JarCastingawsflake
A Snowflake-like service for generating GUIDs within multi-regional AWS environments.
Background
AWSFlake generates a 77-bit unique identifier composed of the following information:
- 41 bits: time since epoch, in millis (this gives us a max of ~69.7 years from the epoch)
- 5 bits: region identifier (1-32, see "regions")
- 16 bits: last two octets of private IP (unique per VPC /16 netmask)
- 15 bits: sequence number; allows multiple IDs to be generated within the same milisecond
The ID is base 62 encoded, which is similar to base 64 encoding minus the non-alphanumeric characters (the charset [0-9A-Za-z] is used).
Regions
The region identifier is an integer, 0-31, that corresponds to an AWS region. This mapping is hardcoded currently and this mapping can be viewed in AWSRegion.kt. Currently there are only 12 AWS regions, but this code can support up to a total of 32.
IP Address (IP)
The AWS EC2 instance private IP is used in the unique identifier generation algorithm. Since largest CIDR block AWS hands out for individual VPCs is /24, that directly implies that the last two octets of an instance's private IP can uniquely identify said instance within a single VPC.
Combined with the region identifier, and you can uniquely identify an EC2 instance in a multi-region AWS cluster. THIS DOES NOT WORK FOR MULTI-VPC CLUSTERS WITHIN THE SAME VPC, as the last octets of the IP address are only guaranteed to be unique within a single VPC.
Usage
As a Library
Perhaps the simplest usage of awsflake is as a library. You can use awsflake-core as a typical Kotlin or Java library (Groovy, Scala, and other JVM languages should also be supported theoretically, but this has not been tested yet).
Gradle
implementation 'me.ccampo:awsflake-core:2.0.0'
Maven
<dependency>
<groupId>me.ccampo</groupId>
<artifactId>awsflake-core</artifactId>
<version>2.0.0</version>
</dependency>
The API is simple - instantiate an IDGenerator with the provided IDGenerator.Builder class, and then use that to start generating IDs.
val epoch = Instant.parse("2016-01-01T00:00:00Z")
val region = AWSRegion.US_EAST_1
val ip = "10.0.128.255"
val generator = IDGenerator.Builder()
.epoch(epoch)
.region(reg)
.ip(ip)
.build()
val id = generator.nextId()
See the source and KDocs for more detailed usage information.
As a Microservice
awsflake can also packaged as a microservice, powered by Micronaut.
In the server directory, there is a README with additonal information, but in general, the most basic use case is to build a docker image of the service, and then call it via http. There is a build-docker.sh script to build the image. Additionally, you can simply build the project, and a fat-jar will be produced which can be executed directly. For example:
# In the service/build/libs directory...
$ java -jar awsflake-server-2.0.0-all.jar
15:43:58.950 [main] INFO io.micronaut.runtime.Micronaut - Startup completed in 4775ms. Server Running: http://localhost:8080
# And then to generate IDs..
$ curl http://localhost:8080/id
57NO5HEVlVCa
Deployment
Once deployed, AWSFlake requires no coordination between nodes. You can scale horizontally with thousands of nodes to generate thousands of IDs per second. It is recommended that at least a 2 node cluster is used, behind a round-robin load balancer.
AWSFlake can be built as a docker container and then deployed onto any of AWS's services that support docker. This includes ECS, Elastic Beanstalk, and plain-old EC2 instances. If using Beanstalk or ECS, PLEASE ENSURE ONLY A SINGLE CONTAINER IS RUN ON EACH INSTANCE. This is necessary to ensure ID uniqueness.
Once you have the docker image built, you can run a container. The following environment variables are accepted as input:
- MICRONAUT_SERVER_PORT: the port on which AWSFlake listens on
Additionally the following variables may be provided for debugging/testing only. If omitted, they will be supplied by the Amazon EC2 metadata service automatically.
- AWSFLAKE_REGION: the region where this is deployed. This is usually not necessary and the service will default to the region where the code is running, using AWS's EC2 metadata service.
- AWSFLAKE_IP: the IPv4 address of the node. Again, this is also optional and will default to the IP returned by the EC2 metadata service.
- AWSFLAKE_EPOCH: the epoch to calculate timestamps against. The more recent, the better. Use the same value across all nodes in your AWSFlake cluster. Defaults to Jan 1 2020.
API
To generate a new ID, simply run the service and send an HTTP request to the /id endpoint. The minLength query parameter can be provided to ensure IDs are returned with a minimum number of characters. Example using curl:
$ curl -X GET "http://localhost:8080/id?minLength=13"
0aDuEi4sFesTo
The generated, base 62 encoded 13 character ID in the above example is 0aDuEi4sFesTo.
Development
Requires Java 8
To Build
macOS or *nix:
./gradlew build
Windows:
gradlew.bat build
Additionally, each sub-module can be built in the same fashion.
Publishing to Bintray
./gradlew -PbintrayUser="..." -PbintrayKey="..." clean build dokkaJar bintrayUpload