Syn (v2)

Syn (short for synonym) is a global Process Registry and Process Group manager for Erlang and Elixir. Syn automatically manages addition / removal of nodes from the cluster, and is also able to recover from net splits.

Introduction

What is a Process Registry?

A global Process Registry allows registering a process on all the nodes of a cluster with a single Key. Consider this the process equivalent of a DNS server: in the same way you can retrieve an IP address from a domain name, you can retrieve a process from its Key.

Typical Use Case: registering on a system a process that handles a physical device (using its serial number).

What is a Process Group?

A global Process Group is a named group which contains many processes, possibly running on different nodes. With the group Name, you can retrieve on any cluster node the list of these processes, or publish a message to all of them. This mechanism allows for Publish / Subscribe patterns.

Typical Use Case: a chatroom.

What is Syn?

Syn is a Process Registry and Process Group manager that has the following features:

• Global Process Registry (i.e. a process is uniquely identified with a Key across all the nodes of a cluster).
• Global Process Group manager (i.e. a group is uniquely identified with a Name across all the nodes of a cluster).
• Any term can be used as Key and Name.
• A message can be published to all members of a Process Group (PubSub mechanism).
• Fast writes.
• Automatically handles conflict resolution (such as net splits).
• Configurable callbacks.
• Processes are automatically monitored and removed from the Process Registry and Process Groups if they die.

Notes

In any distributed system you are faced with a consistency challenge, which is often resolved by having one master arbiter performing all write operations (chosen with a mechanism of leader election), or through atomic transactions.

Syn was born for applications of the IoT field. In this context, Keys used to identify a process are often the physical object's unique identifier (for instance, its serial or MAC address), and are therefore already defined and unique before hitting the system. The consistency challenge is less of a problem in this case, since the likelihood of concurrent incoming requests that would register processes with the same Key is extremely low and, in most cases, acceptable.

In addition, write speeds were a determining factor in the architecture of Syn.

Therefore, Availability has been chosen over Consistency and Syn is eventually consistent.

Setup

For Elixir

Add it to your deps:

defp deps do
  [{:syn, "~> 2.1"}]
end

For Erlang

If you're using rebar3, add syn as a dependency in your project's rebar.config file:

{deps, [
  {syn, {git, "git://github.com/ostinelli/syn.git", {tag, "2.1.1"}}}
]}.

Or, if you're using Hex.pm as package manager (with the rebar3_hex plugin):

{deps, [
  {syn, "2.1.1"}
]}.

Ensure that syn is started with your application, for example by adding it in your .app file to the list of applications:

{application, my_app, [
    %% ...
    {applications, [
        kernel,
        stdlib,
        sasl,
        syn,
        %% ...
    ]},
    %% ...
]}.

API

Example code here below is in Erlang. Thanks to Elixir interoperability with Erlang, the equivalent code in Elixir is straightforward.

Process Registry

To register a process:

syn:register(Name, Pid) ->
    syn:register(Name, Pid, undefined).

syn:register(Name, Pid, Meta) -> ok | {error, Error}.

Types:
    Name = any()
    Pid = pid()
    Meta = any()
    Error = taken

You may re-register a process multiple times, for example if you need to update its metadata. When a process gets registered, Syn will automatically monitor it. You may also register the same process with different names.

Processes can also be registered as gen_server names, by usage of via-tuples. This way, you can use the gen_server API with these tuples without referring to the Pid directly.

Tuple = {via, syn, <<"your process name">>}.
gen_server:start_link(Tuple, your_module, []).
gen_server:call(Tuple, your_message).

To retrieve a Pid from a Name:

syn:whereis(Name) -> Pid | undefined.

Types:
    Key = any()
    Pid = pid()

To retrieve a Pid from a Name with its metadata:

syn:whereis(Key, with_meta) -> {Pid, Meta} | undefined.

Types:
    Key = any()
    Pid = pid()
    Meta = any()

To unregister a previously registered Name:

syn:unregister(Name) -> ok | {error, Error}.

Types:
    Key = any()
    Error = undefined

You don't need to unregister names of processes that are about to die, since they are monitored by Syn and they will be removed automatically. If you manually unregister a process just before it dies, the callback on process exit (see here below) might not get called.

To register a previously registered Name with a different Pid:

syn:unregister_and_register(Name, Pid) ->
    syn:unregister_and_register(Name, Pid, undefined).

syn:unregister_and_register(Name, Pid, Meta) -> ok.

Types:
    Name = any()
    Pid = pid()
    Meta = any()

Due to Syn being eventually consistent, if you were to sequentially unregister/1 a name and register/2,3 a process you might experience a {error, taken} response to the latter, since the unregistration may not have yet properly propagated when the registration call is made. This call ensures that the registration succeeds and propagates properly.

Note that the previously registered process will not be killed and will be demonitored, so that the on_process_exit/4 callback will not be called (even if implemented) when the process dies.

To retrieve the total count of registered names:

syn:registry_count() -> non_neg_integer().

To retrieve the total count of registered names of processes running on a specific node:

syn:registry_count(Node) -> non_neg_integer().

Types:
    Node = atom()

This is a non-optimized call, use for debugging / monitoring purposes only.

Process Groups

There's no need to manually create / delete Process Groups, Syn will take care of managing those for you.

To add a process to a group:

syn:join(GroupName, Pid) ->
    syn:join(GroupName, Pid, undefined).

syn:join(GroupName, Pid, Meta) -> ok.

Types:
    GroupName = any()
    Pid = pid()
    meta = any()

A process can join multiple groups. When a process joins a group, Syn will automatically monitor it. A process may join the same group multiple times, for example if you need to update its metadata, though it will still be listed only once in it.

To remove a process from a group:

syn:leave(GroupName, Pid) -> ok | {error, Error}.

Types:
    GroupName = any()
    Pid = pid()
    Error = not_in_group

You don't need to remove processes that are about to die, since they are monitored by Syn and they will be removed automatically from their groups.

To get a list of the members of a group:

syn:get_members(GroupName) -> [pid()].

Types:
    GroupName = any()

To get a list of the members of a group with their metadata:

syn:get_members(GroupName, with_meta) ->
    [{pid(), Meta}].

Types:
    GroupName = any()
    Meta = any()

The order of member pids in the returned array is guaranteed to be the same on every node, however it is not guaranteed to match the order of joins.

To know if a process is a member of a group:

syn:member(Pid, GroupName) -> boolean().

Types:
    Pid = pid()
    GroupName = any()

To publish a message to all group members:

syn:publish(GroupName, Message) -> {ok, RecipientCount}.

Types:
    GroupName = any()
    Message = any()
    RecipientCount = non_neg_integer()

RecipientCount is the count of the intended recipients.

To call all group members and get their replies:

syn:multi_call(GroupName, Message) ->
    syn:multi_call(GroupName, Message, 5000).

syn:multi_call(GroupName, Message, Timeout) -> {Replies, BadPids}.

Types:
    GroupName = any()
    Message = any()
    Timeout = non_neg_integer()
    Replies = [{MemberPid, Reply}]
    BadPids = [MemberPid]
      MemberPid = pid()
      Reply = any()

Syn will wait up to the value specified in Timeout to receive all replies from the members. The members that do not reply in time or that crash before sending a reply will be added to the BadPids list.

When this call is issued, all members will receive a tuple in the format:

{syn_multi_call, CallerPid, Message}

Types:
    CallerPid = pid()
    Message = any()

To reply, every member must use the method:

syn:multi_call_reply(CallerPid, Reply) -> ok.

Types:
    CallerPid = pid()
    Reply = any()

To get a list of the local members of a group (= running on the node):

syn:get_local_members(GroupName) -> [pid()].

Types:
    GroupName = any()

To get a list of the local members of a group with their metadata:

syn:get_local_members(GroupName, with_meta) ->
    [{pid(), Meta}].

Types:
    GroupName = any()
    Meta = any()

The order of member pids in the returned array is guaranteed to be the same on every node, however it is not guaranteed to match the order of joins.

To know if a process is a local member of a group:

syn:local_member(Pid, GroupName) -> boolean().

Types:
    Pid = pid()
    GroupName = any()

To publish a message to all local group members:

syn:publish_to_local(GroupName, Message) -> {ok, RecipientCount}.

Types:
    GroupName = any()
    Message = any()
    RecipientCount = non_neg_integer()

RecipientCount is the count of the intended recipients.

To retrieve the total count of groups in the cluster:

syn:groups_count() -> non_neg_integer().

This is a non-optimized call, use for debugging / monitoring purposes only.

To retrieve the count of groups that have at least 1 process running on a specific node:

syn:groups_count(Node) -> non_neg_integer().

Types:
    Node = atom()

This is a non-optimized call, use for debugging / monitoring purposes only.

Callbacks

In Syn you can specify a custom callback module if you want to:

• Receive and handle the event of a registered / joined process' exit.
• Customize the method to resolve registry naming conflict in case of net splits or race conditions.

Setup

The callback module can be set in the environment variable syn, with the event_handler key. You're probably best off using an application configuration file.

Elixir

In config.exs you can specify your callback module:

config :syn,
  event_handler: MyCustomEventHandler

In your module you then need to specify the behavior and the callbacks. All callbacks are optional, so you just need to define the ones you need.

defmodule MyCustomEventHandler do
  @behaviour :syn_event_handler

  @impl true
  @spec on_process_exit(
    name :: any(),
    pid :: pid(),
    meta :: any(),
    reason :: any()
  ) :: any()
  def on_process_exit(name, pid, meta, reason) do
  end

  @impl true
  @spec on_group_process_exit(
    group_name :: any(),
    pid :: pid(),
    meta :: any(),
    reason :: any()
  ) :: any()
  def on_group_process_exit(group_name, pid, meta, reason) do
  end

  @impl true
  @spec resolve_registry_conflict(
    name :: any(),
    {pid1 :: pid(), meta1 :: any()},
    {Pid2 :: pid(), meta2 :: any()}
  ) -> pid_to_keep :: pid()
  def resolve_registry_conflict(name, {pid1, meta1}, {pid2, meta2})
    pid1
  end
end

See details about the callback methods here below.

Erlang

In sys.config you can specify your callback module:

{syn, [
    {event_handler, my_custom_event_handler}
]}

In your module you then need to specify the behavior and the callbacks. All callbacks are optional, so you just need to define the ones you need.

-module(my_custom_event_handler).
-behaviour(syn_event_handler).

-export([on_process_exit/4]).
-export([on_group_process_exit/4]).
-export([resolve_registry_conflict/3]).

-spec on_process_exit(
    Name :: any(),
    Pid :: pid(),
    Meta :: any(),
    Reason :: any()
) -> any().
on_process_exit(Name, Pid, Meta, Reason) ->
    ok.

-spec on_group_process_exit(
    GroupName :: any(),
    Pid :: pid(),
    Meta :: any(),
    Reason :: any()
) -> any().
on_group_process_exit(GroupName, Pid, Meta, Reason) ->
    ok.

-spec resolve_registry_conflict(
    Name :: any(),
    {Pid1 :: pid(), Meta1 :: any()},
    {Pid2 :: pid(), Meta2 :: any()}
) -> PidToKeep :: pid().
resolve_registry_conflict(Name, {Pid1, Meta1}, {Pid2, Meta2}) ->
    Pid1.

See details about the callback methods here below.

Callback methods

on_process_exit/4

Called when a registered process exits. It will be called only on the node where the process was running. If a process was registered under n names, this callback will be called n times (1 per registered name).

on_group_process_exit/4

Called when a process in a group exits. It will be called only on the node where the process was running. If a process was part of n groups, this callback will be called n times (1 per joined group).

resolve_registry_conflict/3

In case of net splits or race conditions, a specific Name might get registered simultaneously on two different nodes. In this case, the cluster experiences a registry naming conflict.

When this happens, Syn will resolve this Process Registry conflict by choosing a single process. By default, Syn keeps track of the time a registration takes place with erlang:system_time/0, compares values between conflicting processes and:

• Keeps the one with the higher value (the process that was registered more recently).
• Kills the other process by sending a kill signal with exit(Pid, {syn_resolve_kill, Name, Meta}).

This is a very simple mechanism that can be imprecise, as system clocks are not perfectly aligned in a cluster. If something more elaborate is desired (such as vector clocks), or if you do not want the discarded process to be killed (i.e. to perform a graceful shutdown), you MAY specify a custom handler that implements this resolve_registry_conflict/3 callback. To this effect, you may also store additional data to resolve conflicts in the Meta field, since it will be passed into the callback for both of the conflicting processes.

If implemented, this method MUST return the pid() of the process that you wish to keep. The other process will be not be killed, so you will have to decide what to do with it.

Important Note: the conflict resolution method SHOULD be defined in the same way across all nodes of the cluster. Having different conflict resolution options on different nodes can have unexpected results.

Anti-Entropy

Anti-entropy is a mechanism to force alignment between nodes. It isn't needed with Syn in most cases. However, despite Syn's best efforts and under rare conditions, depending on your cluster topology and other factors, it might be possible that registry and groups get misaligned.

Anti-entropy in Syn is an experimental feature. As with every anti-entropy feature, it comes with a cost: during the forced syncing, the local tables will be rebuilt with data that gets sent from other nodes. This takes time, due to the sending of data across nodes and subsequent storage. As an example, a node that handles 1,000,000 local registry / groups processes will have to send data for ~80MB (depending on your metadata's size) to other nodes. During the syncing, Syn might time out some calls. Your mileage may vary, so it is recommended that you benchmark your use case.

Setup

To activate anti-entropy you need to set in the environment variable syn the key anti_entropy. You're probably best off using an application configuration file. If you do not specify the anti_entropy key, the anti-entropy mechanism will be disabled by default.

Elixir

In config.exs you can specify your anti-entropy settings:

config :syn,
  anti_entropy:
    registry: [interval: 300, interval_max_deviation: 60]
    groups: [interval: 300, interval_max_deviation: 60]

Erlang

In sys.config you can specify your anti-entropy settings:

{syn, [
    {anti_entropy, [
        {registry, [{interval, 300}, {interval_max_deviation, 60}],
        {groups, [{interval, 300}, {interval_max_deviation, 60}]
    ]}
]}

interval specifies in seconds the interval between every anti-entropy syncing, while interval_max_deviation the max deviation in seconds from the interval. For instance, with an interval of 300 seconds and an interval_max_deviation of 60, anti-entropy will be called with an interval range of 240 to 360 seconds.

Manual sync

You can force an anti-entropy sync on the whole cluster by calling the following:

syn:force_cluster_sync(registry | groups) -> ok.

As per the notes above, in normal conditions Syn doesn't need to be manually synced. This function will force a full mesh sync on all of the cluster and is experimental. Use it as a last resort.

Internals

As of v2.1, Syn uses ETS for memory storage and doesn't have any external dependency. Syn has its own replication and naming conflict resolution mechanisms.

Contributing

So you want to contribute? That's great! Please follow the guidelines below. It will make it easier to get merged in.

Before implementing a new feature, please submit a ticket to discuss what you intend to do. Your feature might already be in the works, or an alternative implementation might have already been discussed.

Do not commit to master in your fork. Provide a clean branch without merge commits. Every pull request should have its own topic branch. In this way, every additional adjustments to the original pull request might be done easily, and squashed with git rebase -i. The updated branch will be visible in the same pull request, so there will be no need to open new pull requests when there are changes to be applied.

Ensure that proper testing is included. To run Syn tests you simply have to be in the project's root directory and run:

$ make test