Skip to content
/ mero Public

scalable and lightweight OTP Erlang client for memcached

License

Notifications You must be signed in to change notification settings

AdRoll/mero

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Mero

Mero is a scalable and lightweight OTP Erlang client for memcache. Mero allows interaction with different clusters, specifying different number of pools per server and sharding algorithms per cluster.

Mero achieves high-performance in a number of ways. Mero does not use an Erlang process per socket. Each pool worker adjusts the number of available sockets based on demand and sockets are shared to the point of usage, avoiding copying of terms through the system, as is present in the per-process design common in other pooling libraries. All pools are registered with a local name and its workers are created on startup, removing pool selection bottlenecks and worker creation latency.

If a connection to the memcached server fails there is a mechanism to delay connection retries. All the connections are renewed every time interval.

The storage module is configurable so you can use different protocols or even use a storage different than memcache.

It includes a callback that will be called to notify of specific error events. These events have the form of:

{Id :: list(atoms),
  Args :: list([{Key :: cluster_name | host | port | error_reason,
                 Value :: term()}])
}

Example Ids are:

  • [socket, connect, ok]
  • [socket, connect, error]
  • [socket, send, error]
  • [socket, rcv, error]
  • [socket, controlling_process, error]

Configuration

Please consult mero.app.src to see all the available options. The sharding algorithms available are shard_phash2 and shard_crc32.

  [{cluster_a,
     [{servers, [{"server1", 11211},
                 {"server2", 11211},
                 {"server3", 11211},
                 {"server4", 11211},
                 {"server5", 11211},
                 {"server6", 11211},
                 {"server7", 11211},
                 {"server8", 11211}]},
      {sharding_algorithm, {mero, shard_phash2}}, %% Module and function
      {workers_per_shard, 3},                         %% Number of pools that each server will have
      {pool_worker_module, mero_wrk_tcp_txt}]
  },

  {cluster_b,
     [{servers, [{"localhost", 11211}]},
      {sharding_algorithm, {mero, shard_crc32}},
      {workers_per_shard, 1},
      {pool_worker_module, mero_wrk_tcp_txt}]
  },
  ...
]

Cluster Auto Discovery

Configuration can also be implemented to support auto discovery of the cluster as opposed to hardcoding nodes. Provide the configuration endpoint (AWS Reference) and port as in the following example.

    [{cluster_b,
        [{servers, {elasticache, "ConfigEndpointHostname.com", PortNumber}},
             {sharding_algorithm, {mero, shard_crc32}},
             {workers_per_shard, 1},
             {pool_worker_module, mero_wrk_tcp_binary}]
    },
    ...
]

If instead of using ElastiCache a custom auto discovery is needed you can provide your own function (specified as a MFA) that returns {ok, list({Host :: inet:hostname(), Port :: inet:port_number()})} to be used. Example:

    [{cluster_b,
        [{servers, {mfa, {my_module, my_function, []}}},
             {sharding_algorithm, {mero, shard_crc32}},
             {workers_per_shard, 1},
             {pool_worker_module, mero_wrk_tcp_binary}]
    },
    ...
]
-module(my_module).

-export([my_function/0]).

my_function() ->
  {ok, [{"my_memcached_server", 11211}]}.

ElastiCache Multiple clusters Auto Discovery

We also support the setup of multiple physical clusters with autodiscovery assigned to the same logical cluster. It could be the case in which some of these physical clusters perform better than others, in which you can use a third argument called the ClusterSpeedFactor which has to be a small integer. The default ClusterSpeedFactor is 1.

If an alternate cluster is 2 times faster than the fist one -it can have twice as many cpus or memory-, you can set it up with a ClusterSpeedFactor of 2. This will create 2 times more workers for that "faster cluster", which in practice will send twice as many connections & requests to that cluster than to the other weaker clusters.

    [{cluster_c,
        [{servers,
           {elasticache,
            [{"ConfigEndpointHostname.com", 11211},
             {"ConfigEndpointHostnameAltThatPerforms2TimesBetter.com", 11211, 2},
             {"ConfigEndpointHostnameAltThatPerforms3TimesBetter.com", 11211, 3}]
            }},
         {sharding_algorithm, {mero, shard_crc32}},
         {workers_per_shard, 1},
         {pool_worker_module, mero_wrk_tcp_binary}]}]

Auto Discovery Changes

Mero will also monitor elasticache or your custom provided MFA for changes in the configuration. It will poll the config provider according to the values of conf_monitor_min_sleep and conf_monitor_max_sleep in the configuration. The polling will run at a random interval that will always be between those two values (in milliseconds).

If a cluster configuration change is found after a poll, the connections to the servers of that cluster will be restablished. This is done using OTP supervision principles: Mero maintains a supervision tree per cluster, which is stopped and restarted if its configuration changes.

Using Mero

Mero is a regular OTP application, managed with rebar3. Therefore you can do stuff like...

rebar3 do compile, xref, eunit, ct

...or...

rebar3 test

There are three ways to start this application:

From an erlang shell

$rebar3 shell

> mero:increment_counter(default, <<"key">>).
{ok,1}

> mero:increment_counter(default, <<"key">>).
{ok,2}

> mero:get(default, <<"key">>).
{<<"key">>,<<"2">>}

> mero:set(default, <<"key">>, <<"5">>, 3600, 5000).
ok

> mero:increment_counter(default, <<"key">>).
{ok,6}

> mero:set(default, <<"key">>, <<"50">>, 3600, 5000).
ok

> mero:increment_counter(default, <<"key">>).
{ok,51}

> mero:increment_counter(default, <<"key2">>).
{ok,1}

> mero:increment_counter(default, <<"key3">>).
{ok,1}

> mero:increment_counter(default, <<"key4">>).
{ok,1}

> mero:mget(default, [<<"key">>, <<"key2">>, <<"key3">>, <<"key4">>], 5000).
[{<<"key">>,<<"51">>},
 {<<"key2">>,<<"1">>},
 {<<"key3">>,<<"1">>},
 {<<"key4">>,<<"1">>}]

> mero:set(default, <<"key">>, <<"key">>, 3600, 5000).
ok

> mero:set(default, <<"key2">>, <<"key2">>, 3600, 5000).
ok

> mero:increment_counter(default, <<"key">>).
{error,incr_decr_on_non_numeric_value}
=ERROR REPORT==== 3-Apr-2015::13:57:40 ===
    error: memcached_request_failed
    client: {client,#Port<0.3562>,undefined,
                    {mero,stat_event_callback,
                              [{cluster_name,default},
                               {host,"localhost"},
                               {port,11211}]}}
    cmd: {5,{<<"key">>,<<"1">>,<<"1">>,<<"86400">>}}
    reason: incr_decr_on_non_numeric_value

> mero:mget(default, [<<"key">>, <<"key2">>, <<"key3">>, <<"key4">>], 5000).
[{<<"key">>,<<"key">>},
 {<<"key2">>,<<"key2">>},
 {<<"key3">>,<<"1">>},
 {<<"key4">>,<<"1">>}]

> mero:flush_all(default).
[{default,ok}]


> mero:mget(default, [<<"key">>, <<"key2">>, <<"key3">>, <<"key4">>], 5000).
[{<<"key">>,undefined},
 {<<"key2">>,undefined},
 {<<"key3">>,undefined},
 {<<"key4">>,undefined}]

> mero:add(default, <<"key">>, <<"value">>, 1000, 5000).
ok

> mero:add(default, <<"key">>, <<"value">>, 1000, 5000).
{error,already_exists}

=ERROR REPORT==== 3-Apr-2015::14:03:25 ===
    error: memcached_request_failed
    client: {client,#Port<0.3558>,undefined,
                    {mero,stat_event_callback,
                              [{cluster_name,default},
                               {host,"localhost"},
                               {port,11211}]}}
    cmd: {2,{<<"key">>,<<"value">>,<<"1000">>}}
    reason: already_exists

> {mero:madd(default, [{<<"foo">>, <<"bar">>, 1000},
                       {<<"bar">>, <<"foo">>, 1000},
                       {<<"foo">>, <<"baz">>, 1000}], 5000),
   mero:mget(default, [<<"foo">>, <<"bar">>], 5000)}.
{[ok,ok,{error,already_exists}],
 [{<<"bar">>,<<"foo">>},{<<"foo">>,<<"bar">>}]}

> mero:mcas(default, [{<<"xyzzy">>, <<"bar">>, 0, 360391},
                      {<<"qwer">>, <<"asdfsdf">>, 0, 360390}], 5000).
[ok,{error,already_exists}]

Inside a Node

Set the configuration in the mero.app.src file and start the application inside your OTP node as a regular OTP app.

As an OTP included application

Pass the ClusterConfiguration as a parameter to the supervisor of the application.

mero_sup:start_link([{default,[{servers,[{"localhost",11211}]},
                                   {sharding_algorithm,{mero,shard_crc32}},
                                   {workers_per_shard,1},
                                   {pool_worker_module,mero_wrk_tcp_txt}]}]).

Testing the library against a local memcached server:

Warning: This will erase all the contents of the memcached server it connects to ("localhost" by default).

To run tests using a real memcached server, uncomment the test cases at test/mero_test_with_local_memcached_SUITE.erl