AWS: Use an elastic IP for the master by default

If we don't use an elastic IP, the IP address will be lost if we lose
the master for any reason, and a replacement master will not have the
same IP.  But the master IP is set both in client kubeconfig files and
the master SSL certificate.  Hence the default should be to allocate an
elastic IP for the master.

One complication: AWS doesn't allow tags on elastic IPs, so it is hard
to track the elastic IP so we can delete it as part of kube-down.
Instead, we take the master EBS volume with the elastic IP.  This is a
little odd, but works because the master volume & the master elastic IP
really need to be assigned to the same machine, so might be thought of
as a pair.

Also, we now delete the master EBS volume as part of kube-down, as
people expect kube-down to clean-up everything it creates.
This commit is contained in:
Justin Santa Barbara
2016-01-07 23:31:03 -05:00
parent ca69c2e310
commit 274e589aa6
3 changed files with 84 additions and 38 deletions

View File

@@ -75,9 +75,9 @@ POLL_SLEEP_INTERVAL=3
SERVICE_CLUSTER_IP_RANGE="10.0.0.0/16" # formerly PORTAL_NET
CLUSTER_IP_RANGE="${CLUSTER_IP_RANGE:-10.245.0.0/16}"
MASTER_IP_RANGE="${MASTER_IP_RANGE:-10.246.0.0/24}"
# If set to Elastic IP, master instance will be associated with this IP.
# If set to auto, a new Elastic IP will be acquired
# Otherwise amazon-given public ip will be used (it'll change with reboot).
# If set to an Elastic IP address, the master instance will be associated with this IP.
# Otherwise a new Elastic IP will be acquired
# (We used to accept 'auto' to mean 'allocate elastic ip', but that is now the default)
MASTER_RESERVED_IP="${MASTER_RESERVED_IP:-}"
RUNTIME_CONFIG="${KUBE_RUNTIME_CONFIG:-}"