This section describes our recommended design of a full-stack Daml application.
The above image shows the recommended architecture. Of course there are many ways how you can change the architecture and technology stack to fit your needs, which we’ll mention in the corresponding sections. Note that the Participant Node is integrated into the Daml drivers in some cases rather than being part of the Application Backend. See Daml Ecosystem Overview for more details.
To get started quickly with the recommended application architecture, generate a new project using the
daml new --template=create-daml-app my-project-name
create-daml-app is a small, but fully functional demo application implementing the recommended
architecture, providing you with an excellent starting point for your own application. It showcases
- using Daml React libraries
- quick iteration against the Daml Sandbox.
- deploying your application in the cloud as a Docker container
The backend for your application can be any Daml ledger implementation running your DAR (Daml Archive) file.
We recommend using the Daml JSON API as an interface to your frontend. It is served by the HTTP JSON API server connected to the ledger API server. It provides simple HTTP endpoints to interact with the ledger via GET/POST requests. However, if you prefer, you can also use the gRPC Ledger API directly.
When you use the
create-daml-app template application, you can start a Daml Sandbox together
with a JSON API server by running
daml start --start-navigator=no
in the root of the project. Daml Sandbox exposes the same Daml Ledger API a Participant Node would
expose without requiring a fully-fledged Daml network to back the application. Once your
application matures and becomes ready for production, the
daml deploy command helps you deploy
your frontend and Daml artifacts of your project to a production Daml network. See
Deploying to Daml Ledgers for an in depth manual for specific ledgers.
We recommended building your frontend with the React framework. However, you can choose virtually any language for your frontend and interact with the ledger via HTTP JSON endpoints. In addition, we provide support libraries for Java and Scala and you can also interact with the gRPC Ledger API directly.
We provide two libraries to build your React frontend for a Daml application.
|@daml/react||React hooks to query/create/exercise Daml contracts|
|@daml/ledger||Daml ledger object to connect and directly submit commands to the ledger|
You can install any of these libraries by running
npm install <library> in the
ui directory of
your project, e.g.
npm install @daml/react. Please explore the
create-daml-app example project
to see the usage of these libraries.
daml codegen js .daml/dist/<your-project-name.dar> -o ui/daml.js
types and templates in the DALF and TypeScript typings them. In
ui/package.json refers to these
libraries via the
"create-daml-app": "file:../daml.js/create-daml-app-0.1.0" entry in the
@daml/types and the generated
daml.js libraries provide you with the necessary code to
connect and issue commands against your ledger.
The SDK enables a local development environment with fast iteration cycles:
- The integrated VSCode IDE (
daml studio) runs your Scripts on any change to your Daml models. See Daml Script.
Together, these features can provide you with very tight feedback loops while developing your Daml application, all the way from your Daml contracts up to your web UI. A typical Daml developer workflow is to
- Make a small change to your Daml data model
- Optionally test your Daml code with Daml Script
- Edit your React components to be aligned with changes made in Daml code
- Extend the UI to make use of the newly introduced feature
- Make further changes either to your Daml and/or React code until you’re happy with what you’ve developed
See Your First Feature for a more detailed walkthrough of these steps.
The interaction of a Daml application with the ledger is inherently asynchronous: applications send commands to the ledger, and some time later they see the effect of that command on the ledger.
There are several things that can fail during this time window: the application can crash, the participant node can crash, messages can be lost on the network, or the ledger may be just slow to respond due to a high load.
If you want to make sure that a command is not executed twice, your application needs to robustly handle all the various failure scenarios. Daml ledgers provide a mechanism for command deduplication to help deal with this problem.
For each command application provide a command ID and an optional parameter that specifies the deduplication time. If the latter parameter is not specified in the command submission itself, the ledger will fall back to using the configured maximum deduplication time. The ledger will then guarantee that commands for the same submitting party and command ID will be ignored within the deduplication time window.
To use command deduplication, you should:
- Use generous values for the deduplication time. It should be large enough such that you can assume the command was permanently lost if the deduplication time has passed and you still don’t observe any effect of the command on the ledger (i.e. you don’t see a transaction with the command ID via the transaction service).
- Make sure you set command IDs deterministically, that is to say: the “same” command must use the same command ID. This is useful for the recovery procedure after an application crash/restart, in which the application inspects the state of the ledger (e.g. via the Active contracts service) and sends commands to the ledger. When using deterministic command IDs, any commands that had been sent before the application restart will be discarded by the ledger to avoid duplicate submissions.
- If you are not sure whether a command was submitted successfully, just resubmit it. If the new command was submitted within the deduplication time window, the duplicate submission will safely be ignored. If the deduplication time window has passed, you can assume the command was lost or rejected and a new submission is justified.
For more details on command deduplication, see the Ledger API Services documentation.
Dealing with failures¶
In order to restart your application from a previously known ledger state, your application must keep track of the last ledger offset received from the transaction service or the command completion service.
By persisting this offset alongside the relevant state as part of a single, atomic operation, your application can resume from where it left off.
Failing over between Ledger API endpoints¶
Some Daml Ledgers support exposing multiple eventually consistent Ledger API endpoints where command deduplication works across these Ledger API endpoints. For example, these endpoints might be hosted by separate Ledger API servers that replicate the same data and host the same parties. Contact your ledger operator to find out whether this applies to your ledger.
Below we describe how you can build your application such that it can switch between such eventually consistent Ledger API endpoints to tolerate server failures. You can do this using the following two steps.
First, your application must keep track of the ledger offset as described in the paragraph about crash recovery. When switching to a new Ledger API endpoint, it must resume consumption of the transaction (tree) and/or the command completion streams starting from this last received offset.
Second, your application must retry on
OUT_OF_RANGE errors (see
gRPC status codes)
received from a stream subscription – using an appropriate backoff strategy
to avoid overloading the server. Such errors can be raised because of eventual
consistency. The Ledger API endpoint that the application is newly subscribing
to might be behind the endpoint that it subscribed to before the switch, and
needs time to catch up. Thanks to eventual consistency this is guaranteed to
happen at some point in the future.
Once the application successfully subscribes to its required streams on the new endpoint, it will resume normal operation.
Dealing with time¶
The Daml language contains a function getTime which returns a rough estimate of “current time” called Ledger Time. The notion of time comes with a lot of problems in a distributed setting: different participants might run different clocks, there may be latencies due to calculation and network, clocks may drift against each other over time, etc.
In order to provide a useful notion of time in Daml without incurring severe performance or liveness penalties, Daml has two notions of time: Ledger Time and Record Time:
- As part of command interpretation, each transaction is automatically assigned a Ledger Time by the participant server.
- All calls to
getTimewithin a transaction return the Ledger Time assigned to that transaction.
- Ledger Time is chosen (and validated) to respect Causal Monotonicity: The Create action on a contract c always precedes all other actions on c in Ledger Time.
- As part of the commit/synchronization protocol of the underlying infrastructure, every transaction is assigned a Record Time, which can be thought of as the infrastructures “system time”. It’s the best available notion of “real time”, but the only guarantees on it are the guarantees the underlying infrastructure can give. It is also not known at interpretation time.
- Ledger Time is kept close to “real time” by bounding it against Record Time. Transactions where Ledger and Record Time are too far apart are rejected.
Some commands might take a long time to process, and by the time the resulting transaction is about to be committed to the ledger, it might violate the condition that Ledger Time should be reasonably close to Record Time (even when considering the ledger’s tolerance interval). To avoid such problems, applications can set the optional parameters min_ledger_time_abs or min_ledger_time_rel that specify (in absolute or relative terms) the minimal Ledger Time for the transaction. The ledger will then process the command, but wait with committing the resulting transaction until Ledger Time fits within the ledger’s tolerance interval.
How is this used in practice?
- Be aware that
getTimeis only reasonably close to real time, and not completely monotonic. Avoid Daml workflows that rely on very accurate time measurements or high frequency time changes.
min_ledger_time_relif the duration of command interpretation and transmission is likely to take a long time relative to the tolerance interval set by the ledger.
- In some corner cases, the participant node may be unable to determine a suitable Ledger Time by itself. If you get an error that no Ledger Time could be found, check whether you have contention on any contract referenced by your command or whether the referenced contracts are sensitive to small changes of
For more details, see Background concepts - time.