Guble is a simple user-facing messaging and data replication server written in Go.
Guble is in an early state (release 0.4). It is already working well and is very useful, but the protocol, API and storage formats may still change (until reaching 0.7). If you intend to use guble, please get in contact with us.
The goal of guble is to be a simple and fast message bus for user interaction and replication of data between multiple devices:
- Very easy consumption of messages with web and mobile clients
- Fast realtime messaging, as well as playback of messages from a persistent commit log
- Reliable and scalable over multiple nodes
- User-aware semantics to easily support messaging scenarios between people using multiple devices
- Batteries included: usable as front-facing server, without the need of a proxy layer
- Self-contained: no mandatory dependencies to other services
- Publishing and subscription of messages to topics and subtopics
- Persistent message store with transparent live and offline fetching
- WebSocket and REST APIs for message publishing
- Commandline client and Go client library
- Firebase Cloud Messaging (FCM) adapter: delivery of messages as FCM push notifications
- Docker images for server and client
- Simple Authentication and Access-Management
- Clean shutdown
- Improved logging using logrus and logstash formatter
- Health-Check with Endpoint
- Collection of Basic Metrics, with Endpoint
- Added Postgresql as KV Backend
- Load testing with 5000 messages per instance
- Support for Apple Push Notification services (a new connector alongside Firebase)
- Upgrade, cleanup, abstraction, documentation, and test coverage of the Firebase connector
- GET list of subscribers / list of topics per subscriber (userID , deviceID)
- Support for SMS-sending using Nexmo (a new connector alongside Firebase)
Measured on an old notebook with i5-2520M, dual core and SSD. Message payload was 'Hello Word'. Load driver and server were set up on the same machine, so 50% of the cpu was allocated to the load driver.
- End-2-End: Delivery of ~35.000 persistent messages per second
- Fetching: Receive of ~70.000 persistent messages per second
During the tests, the memory consumption of the server was around ~25 MB.
This is the current (and fast changing) roadmap and todo list:
- Replication across multiple servers (in a Guble cluster)
- Acknowledgement of message delivery for connectors
- Storing the sequence-Id of topics in KV store, if we turn off persistence
- Filtering of messages in guble server (e.g. sent by the REST client) according to URL parameters: UserID, DeviceID, Connector name
- Updating README to show subscribe/unsubscribe/get/posting, health/metrics
- Make notification messages optional by client configuration
- Correct behaviour of receive command with
maxCount
on subtopics - Cancel of fetch in the message store and multiple concurrent fetch commands for the same topic
- Configuration of different persistence strategies for topics
- Delivery semantics: user must read on one device / deliver only to one device / notify if not connected, etc.
- User-specific persistent subscriptions across all clients of the user
- Client: (re-)setup of subscriptions after client reconnect
- Message size limit configurable by the client with fetching by URL
- HTTPS support in the service
- Minimal example: chat application
- Stable JavaScript client: https://github.com/smancke/guble-js
- (TBD) Improved authentication and access-management
- (TBD) Add Consul as KV Backend
- (TBD) Index-based search of messages using GoLucene
We are providing Docker images of the server and client for your convenience.
There is an automated Docker build for the master at the Docker Hub. To start the server with Docker simply type:
docker run -p 8080:8080 smancke/guble
To see available configuration options:
docker run smancke/guble --help
All options can be supplied on the commandline or by a corresponding environment variable with the prefix GUBLE_
.
So to let guble be more verbose, you can either use:
docker run smancke/guble --log=info
or
docker run -e GUBLE_LOG=info smancke/guble
The Docker image has a volume mount point at /var/lib/guble
, so if you want to bind-mount the persistent storage from your host you should use:
docker run -p 8080:8080 -v /host/storage/path:/var/lib/guble smancke/guble
The Docker image includes the guble commandline client guble-cli
.
You can execute it within a running guble container and connect to the server:
docker run -d --name guble smancke/guble
docker exec -it guble /usr/local/bin/guble-cli
Visit the guble-cli
documentation for more details.
Since Go makes it very easy to build from source, you can compile guble using a single command. A prerequisite is having an installed Go environment and an empty directory:
sudo apt-get install golang
mkdir guble && cd guble
export GOPATH=`pwd`
Build and start guble with the following commands (assuming that directory /var/lib/guble
is already created with read-write rights for the current user):
go get github.com/smancke/guble
bin/guble --log=info
go get -t github.com/smancke/guble/...
go test github.com/smancke/guble/...
The following clients are available:
- Commandline Client: https://github.com/smancke/guble/tree/master/guble-cli
- Go client library: https://github.com/smancke/guble/tree/master/client
- JavaScript library: (in early stage) https://github.com/smancke/guble-js
Currently there is a minimalistic REST API, just for publishing messages.
POST /api/message/<topic>
URL parameters:
- userId: The PublisherUserId
- messageId: The PublisherMessageId
You can set fields in the header JSON of the message by providing the corresponding HTTP headers with the prefix X-Guble-
.
Curl example with the resulting message:
curl -X POST -H "x-Guble-Key: Value" --data Hello 'https://127.0.0.1:8080/api/message/foo?userId=marvin&messageId=42'
Results in:
16,/foo,marvin,VoAdxGO3DBEn8vv8,42,1451236804
{"Key":"Value"}
Hello
The communication with the guble server is done by ordinary WebSockets, using a binary encoding.
All payload messages sent from the server to the client are using the following format:
<path:string>,<sequenceId:int64>,<publisherUserId:string>,<publisherApplicationId:string>,<publisherMessageId:string>,<messagePublishingTime:unix-timestamp>\n
[<application headers json>]\n
<body>
example 1:
/foo/bar,42,user01,phone1,id123,1420110000
{"Content-Type": "text/plain", "Correlation-Id": "7sdks723ksgqn"}
Hello World
example 2:
/foo/bar,42,user01,54sdcj8sd7,id123,1420110000
anyByteData
- All text formats are assumed to be UTF-8 encoded.
- Message
sequenceId
s areint64
, and distinct within a topic. The messagesequenceId
s are strictly monotonically increasing depending on the message age, but there is no guarantee for the right order while transmitting.
The client can send the following commands.
Publish a message to a topic:
> <path> [<publisherMessageId>]\n
[<header>\n]..
\n
<body>
example:
> /foo
Hello World
Receive messages from a path (e.g. a topic or subtopic). This command can be used to subscribe for incoming messages on a topic, as well as for replaying the message history.
+ <path> [<startId>[,<maxCount>]]
path
: the topic to receive the messages fromstartId
: the message id to start the replay ** If nostartId
is given, only future messages will be received (simple subscribe). ** If thestartId
is negative, it is interpreted as relative count of last messages in the history.maxCount
: the maximum number of messages to replay
Note: Currently, the fetching of stored messages does not recognize subtopics.
Examples:
+ /foo # Subscribe to all future messages matching /foo
+ /foo/bar # Subscribe to all future messages matching /foo/bar
+ /foo 0 # Receive all message from the topic and subscribe for further incoming messages.
+ /foo 42 # Receive all message with message ids >= 42
# from the topic and subscribe for further incoming messages.
+ /foo 0 20 # Receive the first (oldest) 20 messages within the topic and stop.
# (If the topic has less messages, it will stop after receiving all existing ones.)
+ /foo -20 # Receive the last (newest) 20 messages from the topic and then
# subscribe for further incoming messages.
+ /foo -20 20 # Receive the last (newest) 20 messages within the topic and stop.
# (If the topic has less messages, it will stop after receiving all existing ones.)
Cancel further receiving of messages from a path (e.g. a topic or subtopic).
- <path>
example:
- /foo
- /foo/bar
The server sends status messages to the client. All positive status messages start with >
.
Status messages reporting an error start with !
. Status messages are in the following format.
'#'<msgType> <Explanation text>\n
<json data>
#ok-connected You are connected to the server.\n
{"ApplicationId": "the app id", "UserId": "the user id", "Time": "the server time as unix timestamp "}
Example:
#connected You are connected to the server.
{"ApplicationId": "phone1", "UserId": "user01", "Time": "1420110000"}
This notification confirms, that the messaging system has successfully received the message and now starts transmitting it to the subscribers:
#send <publisherMessageId>
{"sequenceId": "sequence id", "path": "/foo", "publisherMessageId": "publishers message id", "messagePublishingTime": "unix-timestamp"}
Depending on the type of +
(receive) command, up to three different notification messages will be sent back.
Be aware, that a server may send more receive notifications that you would have expected in first place, e.g. when:
- Additional messages are stored, while the first fetching is in progress
- The server decides to meanwhile stop the online subscription and change to fetching, because your client is too slow to read all incoming messages.
-
When the fetch operation starts:
#fetch-start <path> <count>
path
: the topic pathcount
: the number of messages that will be returned
-
When the fetch operation is done:
#fetch-done <path>
path
: the topic path
-
When the subscription to new messages was taken:
#subscribed-to <path>
path
: the topic path
An unsubscribe/cancel operation is confirmed by the following notification:
#canceled <path>
This message indicates, that the message could not be delivered.
!error-send <publisherMessageId> <error text>
{"sequenceId": "sequence id", "path": "/foo", "publisherMessageId": "publishers message id", "messagePublishingTime": "unix-timestamp"}
This notification has the same meaning as the http 400 Bad Request.
!error-bad-request unknown command 'sdcsd'
This notification has the same meaning as the http 500 Internal Server Error.
!error-server-internal this computing node has problems
Messages can be hierarchically routed by topics, so they are represented by a path, separated by /
.
The server takes care, that a message only gets delivered once, even if it is matched by multiple
subscription paths.
The path delimiter gives the semantic of subtopics.
With this, a subscription to a parent topic (e.g. /foo
)
also results in receiving all messages of the subtopics (e.g. /foo/bar
).