This release fixes a number of stability issues with 1.1.2. We recommend that all users
of Chef Backend 1.1.2 and prior upgrade.

## Enhancements

- Fail follower health check if the node falls too far
behind leader. This is configurable with the
`leaderl.health_check.max_bytes_behind_leader` option.

- The following etcd configuration options are now
available for customization in `/etc/chef-backend/chef-backend.rb`:

    - `etcd.election_timeout`
    - `etcd.heartbeat_interval`
    - `etcd.snapshot_count`


- Bump etcd to 2.3.7

- Upgrade to postgresql 9.5.5

- Health Checks can now fail up to a configurable number
  of times before triggering a failover. Users can configure the
  maximum failures for each service via the following configuration
  settings:

    - `leaderl.health_check.max_elasticsearch_failures`
    - `leaderl.health_check.max_etcd_failures`
    - `leaderl.health_check.max_pgsql_failures`

  If you have manually set these flags in chef-backend.rb, your
  overrides will still be respected. Please remove these options if
  you'd prefer to use the new defaults instead.

- Add a `--force-basebackup` option to the `pgsql-follow`
  command. This allows for a more straightforward recovery procedure
  in cases where human intervention is needed.

- Report 503 from /leaderl if a user-specified number of
  followers haven't initiated replication connections.This is
  tunable with the `leaderl.required_active_followers` configuration
  option (default: 0)

- No longer set shmmax sysctl's as they are no longer
  required by postgresql 9.5. If you are upgrading from a previous
  installation, remember to remove postgresql.shmmax and
  postgresql.shmall from your chef-backend.rb.

- Update configuration defaults based on customer
  feedback. We've changed the following configuration defaults:

  - `leaderl.health_check.interval_seconds` from 2 to 5 seconds.
  - `leaderl.leaderl_ttl_seconds` from 10 to 30.
  - `etcd.heartbeat_interval` from 100 to 500 milliseconds.
  - `etcd.election_timeout` from 1000 to 5000 milliseconds.

  The new defaults have proven to reduce spurious failovers when
  deploying Chef Backend to various cloud providers. If you have
  manually set these flags in chef-backend.rb, your overrides will
  still be respected. Please remove these options if you'd prefer
  to use the new defaults instead.

## Bug Fixes

- `chef-server-ctl cluster-status` shouldn't fail because of
missing leader key.

- Force a checkpoint after promoting the local postgresql
instance, avoiding some cases of postgresql followers failing to
start syncrounous replication after a failover.

- chef-backend-ctl restore shouldn't fail with a
NoMethodError.

- Wait for postgresql to be started via a test connection
rather than a simple process check during the promotion and follow
processes. This fixes a number of cases were postgresql would fail
to promote during a failover event.

- Ensure that the leader key keeps the leader key as long as
it is still in the `promoting` state. This fixes some cases where
the cluster would become leaderless after a failover event because
the elected follower was slow to promote.

- Don't erroneously report etcd as down in `chef-backend-ctl
status` in cases where etcd is up but there is no leaderl leader.