Iroha 2

Appearance

Triggers ​

Certain things, such as changing the state of an entity, committing a block or executing an Iroha Special Instruction (ISI), can emit events, and you can attach triggers to these events.

A trigger is a fairly basic entity that can be registered. Just like with Accounts, to register a trigger, you submit a RegisterBox::Trigger, which contains the necessary information:

  • an account ID, which should ideally be a brand new account that you register in the same transaction
  • an executable, which itself is either a Vec<Instruction> or a WASM blob
  • an EventFilter[1], which is something that combs through all[2] events and returns true when it finds the matching event to start the execution

Let's take a closer look at how triggers work.

The Anatomy of a Trigger ​

A trigger has roughly the following form:

rust
struct Trigger {
  id: TriggerId,
  action: Action,
}

Trigger.id ​

The TriggerId is a simple wrapper around a single Name, a string with no whitespaces and no reserved characters (@, #, $).

A typical domain-scoped trigger looks like trigger_id$domain_name, while a bare trigger looks like @@trigger_id, which makes these names easy to parse.

Trigger.action ​

An Action is the heart of the trigger. It is defined like this:

rust
struct Action {
  executable: Executable,
  repeats: Repeats,
  technical_account: AccountId,
  filter: EventFilter,
  metadata: Metadata,
}

Action.executable ​

The executable linked to this action, either a Vec<Instruction> or a WASM binary.

Action.repeats ​

The Repeats is a universal enumeration of all possible repetition schemes.

rust
enum Repeats {
  Indefinitely,
  Exactly(u32),
}

Action.technical_account ​

A technical account is the account that would (in theory) be responsible for the execution environment and be the authority for Instruction execution.

For now, you can leave this as the account that registered the trigger. If you have been following the tutorial, this is alice@wonderland. However, later on we will show you why you'd want to create a brand new account for those purposes.

INFO

Note that you can only use the account that already exists in order to be able to register a new trigger.

Action.filter ​

A filter is what determines what kind of trigger you're dealing with. All triggers respond to events, but the precise type of event that activates a trigger depends on which EventFilter was used.

The reason why we chose this architecture is simple; front end code has an abundance of event filters, and so, your knowledge of filters is transferable to writing smart contracts.

Action.metadata ​

This Metadata is the same kind of Metadata that can be attached to accounts, domains, assets, or transactions. This is the storage for trigger data.

You can learn more about metadata in a dedicated section.

How Triggers Work ​

As we already said, the filter that is used to register a trigger determines what kind of trigger this is. It is, of course, also determines how the trigger works, e.g. when it is executed. We will go into more details about the types of triggers in just a moment.

First, we shall point out that there two other characteristics of a trigger that determine how this trigger works: its scope and repetition schema.

Scope ​

Triggers can be scoped and un-scoped. Iroha supports both un-scoped system-wide triggers as well as domain-scoped triggers. Since system-wide triggers scan all events, and domain-scoped triggers only scan events emitted in a certain domain, it is highly recommended to use domain-scoped triggers where possible.

INFO

Be mindful of the limitations. Currently triggers don't check for permissions , so they can do things your account wouldn't be allowed to. Un-scoped triggers process every event, and the amount of work grows quadratically.

Domain-scoped Triggers ​

While un-scoped triggers check all events of a specified type, domain-scoped triggers only look for events in a given domain. These triggers are more optimised compared to un-scoped triggers.

You can use FindTriggersByDomainId query to find triggers for the given domain.

When you register a domain-scoped trigger, you need to add the domain id to the trigger id using $ symbol: my_trigger$my_domain.

Repetition Schema ​

Each such trigger can be set to repeat either Indefinitely or Exactly(n) times, where n is a 32-bit integer. Once the number of repetitions reaches zero, the trigger is gone. That means that if your trigger got repeated exactly n times, you can't Mint new repetitions, you have to Register it again, with the same name.

After a trigger is repeated for the last time, i.e. the execution count reaches 0, the trigger should be un-registered.

Types of Triggers ​

We shall cover the following basic types of triggers and provide you with the detailed information on how to use each of them:

All triggers are essentially event triggers. The type of a trigger is determined by the type of an event that trigger is associated with. This, in turn, is determined by the filter used to register a trigger.

Data Triggers ​

This category includes the largest variety of triggers. The events that are associated with this trigger type account for the vast majority of events in Ethereum. These are data-related events, such as: an account got registered, an asset got transferred, the Queen of Hearts decided to burn all of her assets.

Time Triggers ​

Time triggers behave slightly differently compared to data triggers. There are two sub-types of this type: scheduled triggers and pre-commit triggers.

Instead of processing all the events generated by normal transactions, all time triggers process one event: the block formation event.

The filters of scheduled triggers are only interested in the timestamp provided in that event, but not the block height, and not the current time. They are executed according to a certain schedule. Pre-commit triggers, on the other hand, are executed right before a block is committed.

Scheduled Triggers ​

When going through consensus, all peers must agree on which triggers got executed. Scheduled triggers can't use real time, because then you can easily create a situation when they would never agree: e.g. by giving the Repeats::Indefinitely trigger a period that is smaller than the time it takes to pass consensus. It's really that simple.

So instead of using the actual current time at each peer, we use the time when the block got started plus a small offset. All triggers before that point in time get executed. All triggers that would be executed after that time wait for the next block.

Why we use the offset

The reason why we add this offset has to do with Iroha being best effort.

Imagine if we didn't have the offset... Normally, triggers would be set to nice round numbers; e.g. 12:00, 12:05, 11:55, etc. (as opposed to e.g. 11:59). However, the consensus can start at any point in time and could last a while.

Suppose that the block started to form at 11:56 and consensus finished at 12:03 (which is optimistically quick). Let's consider different scenarios:

  • If your trigger was supposed to run at 11:55, you'd be happy, since your trigger got executed just 1 minute late.
  • If your trigger was supposed to run at 12:05, it will run in the next block, not the one that was formed at 11:56. If you're the author and you're looking at the time stamp of 12:03, it makes sense, your trigger wasn't supposed to run yet.
  • For the trigger scheduled for 12:00, the situation is different. You look at the clock, you see 12:03, which is when the blockchain explorer shows you the block data was committed, but you don't see your trigger. It was supposed to run, but didn't.

So, the offset is meant to anticipate when the block would get added to the chain, so that people who were just 4 minutes early don't have to be potentially several hours late.

Because more triggers get executed sooner, your throughput is also infinitesimally smaller.

We could also say "you should aim to execute your trigger slightly earlier than consensus starts", but people writing smart contracts already have too much to worry about.

Pre-commit Triggers ​

This is a variant of timed triggers that gets run before blocks with transactions get committed. It leaves a special event to be triggered in the next block. Effectively, it's a delayed pre-commit that can track the behaviour of transactions in the pipeline.

INFO

These triggers are not meant for restricting the execution of transactions.

If you want to stop your users from transferring more than X amount of Y to user Z, you really want a permission. While you could hack the pre-commit triggers to emulate the desired behaviour, this is not economical neither in terms of gas fees nor computation.

Until Iroha 2 supports WASM-based permissions validators, however, your only choice is pre-commit triggers.

By-call Triggers ​

These triggers only get executed once the CallTrigger(trigger_name) instruction is executed. They can be useful if you want to achieve dynamic linkage between different smart contract modules.

Space is precious, so you want to use as little of it as you can. Thus, you follow the UNIX design philosophy, and instead of creating one large smart contract, you create many smaller ones, and re-use as much logic as you can.

INFO

Of course, this is a rather exotic use case, so it shall be implemented last.

Event Triggers by Example ​

Now that we've gotten the theory out of the way, we want to sit down with the Mad Hatter, the March Hare, and the Dormouse and see if we can spin. Let's start with an event trigger that shows the basics.

1. Register accounts ​

We have mad_hatter@wonderland, dormouse@wonderland, march_hare@wonderland all of which share the fixed-point asset of tea#wonderland. The Mad Hatter has the tea pot, while the rest have a single cup of tea. When alice@wonderland had arrived, she got a nice cup of tea as well.

The way we get them in Rust code looks like this:

rust
let tea = AssetDefinitionId::new("tea", "wonderland")?;
let mad_hatter = AccountId::new("mad_hatter", "wonderland")?;
let dormouse = AccountId::new("dormouse", "wonderland")?;
let march_hare = AccountId::new("march_hare", "wonderland")?;
vec![
  RegisterBox::new(IdentifiableBox::from(NewAccount::new(mad_hatter.clone()))),
  RegisterBox::new(IdentifiableBox::from(NewAccount::new(march_hare.clone()))),
  RegisterBox::new(IdentifiableBox::from(NewAccount::new(dormouse.clone()))),
  RegisterBox::new(IdentifiableBox::from(AssetDefinition::new_fixed(tea.clone()))),
  MintBox::new(Value::Fixed(100.0_f64.try_into()?), IdBox::AssetId(AssetId::new(tea.clone(), mad_hatter.clone())))
  MintBox::new(Value::Fixed(1.0_f64.try_into()?), IdBox::AssetId(AssetId::new(tea.clone(), march_hare.clone())))
  MintBox::new(Value::Fixed(1.0_f64.try_into()?), IdBox::AssetId(AssetId::new(tea.clone(), dormouse.clone())))
  MintBox::new(Value::Fixed(1.0_f64.try_into()?), IdBox::AssetId(AssetId::new(tea.clone(), alice.clone())))
]

2. Register a trigger ​

We want a smart contract that transfers some tea from mad_hatter@wonderland to alice@wonderland when her tea reduces by a single cup.

For that we need to register a trigger. The boilerplate is straightforward:

rust
let id = TriggerId::new(Name::new("refresh_tea"));

let metadata = Metadata::new();

let executable = vec![
    TransferBox::new(
      IdBox::AssetId(AssetId::new(tea.clone(), mad_hatter.clone())),
      Value::Fixed(1_f64.try_into()?),
      IdBox::AssetId(AssetId::new(alice.clone(), mad_hatter.clone())),
    )
];

let repeats = Repeats::Indefinitely;

let technical_account = mad_hatter.clone();

let filter = _ // ...

3. Define an event filter ​

The event filter is where we need to spend some time and think. So far we've seen the Pipeline variety of filters. This time around, the filter is a Data kind. This type of filter is a tuple with a single variant, which is a FilterOpt of an EntityFilter:

  • FilterOpt stands for Optional Filter. It can either AcceptAll or accept BySome of another Filter.
  • An EntityFilter is a filter that matches ByAccount in our case, but can match by many other means. It wraps an AccountFilter, which matches various events produced on accounts.

What we want to do is create an event filter for when alice@wonderland drinks some of her tea, or, in other words, reduces the tea asset by any amount. To do this with the current API, we need to work bottom up.

An IdFilter is a filter that .matches(event) == true if and only if the identities are exactly the same. Everything that has an Id has a corresponding IdFilter.

INFO

An IdFilter is a parametric structure, an IdFilter that works on Peers has the type IdFilter<PeerId> and is not the same type as an IdFilter that works with AccountId; IdFilter<AccountId.

Now if we wanted a filter that will match whenever tea gets reduced, either through a Transfer or a Burn instruction, we need an AssetFilter. It needs to look at what the Id of the asset is, hence IdFilter<AssetDefinitionId> and ByRemoved.

rust
use FilterOpt::{BySome, AcceptAll};

let asset_filter = AssetFilter::new(BySome(IdFilter(tea.clone())), BySome(AssetEventFilter::ByRemoved));

So far so good?

Next, we want a filter that looks for changes in an asset for an account. Specifically:

rust
let account_filter = AccountFilter::new(BySome(IdFilter(alice.clone())), BySome(asset_filter));

Now, because of the way that parity_scale_codec works, we need to wrap this in various boxes.

rust
let filter = EventFilter::Data(BySome(EntityFilter::ByAccount(account_filter)));

4. Create a Trigger instance ​

After this somewhat laborious filter combination, we can create an Action

rust
let action = Action {
    executable, repeats, technical_account, filter, metadata
}

Which allows us to create an instance of a Trigger.

rust
let trigger = Trigger::new(id, action);

5. Create a transaction ​

Finally, in order to get said trigger onto the blockchain, we create a transaction with the following single instruction:

rust
Instruction::Register(RegisterBox::new(IdentifiableBox::Trigger(Box::new(trigger))));

How it works ​

The technical details of the created transaction are summarised as follows:

  • The (normal) instructions that either got submitted from WASM or directly from the client get executed. If there were any triggers that should have been registered, they get registered.
  • Using the total set of events that got generated during the execution of instructions, the triggers (including some that got registered just this round) get executed.
  • The events produced in the previous step get scheduled for execution in the next block.

INFO

The reason why the events caused by the execution of triggers get processed in the next block is that we don't want two triggers to inadvertently cause an infinite loop of instruction execution and break consensus.

Now each time Alice drinks some tea, the Mad Hatter pours in a whole new cup. The keen eyed among you will have noticed that the amount that Alice drinks is irrelevant to how much tea will be transferred. Alice may take a tiny sip and still be poured a whole new cup's worth.

INFO

We intend to address this issue in the future so that an emitted event also has an attached Value. We also intend to provide more event filter types. For example, we will have filters that match when the asset:

  • Decreases by any amount (current behaviour)
  • Decreases by more than (or exactly) the specific amount in one instruction
  • Decreases to below a certain threshold

Only the first type of event filter is implemented now, and the other two can be emulated using a WASM smart contract as the Executable.

Why not WASM

The above observation can be generalised. WASM can do any logic that a Turing complete machine could, using the data available via queries. So in theory for event-based triggers, you could create an AcceptAll event filter and do all of the processing using the key-value store as persistent storage, and then, determining if you want to execute using easy-to-understand Rust code, and not our admittedly cumbersome, EventFilters.

We don't want that. WASM takes up significantly more space, and takes longer to execute compared to plain ISI, which are slower than EventFilters. We want you to want to use the EventFilters because they would make the process much more efficient, and we are working tirelessly to make the experience of using event filters much less gruelling.

However, as was mentioned previously on several occasions, implementing a feature properly takes time and effort. Ergonomics must be balanced against safety and reliability, so we cannot just make things easier to use. We want them to retain many of the advantages of strong typing.

This is all a work in progress. Our code is in flux. We need time to play around with a particular implementation to optimise it.

Supported ISI ​

All Iroha Special Instructions work with triggers, specifically:

  • Register<Trigger>: Create a trigger object and subscribe it to global events.

  • Unregister<Trigger>: Remove a trigger from the World State View and stop passing events through it.

  • Mint<Trigger, u32>: For triggers that repeat a certain number of times, increase the number of times that the trigger gets executed. Can be done from inside the executable of the trigger.

  • Burn<Trigger, u32>: For triggers that repeat a certain number of times, decrease the number of times that the trigger gets executed.

    WARNING

    If the number provided is greater than the remaining number of repetitions, the instruction will fail to execute, and the transaction that it is part of will be rejected.

You can learn more about Iroha Special Instructions in the dedicated section.

Supported Queries ​

We list supported queries for triggers when we talk in more detail about queries in the next chapter.

  1. The documentation on the EventFilter types is under construction, as we are likely to make major changes to that particular architecture. For now, suffice it to say that you can look at the source code in iroha_data_model and see a few particularly interesting applications. ↩︎

  2. This behaviour is likely to change in future releases.

    ↩︎