Skip to content
This repository has been archived by the owner on Apr 24, 2021. It is now read-only.

Security: googlearchive/compute-hadoop-java-python

Security

docs/SECURITY

User <-> Coordinator:
  - Encrypted by HTTPS
  - The coordinator authenticates the user based on a secret token created by
    tools/gen_passwd.py. The secret is passed to the coordinator upon creation
    and served via the metadata server, all of which are secure channels.
  - WARNING: The user does NOT know they are talking to the true coordinator,
    since the coordinator uses a self-signed certificate. For this reason, the
    system is vulnerable to man-in-the-middle attacks.

Coordinator <-> Snitch:
  - Encrypted by HTTPS
  - Authenticated by the source host's address. All instances have internal IPs
    in the 10.* range.

Hadoop nodes:
  - All instances have external IPs. Only incoming ssh traffic is enabled by
    the default firewall, for debugging.
  - There is no extra security implemented on top of Hadoop.

Service accounts:
  - See https://developers.google.com/compute/docs/authentication for an
    introduction to service accounts.
  - The coordinator runs with RW Compute and Google Storage permissions.
  - All slaves run with RO Google Storage permissions.
  - The hadoop-namenode instance runs with RW Google Storage permissions.

There aren’t any published security advisories