Ably’s realtime platform enables data to be streamed between Internet-connected devices such as browsers, phones, servers and IoT (Internet of Things) devices in milliseconds. The Ably platform provides developers with enterprise scale messaging by delivering a highly-available service, message delivery guarantees and low latency on a global basis. Round trip latencies are typically less than 60ms.
At its core, Ably is a cloud-based pub/sub Platform-as-a-Service ensuring any device publishing messages to Ably will be received in real time by any number of subscribing devices. But it is more than that. Ably makes it possible for developers to build apps and infrastructure that can communicate in realtime without the worries of managing scale, latency, data durability, integrity and storage, seamless connection recovery, device interoperability, network outages, encryption, security and authentication, throttling, and denial of service attacks, to name a few. Additionally, our platform supports both pub/sub and message-queue patterns.
In order to understand how Ably works, and why pub/sub is only one part of the problem we solve for developers, we explain the key concepts in the following sections.
The Ably service organizes the traffic within any application into named channels. Clients connected to Ably can either be publishers (they push messages with data to Ably), subscribers (they wait for messages to be pushed from Ably to them), or both. Messages are always published over a named channel. Channels are used to filter messages, so whilst billions of messages may be delivered by Ably, subscribers will only receive the messages on the channels they subscribe to.
Channels are uniquely identified by a string specified when publishing to or attaching to a channel. Publishers and subscribers are completely decoupled: a publisher can publish a message without any subscribers on the channel; subscribers can listen on channels that don’t yet have publishers; arbitrarily many subscribers can receive a single message published on a channel. In other words, Ably channels support one-to-many (fan-out), many-to-one (fan-in), and many-to-many.
The following diagram illustrates a very simple use case for channels. The server and vehicle are publishing messages on channels without any knowledge of subscribers. One mobile device is both a subscriber and publisher and is publishing its location, but also subscribing for alerts. All other devices are subscribed to just one channel.
All messages received by Ably are immediately stored in RAM in three separate physical locations for redundancy. They are then persisted as follows:
- In server RAM for 2 minutes in every location that the channel is active
- On disk in three locations for 24 – 72 hours if the channel is configured to persist messages
Whilst Ably is used primarily by clients to receive messages in real time, Ably provides a history API that allows clients to retrieve older messages from memory and/or disk.
Presence provides awareness of other clients that are connected to Ably and present on a channel. Each member present on a channel has a unique client identifier and an optional payload that can be used to describe the member’s status or attributes. Presence allows you to quickly build apps such as chat rooms or multiplayer games as Ably will automatically keep track of who is present in real time across any device. Clients can also subscribe to the presence events and members on a channel without being present themselves.
There are three presence operations,
enter for new members,
update allowing the payload data associated with a member to be updated and announced to everyone, and
leave for members that have requested to leave or who have left implicitly as a result of their connection being disconnected.
The complete set of members present and their optional payload is stored by Ably in server RAM in at least three locations. Like messages, presence events such as
leave are persisted in RAM for 2 minutes, and optionally to disk in three locations for 24 – 72 hours if the channel is configured to persist messages
Whilst pub/sub channels are well suited to fan-out, fan-in or one-to-one communication between servers and/or devices, they are less well suited to servers that want to consume realtime data and distribute that workload between server nodes. Consuming realtime data in a scalable, reliable and stateless way requires a different strategy, and our Ably Integrations deliver that through streams and message queues.
Ably Integrations are powered by a rules engine that enables realtime data to be streamed into our Ably Queue service or any third party queue, streaming provider or endpoint.
- Message queues allow you to distribute the message processing work between your servers to enable horizontal scaling and reliability for applications that need to respond to realtime data as it happens. Find out more
- Firehose allows you to stream realtime data published within the Ably core platform directly to another streaming or queueing service such as Amazon Kinesis or Amazon SQS. Find out more
- Webhooks allow us to notify your servers over HTTP or trigger server-less functions in real time when devices become present, channels become active, or messages are published. Find out more about Webhooks
Ably supports two forms of authentication described below. For an in-depth explanation, view the authentication documentation.
Basic authentication is the simplest form of authentication, allowing clients to communicate with Ably by including the complete private API key within the URL or request headers. To mitigate the risks of sending a private key over the Internet, basic authentication is only permitted over an encrypted TLS connection. Private API keys and their capabilities (permissions) are managed within an app’s dashboard. In most cases, we do not recommend that basic authentication is used as it requires that you share your private API key with the client that is connecting to Ably. Our recommendation is to only use basic authentication on your trusted servers.
Token authentication provides a means for clients to connect to Ably without a private API key. Tokens have a time-to-live (they expire) ensuring that any compromise of a token will have a limited impact. Token authentication is also used to identify clients and provide fine-grained access control on a per-client basis.
Most often, token authentication is implemented as follows: an Ably TokenRequest is generated by your server; it is signed using your private API key and then passed to the client; the realtime client uses this Ably TokenRequest to request an Ably Token from Ably directly. This approach is recommended as at no point does your server need to communicate directly with Ably and ever communicate the secret API key.
An equally viable method is to generate a JSON Web Token (JWT) on your server; it is signed using your private API key and then passed to the client; the realtime client uses this JWT to authenticate with Ably. A JWT which has been constructed to be compatible with Ably is known as an Ably JWT. Any third party software can be used to generate this JWT using the Ably private API key to sign it.
A client using token authentication is considered to either anonymous, or identified if a client ID exists in the token. All messages, presence state and connection state for identified clients contain the trusted client ID and are accessible by other clients. As a private key is needed to generate a token for a client with a client ID, the client ID can be trusted by other clients. Find out more on identified clients.
Tokens are limited to the capabilities of the API key used to create the token. The capabilities of a token may be limited further using fine-grained permissions using a combination of the operation (such as publish, subscribe, presence) and the channel(s). Find out more on token capabilities.
The Ably service, running in over 16 datacenters and 175+ edge acceleration points-of-presence globally, provides a true mesh distributed system ensuring that there is both no single point of congestion and no single point of failure. Ably is designed to always route messages using the least number of network hops minimizing latency and ensuring maximum performance for clients no matter their location.
The diagram below explains how Ably solves the challenge of efficient global routing at all times:
- The publish only server located in New York is routed to the nearest datacenter (US East) using our latency based routing;
Msg Apublished to US East is routed directly to clients in US East, and once to every other datacenter with clients subscribed for messages; Clients in all other regions subscribed for messages receive
Msg Adirectly from the datacenter they are connected to
- Publish & subscribe client in London is routed to the nearest data (EU West) using our latency based routing;
Msg Bpublished to EU West is routed directly to clients in EU West, and once to every other datacenter with clients subscribed for messages; Clients in all other regions subscribed for messages receive
Msg Bdirectly from the datacenter they are connected to
It is common for devices to have constantly changing network conditions as a result of moving from a mobile data network to a WiFi network, being in a tunnel for a short period, or perhaps due to intermittent network issues. We believe that developers shouldn’t have to worry about service continuity between brief connection loss. As such, Ably ensures that the connection state for a client is retained on our servers whenever a client is disconnected abruptly. By retaining a client’s connection state on our servers, this allows us to replay all channel activity back to a client as soon as it reconnects.
Ably provides the following assurances in regards to connection state recovery:
- Any client that is able to reestablish a connection within 2 minutes will remain attached on all channels and will not lose any messages published by other clients whilst disconnected;
- All messages published on channels whilst the client was disconnected will be replayed to the recovered client in the order they were published;
- There is no upper limit in regards to the number of messages that are queued for disconnected clients, however if a threshold to ensure system stability is exceeded then an error is emitted;
- Connection state recovery provides certainty and is thus binary in its behavior – it either succeeds and all operations continue as if the connection was never disconnected, or it fails by detaching the channel or emitting an error on the channel so that the developer is made aware of the state loss and is able to respond accordingly
The Ably global platform is designed to provide rigorous uptime guarantees. This is possible because redundancy has been addressed in not just every area of our systems, but also within the client libraries used by our customers. Our redundancy is best depicted in the following diagram:
- Connection state recovery
- ensures that clients disconnected abruptly from Ably can resume their connection
- Proactive health checked DNS
- our DNS TTL is kept very low allowing us to route traffic away from unhealthy datacenters in 60s once our monitoring systems detect an issue
- Secondary domain endpoints
- we operate a completely isolated backup domain that is used by our client libraries if the primary
ably.ioendpoint is failing
- Host fallback
- clients that are unable to connect to their closest datacenter using the primary
ably.iodomain will automatically fallback to an alternative datacenter using the secondary domain
- Auto scaling
- as load on the system increases, more servers are automatically provisioned in every part of the system that requires additional capacity
- Load balancers
- our load balancers are elastic and scale to meet demand, but are also responsible for intelligently routing traffic to existing and new frontends that are coming online
- our frontends do not store any state thus ensuring that frontends can come online quickly and service new requests, but also go offline easily without data loss
- Self-healing cluster
- as problems are detected in the system, they are isolated or remedied by our automated health servers
- Data replicas
- all data is stored in at least three datacenters across at least two regions ensuring data availability through any imaginable failure
- Multiple availability zones
- in every region our our data is replicated between servers in at least two independent datacenters ensuring outages in one datacenter cannot cause data loss
- Multiple regions
- ensuring that data is always stored in at least two regions protects against complete region outage or network partitioning
- Edge acceleration points-of-presence
- servers running in 175+ locations globally that accept connections close to where users are (reducing latency) and accelerate traffic to an Ably datacenter using dedicated network connections (not over the Internet)
By default all of our client libraries use TLS when communicating with Ably over REST or via our Realtime transports such as WebSockets. We do not charge customers for using TLS and actively encourage everyone to always use TLS as it provides a secure transport for communication with Ably with very little practical overhead.
Whilst TLS encryption ensures that messages in transit to and from Ably cannot be intercepted, inspected, or tampered with, it does not ensure that the Ably service itself is unable to (in theory) inspect your messages and their content. If you want to ensure that all messages are encrypted and complete opaque and inaccessible to the Ably service, we provide built-in private key symmetric encryption within our client libraries.
Most of our client libraries use a WebSocket to establish a realtime connection to Ably, and use a simple HTTP request for all other REST operations including authentication.
- WebSockets (supported by 94% of browsers globally as of Dec 2017)
- XHR streaming
- XHR polling
- JSONP polling