Ekka

Ekka - Autocluster for EMQX Broker. Ekka helps building a new distribution layer for EMQ X R2.3+.

----------             ----------
|  EMQX  |<--- MQTT--->|  EMQX  |
|--------|             |--------|
|  Ekka  |<----RPC---->|  Ekka  |
|--------|             |--------|
| Mnesia |<--Cluster-->| Mnesia |
|--------|             |--------|
| Kernel |<----TCP---->| Kernel |
----------             ----------

Node discovery and Autocluster

Ekka supports erlang node discovery and autocluster using various strategies:

The configuration example files are under 'etc/' folder.

Cluster using static node list

Cuttlefish style config:

cluster.discovery = static

cluster.static.seeds = ekka1@127.0.0.1,ekka2@127.0.0.1

Erlang config:

{cluster_discovery,
  {static, [
    {seeds, ['ekka1@127.0.0.1', 'ekka2@127.0.0.1']}
  ]}},

Cluster using DNS A records

Cuttlefish style config:

cluster.discovery = dns

## DNS name.
##
## Value: String
cluster.dns.name = localhost

## The App name is used to build 'node.name' with IP address.
##
## Value: String
cluster.dns.app = ekka

Erlang config:

{cluster_discovery,
  {dns, [
    {name, "localhost"},
    {app, "ekka"}
  ]}},

Cluster using etcd

Cuttlefish style config:

cluster.discovery = etcd

## Etcd server list, seperated by ','.
##
## Value: String
cluster.etcd.server = http://127.0.0.1:2379

## The prefix helps build nodes path in etcd. Each node in the cluster
## will create a path in etcd: v2/keys/<prefix>/<cluster.name>/<node.name>
##
## Value: String
cluster.etcd.prefix = ekkacl

## The TTL for node's path in etcd.
##
## Value: Duration
##
## Default: 1m, 1 minute
cluster.etcd.node_ttl = 1m

## Path to a file containing the client's private PEM-encoded key.
##
## Value: File
##
## cluster.etcd.keyfile = {{platform_etc_dir}}/certs/client-key.pem

## Path to the file containing the client's certificate
##
## Value: File
##
## cluster.etcd.certfile = {{platform_etc_dir}}/certs/client.pem

## Path to the file containing PEM-encoded CA certificates. The CA certificates
## are used during server authentication and when building the client certificate chain.
##
## Value: File
##
## cluster.etcd.cacertfile = {{platform_etc_dir}}/certs/ca.pem

Erlang config:

{cluster_discovery,
  {etcd, [
    {server, ["http://127.0.0.1:2379"]},
    {prefix, "ekkacluster"},
    %%{ssl_options, [
    %%    {keyfile, "path/to/client-key.pem"},
    %%    {certfile, "path/to/client.pem"},
    %%    {cacertfile, "path/to/ca.pem"}
    %%]},
    {node_ttl, 60000}
  ]}},

Cluster using Kubernates

Cuttlefish style config:

cluster.discovery = k8s

## Kubernates API server list, seperated by ','.
##
## Value: String
## cluster.k8s.apiserver = http://10.110.111.204:8080

## The service name helps lookup EMQ nodes in the cluster.
##
## Value: String
## cluster.k8s.service_name = ekka

## The name space of k8s
##
## Value: String
## cluster.k8s.namespace = default

## The address type is used to extract host from k8s service.
##
## Value: ip | dns | hostname
## cluster.k8s.address_type = ip

## The app name helps build 'node.name'.
##
## Value: String
## cluster.k8s.app_name = ekka

## The suffix added to dns and hostname get from k8s service
##
## Value: String
## cluster.k8s.suffix = pod.cluster.local

Erlang config:

{cluster_discovery,
  {k8s, [
    {apiserver, "http://10.110.111.204:8080"},
    {namespace, "default"},
    {service_name, "ekka"},
    {address_type, ip},
    {app_name, "ekka"},
    {suffix, "pod.cluster.local"}
  ]}}

Network partition and Autoheal

Autoheal Design

When network partition occurs, the following steps to heal the cluster if autoheal is enabled:

1. Node reports the partitions to a leader node which has the oldest guid.

2. Leader node create a global netsplit view and choose one node in the majority as coordinator.

3. Leader node requests the coordinator to autoheal the network partition.

4. Coordinator node reboots all the nodes in the minority side.

Enable autoheal

Erlang config:

{cluster_autoheal, true},

Cuttlefish style config:

cluster.autoheal = on

Lock Service

Ekka implements a simple distributed lock service in 0.3 release. The Lock APIs:

Acquire lock:

-spec(acquire(resource()) -> {boolean(), [node()]}).
ekka_locker:acquire(Resource).

-spec(acquire(atom(), resource(), lock_type()) -> lock_result()).
ekka_locker:acquire(ekka_locker, Resource, Type).

Release lock:

-spec(release(resource()) -> lock_result()).
ekka_locker:release(Resource).

-spec(release(atom(), resource()) -> lock_result()).
ekka_locker:release(Name, Resource).

The lock type:

-type(lock_type() :: local | leader | quorum | all).

Cluster without epmd

The ekka 0.6.0 release implements erlang distribiton without epmd.

See: http://erlang.org/pipermail/erlang-questions/2015-December/087013.html

For example:

## Dist port: 4370
erl -pa ebin -pa _build/default/lib/*/ebin -proto_dist ekka -start_epmd false -epmd_module ekka_epmd -name node1@127.0.0.1 -s ekka
## Dist port: 4371
erl -pa ebin -pa _build/default/lib/*/ebin -proto_dist ekka -start_epmd false -epmd_module ekka_epmd -name node2@127.0.0.1 -s ekka
## Dist port: 4372
erl -pa ebin -pa _build/default/lib/*/ebin -proto_dist ekka -start_epmd false -epmd_module ekka_epmd -name node3@127.0.0.1  -s ekka

The erlang distribution port can be tuned by ekka inet_dist_base_port env. The default port is 4370.

License

Apache License Version 2.0

Author

EMQX Team.