s3://s3.amazonaws.com/foobar/dists/s3repo/main/binary-amd64/Packages
  #
  # to correspond with a sources.list entry of
  # deb s3://ACCESSKEY:[SECRET]@s3.amazonaws.com/foobar s3repo main
  s3path="${REPO_BUCKET}dists/${REPO_NAME}/main/binary-amd64/"
  echo "Publishing to ${s3path}"
  cd "$1" || croak "Could not cd to $1"

  echo "Publishing debs..."
  aws s3 sync . "${s3path}" || croak "Could not sync debs to ${s3path}"

  echo "Massaging Packages.gz into shape..."
  mkdir -p "${SCRATCH_DIR}"
  # Annoyingly, while we have to muck with the s3path to humor the s3 apt
  # transport plugin, the filenames in Packages.gz are just ./debName.deb
  # which of course causes the plugin to try to find the debs in
  # s3://foobar/debName, which clearly won't work.
  #
  # We can't just rewrite Packages.gz in place on the apt-server, or
  # connections via http will break with bad paths instead, so we're going
  # to edit a copy to have paths compliant with the sources.list described
  # above and then push _that_ Packages.gz to s3 instead.
  #
  # TODO: Figure out a better solution so we can get rid of this ugly hack.
  gzip -dc Packages.gz | \
    sed "s/^Filename: \./Filename: dists\/$REPO_NAME\/main\/binary-amd64/" > "${SCRATCH_DIR}/Packages"
  gzip -9v "${SCRATCH_DIR}/Packages"
  s3cmd --config="${S3_CREDENTIAL_PATH}" put "${SCRATCH_DIR}/Packages.gz" "${s3path}" || croak "Could not update ${s3path}/Packages.gz"
  # gzip -dc "${SCRATCH_DIR}/Packages.gz"
  rm -fr "${SCRATCH_DIR}"
}