Datomic Documentation
«

Ions Reference

This page covers the basic steps of ion development:

It also covers more advanced topics:

Prerequisites

Before you begin developing an ion application, you will need:

Developing Ions

This section covers the things you need to know when writing ion code:

Project Structure

You develop ions in a tools.deps project organized as follows:

  • The project must be a git repository – Ions use git SHAs to uniquely name a reproducible revision.
  • There should be a single deps.edn file located at the project root. Multiple deps.edn files are incompatible with the use of SHAs to uniquely name a revision.
  • The :deps sections of deps.edn must include the client-cloud and ion libraries.
  • The deps.edn classpath must include a resource named datomic/ion-config.edn. This resource configures the entry points for an application.

For an example, check the sample project deps.edn and ion-config.edn.

Ion Clients

In an ion application, your client code and Datomic run in the same process on a compute node. To support this while also supporting development and testing, you can use the :ion server type as an argument to client. When you create a client with :ion server-type, Datomic will create a client based on where your code is running:

  • If the code is running on a Cloud node, Datomic will ignore the rest of the argument map to client and create an in-memory ion client
  • Otherwise, Datomic will use the rest of the argument map to create a client as for the :cloud server-type

Clients created with the :ion server type support the synchronous client API only.

The following example shows how to create a client that connects to the inventory-dev system in us-east-1 during development. When deployed as an ion, the same code will create an in-memory client on the system it is deployed to. 

(require '[datomic.client.api :as d])

(def cfg {:server-type :ion
          :region "us-east-1"
          :system "inventory-dev"
          :endpoint "https://ljfrt3pr18.execute-api.us-east-1.amazonaws.com"})

(def client (d/client cfg))

Ion Entry Points

Ion applications are arbitrary Clojure code, exposed to consumers via one or more entry points. An entry point is a function with a well-known signature. There are five types of entry points for different callers, each with a different function signature.

Lambda and HTTP direct are external entry points. They expose AWS lambdas and web services, respectively.

Internal entry points are callbacks that extend the Datomic Client API with your code. They include transaction functions, query functions, and pull xforms.

Entry point Called from Input Output
Lambda AWS lambda :input JSON, :context map String, InputStream, ByteBuffer, or File
HTTP Direct Web client Web request Web response
Transaction fn Inside a transaction Transaction data Data
QQuery fn Inside query Data Data
xform Pulled attribute value Data Data

Lambda Entry Point

A lambda entry point is a function that takes a map with two keys:

  • :input is a String containing the input JSON payload.
  • :context contains all the data fields of the AWS lambda context as a Clojure map, with all keys (including nested and dynamic map keys) converted to keywords.

A lambda entry point can return any of String, ByteBuffer, InputStream, or File; or it can throw an exception to signal an error.

For example, the following echo function simply echos back its invocation: 

(defn echo
  [{:keys [context input]}]
  input)

Ion deploy automatically creates AWS lambdas that forward requests to your entry points. Your entry points have access to all the resources of your Datomic system and are not constrained by the execution context of AWS lambdas. However, entry point responses must still flow back through AWS lambda to callers, so lambda entry points are bound by the AWS lambda limits on response size (6MB) and response timeout (900 seconds).

HTTP Direct Entry Point

A web entry point is a function that takes the following input map:

Key Req'd Value Example
:body No InputStream "hello"
:headers Yes Map string->string {"x-foo" "bar"}
:protocol Yes HTTP protocol "HTTP/1.1"
:remote-addr Yes Caller host "example.com"
:request-method Yes HTTP verb as keyword :get
:scheme No :http or :https :https
:server-name Yes Server host name "example.com"
:server-port No TCP port 443
:uri Yes Request URI /

And returns:

Key Req'd Value Example
:body No InputStream or String "goodbye"
:headers Yes Map string->string {"x-foo" "bar"}
:status Yes HTTP status code 200

The input and output maps are a subset of the Clojure ring spec, and many web applications should be able to run unmodified as HTTP direct ions.

The ion-starter project includes an example HTTP direct ion.

ion-config.edn

The datomic/ion-config.edn resource configures the application's entry points. Ion-config is an EDN map with the following keys:

  • :app-name is string name of a Datomic application.
  • :allow is a vector of fully qualified symbols naming query or transaction functions. When you deploy an application, Datomic will automatically require all the namespaces mentioned under :allow.
  • :xforms is a vector of fully qualified symbols naming functions for use in xform.
  • :lambdas is a map configuring AWS lambda entry points
  • :http-direct is a map configuring an HTTP direct entry point.

:lambdas

:lambdas is a map from lambda names (keywords) to lambda configurations.

For each entry in the :lambdas map, ion deploy will create an AWS lambda named $(group)-$(name), where group is the :group key you used to invoke deploy, and name is the name is the lambda name. AWS lambda names are limited to 64 characters, so make sure that you choose a lambda name that will not exceed this when appended to your group name.

A lambda configuration supports the following keys:

Key Required Value Example Default
:fn Yes Symbol datomic.ion.starter/echo  
:description No String "echo" ""
:timeout-secs No Positive int 60 60
:concurrency-limit No Positive int or :none 20 20
:integration No :api-gateway/proxy :api-gateway/proxy  

:concurrency-limit sets AWS reserved concurrency. Set it :none if you want the AWS lambda to use unreserved concurrency.

The :integration key customizes integration between the lambda and AWS. The only value currently supported is :api-gateway/proxy. Set this only for web Lambda proxies.

HTTP Direct Configuration

The HTTP Direct configuration map supports the following keys:

Key Required Description Default
:handler-fn Yes Fully qualified handler function name  
:pending-ops-queue-length No Number of operations to queue 100
:processing-concurrency No Number of concurrent operations 16
:pending-ops-exceeded-message No Return message when queue is full "Throttled"
:pending-ops-exceeded-code No HTTP status code returned when queue is full 429

Nodes will enqueue up to :pending-ops-queue-length HTTP direct requests, servicing them with :processing-concurrency threads. When the ops queue is full, nodes will reject requests with :pending-ops-exceeded-messsage and :pending-ops-exceeded-code.

You can use the HttpDirectOpsPending and HttpDirectThrottled metrics to monitor your application and autoscale query groups.

Ion-Config Example

The example below is taken from the ion-starter project: 

{:allow [;; transaction function
         datomic.ion.starter/create-item

         ;; query function
         datomic.ion.starter/feature-item?]            

 ;; AWS Lambda entry points
 :lambdas {:echo
           {:fn datomic.ion.starter/echo
            :description "Echos input"}
           :get-tutorial-schema
           {:fn datomic.ion.starter/get-tutorial-schema
            :description "returns the schema for the Datomic docs tutorial"}
           :get-items-by-type
           {:fn datomic.ion.starter/get-items-by-type
            :description "Lambda handler that returns items by type"}}

 ;; HTTP Direct entry point
 :http-direct {:handler-fn datomic.ion.starter/items-by-type}
 :app-name "<YOUR-APP-HERE>"}

JVM Settings

When developing and testing locally, it can be helpful to match the JVM settings that your code will run under in Datomic Cloud.

Datomic Cloud runs the Amazon Coretto 11 version of Java.

Instance type Heap (-Xmx) Stack (-Xss)
t3.small 1189m 512k
t3.medium 2540m 1m
t3.large 5333m 1m
t3.xlarge 10918m 1m
i3.large 10520m 1m
i3.xlarge 21280m 1m

Additionally, all instances use the following flags: 

-XX:+UseG1GC -XX:MaxGCPauseMillis=50 -Dclojure.spec.skip-macros=true

Push

Ion /push/package your application into runnable artifacts. Push creates an AWS CodeDeploy application revision in S3 that can later be deployed to one or more compute groups. Ion push understands your dependencies at a granular level, so e.g. all revisions can share the same copy of a common library. This is more efficient than "uberjar" style deployment.

To push an application revision, call datomic.ion.dev with an :op of :push: 

clojure -A:ion-dev '{:op :push (options)}'
Keyword Required Value Example
:op Yes :push :push
:uname No Unreproducible name "janes-wip"
:creds-profile No AWS profile name "janes-profile"
:region No AWS region "us-east-1"

Push will ensure that your application's dependencies (Git and Maven libs) exist in your ion code bucket: 

s3://$(ion-code-bucket)/datomic/libs

And will create an application revision located at: 

s3://$(ion-code-bucket)/datomic/apps/$(app-name)/$(git-sha|uname).zip

On success, push returns a map with the following keys:

Keyword Required Value
:rev No Git SHA for the commit that was pushed
:uname No Unreproducible name for the push
:deploy-groups Yes List of available groups for deploy
:deploy-command Yes Sample command for deploy
:doc No Documentation
:dependency-conflicts No Map describing conflicts

Dependency Conflicts

Because ions run on the same classpath as Datomic Cloud, it is possible for ion dependencies to conflict with Datomic's own dependencies. If this happens:

  • Datomic's dependencies will be used
  • The return from push will warn you with a :dependency-conflicts map
  • You can add the :deps from the conflicts map to your local project so that you can test against the libraries used by Datomic Cloud

The Datomic team works to keep Datomic's dependencies up-to-date. If you are unable to resolve a dependency conflict, please contact support.

Unreproducible Push

By default, an ion push is reproducible, i.e. built entirely from artifacts in git or maven repositories. For a push to be reproducible, your git working tree must be clean, and your deps.edn project cannot include any local/root dependencies. A reproducible push is uniquely named by the SHA of its git commit.

In some situations, you may want to push code that Datomic does not know to be reproducible. For example:

  • You are testing work-in-progress that does not have a git commit
  • You are implementing your own approach to reproducibility

For these situations, ions permit an unreproducible push. Since an unreproducible push has no git SHA, you must specify an unrepro name (:uname).

You are responsible for making the uname unique within your application. If unames are not unique, ions will be unable to automatically roll back failed deploys.

Deploy

Ion deploy deploys a pushed revision to a compute group.

When you deploy, Datomic will use AWS Step Functions to

  • CodeDeploy the code for the application onto each instance in the compute group.
  • Automatically roll back to the previous deployment if the application does not deploy correctly (e.g. if loading a namespace throws an exception).
  • Ensure that active databases are present in memory before routing requests to newly updated nodes.
  • Ensure that all the lambdas requested via the ion-config.edn map exists and are configured correctly.

Node deployment happens one node at a time, so a Datomic system will remain available with N-1 members active throughout a deployment.

To deploy an application, call datomic.ion.dev with an :op of :deploy: 

clojure -A:ion-dev '{:op :deploy (options)}'
Keyword Required Value Example
:op Yes :deploy :deploy
:group Yes C ompute group name "my-datomic-compute"
:rev (Or :uname) Output from push "6468765f843d70e01a7a2e483405c5fcc9aa0883"
:uname (Or :rev) Input to push "janes-wip"
:creds-profile No AWS profile name "janes-profile"
:region No AWS region "us-east-1"

The time to complete an entire deployment is the sum of:

  • the sum of times to deploy to each node
  • the time to (re)configure lambdas

The time to deploy to a single node is the sum of

  • The time to copy new code and deps from S3 to the node
  • The time to stop and restart the Datomic Cloud process
  • The time to load active databases into memory

Deployment to a single node can take from 20 seconds up to several minutes, depending on the number and size of active databases.

You can monitor a deployment with the deploy-status command or from the AWS Console.

Deploy-Status

To check the status of a deploy, call datomic.ion.dev with an :op of :deploy-status: 

clojure -A:ion-dev '{:op :deploy-status ...}'
Keyword Required value Example
:op Yes :deploy-status :deploy-status
:execution-arn Yes Output from deploy "arn:aws:states:us-east-1:123456789012:execution:datomic-compute:datomic-compute-1526506240469"
:creds-profile No AWS profile name "janes-profile"
:region No AWS region "us-east-1"

Deploy status returns the the Step Functions reference of the overall ion deploy and the code deploy as a map: 

{:deploy-status "SUCCEEDED"
 :code-deploy-status "SUCCEEDED"}

On success, both the overall ion deploy and code deploy status will eventually, return with "SUCCEEDED"

Console Status

You can monitor a deploy in near real-time from the Step Functions Console. Look for a state machine named datomic-$(group).

The CodeDeploy step can be monitored in the CodeDeploy Deployments console.

Invoking Ions

Ion entry points that extend the power of the Client API are described in the relevant documentation sections:

Web service and lambda invocations are described below.

Invoking Web Services

Ion web services are exposed at the URL https://$(IonApiGatewayEndpoint), where IonApiGatewayEndpoint is in your compute group's template outputs. You can invoke a web service with any HTTPS client, e.g. curl: 

curl https://$(IonApiGatewayEndpoint)

If your compute template does not have an IonApiGatewayEndpoint, see older versions of Datomic.

Invoking Lambdas

When you deploy an application with lambda ions, Datomic will create AWS lambdas named: 

$(group)-$(name)

Where group is the :group key you used to invoke deploy, and name is the name under the :lambdas key in ion-config.edn.

The tutorial includes an example of invoking an ion Lambda via the AWS CLI.

Configuring Compute Groups

Ions typically have some amount of configuration, e.g. the name of a Datomic database that they use. If you are deploying your ion application to a single compute group, then you might choose to keep such configuration in Clojure vars in your application source code. This is easy, if not simple:

  • Everything is in a single place. For small services, it may even fit in a single screen of Clojure code.
  • Code and configuration are versioned together on every ion push.

If you have more complex ions, potentially deployed to multiple compute groups (or even multiple systems), there are several challenges to consider:

  • Configuration values may have a lifecycle that is independent of application source code. They should not be hard-coded in the application and should be managed separately from the source code.
  • Applications need a way to obtain their configuration values at runtime.
  • Configuration values may be sensitive, and should not be stored or conveyed as plaintext.
  • Configuration values may need to be secured at a granular level.

Datomic Cloud provides tools to help with these challenges.

Configuration Scope Source IAM
App-info map Compute group Datomic No
Environment map Compute group User No
Parameters Arbitrary User Yes

App-Info Map

If you deploy the same ion to different compute groups, your code may want to be conditional based on which group it is running in. You can discover your compute group at runtime with get-app-info, which takes no arguments, and returns a map that will include at least these keys:

For example, the following excerpt from deploy-monitor gets the app-name: 

(get (ion/get-app-info) :app-name)

When running outside Datomic Cloud, get-app-info returns the value of the DATOMIC_APP_INFO_MAP environment variable, read as EDN.

If you want to add your own per-compute-group configuration, you can create an environment map.

Environment Map

You create an environment map when you first create a compute group. For example, the environment map below specifies that a compute group is for a :staging deployment environment:

environment-map.png

You can then retrieve the environment map at runtime with get-env, which takes no arguments, and returns the environment map.

For example, the following excerpt from deploy-monitor retrieves the deployment environment specified above: 

(get (ion/get-env) :env)

You can set the environment map for local development via the DATOMIC_ENV_MAP environment variable. When running outside Datomic Cloud, get-env returns the value of DATOMIC_ENV_MAP, read as EDN.

If you need to change the environment map for a running compute group, you can update the group's CloudFormation stack.

If you need a more flexible lifecycle or granular security, you can use parameters as described below.

Parameters

  • A parameter is a named slot known to application code that can be filled in with a specific parameter value at runtime
  • The AWS Systems Manager Parameter Store provides an implementation of parameters that support an independent lifecycle, encryption for sensitive data, and IAM security over a hierarchical naming scheme
  • Inside a given AWS account and region, all Datomic cluster nodes have read permission on parameter store keys that begin with datomic-shared
  • The ion library provides get-params as a convenience for reading parameter store parameters

Get-Params

The get-params function is a convenience abstraction over GetParametersByPath. get-params take a map with a :path key, and it returns all the parameters under the path as a map from the parameter name string to parameter value string, decrypting if necessary.

For example, the call to get-params below returns all the parameters under the path /datomic-shared/prod/deploy-monitor. 

(ion/get-params {:path "/datomic-shared/prod/deploy-monitor"})

Per-System Permissions

The datomic-shared parameter prefix is readable by all Datomic systems in a particular AWS account and region. If you want more granular permissions, you can choose your own naming convention (under a different prefix), and explicitly add permissions to the IAM policy for your Datomic nodes.

Configuration Example

The deploy-monitor sample application demonstrates using app-info, the environment map, and parameters together. First, the environment and app-info map are used to create a key prefix following the the naming convention: 

/datomic-shared/(env)/(app-name)/
  • datomic-shared is a prefix readable by all Datomic nodes
  • env is taken from the environment map, and differentiates different environments for the same application, e.g. "ci" vs. "prod"
  • app-name is the ion app-name

Then, this prefix is used to name configuration keys in the parameter store. Deploy-monitor needs four parameters:

  • A Datomic database name
  • A Slack channel
  • Two encrypted tokens for Slack

Given this convention, the following AWS CLI commands will create the parameters for the "prod" environment: 

# actual parameter values not shown
aws ssm put-parameter --name /datomic-shared/prod/deploy-monitor/db-name --value $DB_NAME --type String
aws ssm put-parameter --name /datomic-shared/prod/deploy-monitor/channel --value $CHANNEL --type String
aws ssm put-parameter --name /datomic-shared/prod/deploy-monitor/bot-token --value $BOT_TOKEN --type SecureString
aws ssm put-parameter --name /datomic-shared/prod/deploy-monitor/verification-token --value $VERIFICATION_TOKEN --type SecureString

You could use similar commands to create parameters for additional environments.

At runtime, deploy-monitor uses get-app-info and get-env to load the information needed to create a Parameter Store path, and then reads all of its parameter values with get-params: 

(def get-params
  "Returns the params under /datomic-shared/(env)/(app-name)/, where

env       value of get-env :env
app-name  value of get-app-info :app-name"
  (memoize
   #(let [app (or (get (ion/get-app-info) :app-name) (fail :app-name))
          env (or (get (ion/get-env) :env) (fail :env))]
      (ion/get-params {:path (str "/datomic-shared/" (name env) "/" app "/")}))))

Naming Applications

When you create a compute group, you specify an AWS CodeDeploy application name to be used for deploying ions. By default, this name is the same as the name of your compute group.

Override the default and choose a common application name if you plan to have multiple compute groups run the same code, e.g. different development stages of the same application. For example, you might have two systems "inventory-staging" and "inventory-production", with the intent that code deployed to production has always been deployed and tested in staging first. You can give the compute groups in both systems a shared application name "inventory" when you create their compute stacks.

The application name cannot be changed after a compute group is created. If you need a different application name, simply delete the compute group and create a new one.

Best Practices

  • Ions are designed to let you do most of your development and testing at a Clojure REPL, with a much faster feedback look than push/deploy. Take advantage of this wherever possible.
  • You do not need ions, or a Datomic Cloud system, to get develop and test Datomic applications. You can start with dev-local, and when you are ready for ions, make your code portable with divert-system.
  • Do not store AWS credentials in the Parameter Store, as Datomic Cloud fully supports IAM roles.
  • The tools.deps classpath is the source of truth for an ion revision. If you need a file for local development that you do not want deployed, you should add that file's directory with a tools.deps alias. (You cannot e.g. "subtract" the file with a .gitignore. Ions use git only for calculating a SHA.)
  • Provision AWS lambda concurrency to mitigate cold starts.
  • If you have dependencies that are needed only in dev and test, place them under a tools.deps alias so that you do not deploy them to production.

Older Versions of Datomic Cloud

Datomic versions 781-9041 (2021-03-02) and older do not manage an API Gateway for ion applications. If you are running an older version of Datomic, the easiest way to deploy web services is to upgrade Datomic and follow the instructions above.

If this is not possible, the instructions below cover managing your own API Gateway. The instructions vary based on your system topology. For Datomic Solo, you can connect an API Gateway to a web lambda proxy. For Production, you can connect an API Gateway to your Network Load Balancer.

Web Lambda Proxies

If you are running the Solo Topology, version 781-9041 and older, you can expose a web service using API Gateway Lambda Proxy integration.

To deploy a Lambda proxy web endpoint, you will need to:

Create a Web Lambda Proxy Entry Point

Given a var that implements a web entry point, you can create a Lambda proxy entry point by calling ionize. Ionize takes a web entry point function and returns a function that implements the special contract required by a lambda proxy.

The ion-starter sample project has an example: 

(def get-items-by-type-lambda-proxy
  (apigw/ionize get-items-by-type))

You should then name this lambda proxy as a Lambda entry point (not an HTTP Direct entry point!) in your ion-config.

Again, the ion-starter sample project has an example.

The input map for web lambda proxies includes the following additional keys:

Key Req'd Value Example
:json No api-gateway/json {"resource":"{proxy+}","path":"datomic","httpMethod":"POST" … }
:data No api-gateway/data {"Path": "/datomic","Querystringparameters": null, "Pathparameters": {"Proxy": "datomic"}…

Configure an API Gateway for a Web Lambda Proxy

After you push and deploy an ion application with a web Lambda Proxy entry point, you can create an API gateway as follows:

  • Go to the AWS API gateway console to create a new AWS API gateway.
    • If this is the first AWS API Gateway you are creating, choose "Get started".
    • Then click "Ok" and choose the "New API" radio button.
    • If you have existing AWS API Gateways, choose "Create API".
  • Give your API a unique name, e.g. "ion-get-items-by-type". Leave the other default settings.
  • Click "Create API" in the bottom right.
  • Under the "Actions" dropdown, choose "Create resource".
  • Check the box "Configure as proxy resource."
    • Other fields will be updated with defaults. Leave these as is.
  • If you will need to support CORS, then select the "Enable API Gateway CORS" option. This will set up an OPTIONS method with a Mock Integration Request which can be changed to fit your needs.
  • Click "Create resource".
  • Set the lambda function to the name of your Lambda proxy and click "Save".
    • Note that the autocomplete does not always work correctly on this step, so trust and verify your own spelling.
  • Choose "Ok" to give API Gateway permission to call your lambda.
  • Under your API on the left side of the UI, click on the bottom choice "Settings".
  • Choose "Add binary media type", and add the */* type, then "Save changes".

You can then deploy your API gateway on the public internet:

  • Select your API name (if you have not already), and choose Deploy API under the Actions dropdown.
  • Choose "New stage" as the deployment stage.
  • Choose a "Stage name", then choose "Deploy".

The top of the Stage Editor with show the Invoke URL for your deployed app. Your service will be available at the path https://$(Invoke URL)/datomic.

API Gateway to Network Load Balancer

If you are running the Production Topology, version 781-9041 and older, you can expose a web service by connecting an API Gateway to a compute group's Network Load Balancer as follows: