ALOHA

What's with the name?

Pretty obvious, duh!

ALOHA == zALOra jmemcacHed infinispAn

aloha replaces your traditional memcached setup with a clusterable, HA memory grid, which still speaks and behaves like memcached

tl;dr – Give me Memcached!

mvn clean package && java -jar target/aloha-*.jar

This gives you a single local aloha instance with one memcached container listening on 11211

The Architecture

The setup is built on top of these libraries:

• Infinispan for the memory grid
• spring-boot because it's great!

Configuration

aloha has 2 configuration files:

• bootstrap.yml
• config/application.yml

bootstrap.yml

Currently only the app name is defined here. If you wanted to add support for a config server, that would be the place to go.

config/application.yml

This is the heart of aloha, you can configure nearly every aspect of the app. You can override every setting in the application.yml via command-line properties, e.g. turn on the socket server:

java -Dsocket.enabled=true -jar aloha.jar

infinispan

infinispan:
  cluster:
    jgroups:
    name: Kamehameha

  cache:
    name: memcachedCache
    mode: LOCAL
    numOwners: 3 # Amount of machines in the cluster storing the key
    stateTransferChunkSize: 128
    lock:
      timeout: 30 # Lock acquisition timeout in seconds
      concurrency: 1024

By default Aloha is configured to provide a single local cache which can be accessed with the Memcached protocol. Internally infinispan uses a Cache<String, byte[]> type to store the data.

If you let jgroups point to a jgroups file (Aloha comes with two: /default-configs/default-jgroups-(tcp|udp).xml), you can use the clustering features of infinispan, which is documented here: http://infinispan.org/docs/stable/user_guide/user_guide.html#clustering

To start a REPL_SYNC cluster, you can run ./start-node1 and ./start-node2.

memcached

Since Infinispan 9.1.1, we can use the Memcached server, which comes with Infinispan again, because a bug we reported was fixed in this version.

memcached:
  enabled: true
  host: 0.0.0.0
  port: 11211
  idleTime: -1
  recvBufSize: 0
  sendBufSize: 0
  tcpNoDelay: true

In theory you can configure the name of the cache to use, but unless it's set to "memcachedCache", the server fails to boot up. The configuration parameters in the application.yml represent the defaults of configuration-builder.

hotrod

By default the hotrod server is disabled, it can be enabled via a cli parameter or by changing the config file directly. We increased the timeouts, because it solved problems we had on shitty networks. Like on AWS.

hotrod:
  enabled: false
  host: 0.0.0.0
  topologyLockTimeout: 30000 # ms
  topologyReplTimeout: 30000 # ms

socket

We added a netty-based unix domain socket server to eliminate latency when the server is running on the same machine than the client.

I'm sure there's plenty of potential for optimization, so PRs are welcome! The socket is read-only and it supports get and multi-get. Quick test on the cli using openbsd's netcat:

Single Get:

echo mykey | nc -U /tmp/sockythesock.sock

The result of a single get is only the value.

Multi Get:

echo "mykey1 mykey2" | nc -U /tmp/sockythesock.sock

The result of a multi-get is a bit more verbose:

mykey1
myvalue1
mykey2
myvalue2

Configuration is self-explaning:

socket:
  enabled: false
  path: /tmp/sockythesock.sock

Health

For now we only have the health overview provided by spring at /health, which shows you if the system is up or not.

Jolokia

For some more insight into the system and the cluster status, you can query jolokia at /jolokia