Periodic tasks and timers

Advanced

Concept

Overview

Unlike other blockchains, ICP canisters can automatically execute tasks after a specified delay or interval.

There are two ways to schedule an automatic canister execution on ICP:

1. Timers: single-expiration or periodic canister calls with specified minimum timeout or interval. A canister can implement multiple timers.
2. Heartbeats: legacy periodic canister invocations with intervals close to the blockchain finalization rate (1s). Heartbeats are supported by ICP for backward compatibility and some very special use cases. Newly developed canisters should prefer using timers over the heartbeats.

Timers

Timers are implemented on two layers:

• The protocol level implementation: the Internet Computer protocol supports minimalistic on-shot global timer per canister via ic0.global_timer_set() system API call and canister_global_timer handler (see the Internet Computer interface specification).

• The CDK timers library level: the library wraps the minimalistic protocol implementation, adding multiple and periodic timers on top. Canister developers can enjoy the familiar timers functionality using the CDK timers library for:

• Rust

• Motoko

• TypeScript (Azle)

• Python (Kybra)

Internally the CDK timers libraries do the following:

1. The library keeps a global list of multiple and periodic tasks inside the canister.
2. Calls the ic0.global_timer_set() system API to schedule the next task from the list.
3. Implements the canister_global_timer method with the following logic:
   • For each expired task, the handler initiates a self canister call to isolate the tasks from each other and from the library code. Note, the normal inter-canister call costs costs apply.
   • Reschedules periodic tasks at the end of their execution.
   • Calls the ic0.global_timer_set() system API to schedule the next task.

The library does not handle the canister upgrades. It is up to the canister developer to serialize the timers in the canister_pre_upgrade hook and reactivate the timers in the canister_post_upgrade method if necessary.

For the code composability reasons, i.e. to be able to use different libraries with timers in a single project, canister developers are encouraged to use the CDK timers libraries over the protocol level API or the heartbeats.

Single timer example

Motoko

import { print } = "mo:base/Debug";
import { recurringTimer } = "mo:base/Timer";

actor Alarm {
  let N = 5;
  private func ring() : async () {
    print("Motoko Timer Ring!");
  };
  ignore recurringTimer<system>(#seconds N, ring);
};

Rust

use std::time::Duration;

const N: Duration = Duration::from_secs(5);

fn ring() {
    ic_cdk::println!("Rust Timer Ring!");
}

#[ic_cdk::init]
fn init() {
    let _timer_id = ic_cdk_timers::set_timer_interval(N, ring);
}

#[ic_cdk::post_upgrade]
fn post_upgrade() {
    init();
}

TypeScriptBeta

import { IDL, setTimer, update } from 'azle';

export default class {
    @update([], IDL.Nat64)
    createTimer(): bigint {
        const timerId = setTimer(1_000n, () =>
            console.log('timer callback called')
        );

        return timerId;
    }
}

PythonBeta

from kybra import (
    Duration,
    ic,
    TimerId,
    update,
)


@update
def set_timer(delay: Duration) -> TimerId:
    return ic.set_timer(delay, timer_callback)


def timer_callback():
    ic.print("timer_callback")

Multiple timers example

To use multiple timers in a canister, simply create multiple timer definitions:

Motoko

In Motoko, call the recurringTimer function multiple times, setting different durations and callback functions for each timer:

import { print } = "mo:base/Debug";
import { recurringTimer } = "mo:base/Timer";

actor Alarm {

  let N1 = 5;
  let N2 = 10;

  private func ring1() : async () {
    print("Motoko Timer 1 Ring!");
  };
  private func ring2() : async () {
    print("Motoko Timer 2 Ring!");
  };
  ignore recurringTimer<system>(#seconds N1, ring1);
  ignore recurringTimer<system>(#seconds N2, ring2);
};

Rust

In Rust, call the set_timer_interval function multiple times, setting different durations and callback functions for each timer:

use std::time::Duration;

const N: Duration = Duration::from_secs(5);

fn ring() {
    ic_cdk::println!("Rust Timer Ring!");
}

#[ic_cdk::init]
fn init() {
    let _timer_id = ic_cdk_timers::set_timer_interval(N, ring);
}

#[ic_cdk::post_upgrade]
fn post_upgrade() {
    init();
}

TypeScriptBeta

import {
    call,
    clearTimer,
    IDL,
    query,
    setTimer,
    setTimerInterval,
    update
} from 'azle';

const StatusReport = IDL.Record({
    single: IDL.Bool,
    inline: IDL.Int8,
    capture: IDL.Text,
    repeat: IDL.Int8,
    singleCrossCanister: IDL.Vec(IDL.Nat8),
    repeatCrossCanister: IDL.Vec(IDL.Nat8)
});
type StatusReport = {
    single: boolean;
    inline: number;
    capture: string;
    repeat: number;
    singleCrossCanister: Uint8Array;
    repeatCrossCanister: Uint8Array;
};

const TimerIds = IDL.Record({
    single: IDL.Nat64,
    inline: IDL.Nat64,
    capture: IDL.Nat64,
    repeat: IDL.Nat64,
    singleCrossCanister: IDL.Nat64,
    repeatCrossCanister: IDL.Nat64
});
type TimerIds = {
    single: bigint;
    inline: bigint;
    capture: bigint;
    repeat: bigint;
    singleCrossCanister: bigint;
    repeatCrossCanister: bigint;
};

let statusReport: StatusReport = {
    single: false,
    inline: 0,
    capture: '',
    repeat: 0,
    singleCrossCanister: Uint8Array.from([]),
    repeatCrossCanister: Uint8Array.from([])
};

export default class {
    @update([IDL.Nat64])
    clearTimer(timerId: bigint): void {
        clearTimer(timerId);
        console.info(`timer ${timerId} cancelled`);
    }

    @update([IDL.Nat64, IDL.Nat64], TimerIds)
    setTimers(delay: bigint, interval: bigint): TimerIds {
        const capturedValue = '🚩';

        const singleId = setTimer(delay, oneTimeTimerCallback);

        const inlineId = setTimer(delay, () => {
            statusReport.inline = 1;
            console.info('Inline timer called');
        });

        const captureId = setTimer(delay, () => {
            statusReport.capture = capturedValue;
            console.info(`Timer captured value ${capturedValue}`);
        });

        const repeatId = setTimerInterval(interval, () => {
            statusReport.repeat++;
            console.info(`Repeating timer. Call ${statusReport.repeat}`);
        });

        const singleCrossCanisterId = setTimer(
            delay,
            singleCrossCanisterTimerCallback
        );

        const repeatCrossCanisterId = setTimerInterval(
            interval,
            repeatCrossCanisterTimerCallback
        );

        return {
            single: singleId,
            inline: inlineId,
            capture: captureId,
            repeat: repeatId,
            singleCrossCanister: singleCrossCanisterId,
            repeatCrossCanister: repeatCrossCanisterId
        };
    }

    @query([], StatusReport)
    statusReport(): StatusReport {
        return statusReport;
    }
}

function oneTimeTimerCallback(): void {
    statusReport.single = true;
    console.info('oneTimeTimerCallback called');
}

async function singleCrossCanisterTimerCallback(): Promise<void> {
    console.info('singleCrossCanisterTimerCallback');

    statusReport.singleCrossCanister = await getRandomness();
}

async function repeatCrossCanisterTimerCallback(): Promise<void> {
    console.info('repeatCrossCanisterTimerCallback');

    statusReport.repeatCrossCanister = Uint8Array.from([
        ...statusReport.repeatCrossCanister,
        ...(await getRandomness())
    ]);
}

async function getRandomness(): Promise<Uint8Array> {
    return await call('aaaaa-aa', 'raw_rand', {
        returnIdlType: IDL.Vec(IDL.Nat8)
    });
}

PythonBeta

from kybra import (
    Async,
    blob,
    CallResult,
    Duration,
    ic,
    match,
    nat8,
    query,
    Record,
    TimerId,
    update,
    void,
)
from kybra.canisters.management import management_canister


class StatusReport(Record):
    single: bool
    inline: nat8
    capture: str
    repeat: nat8
    single_cross_canister: blob
    repeat_cross_canister: blob


class TimerIds(Record):
    single: TimerId
    inline: TimerId
    capture: TimerId
    repeat: TimerId
    single_cross_canister: TimerId
    repeat_cross_canister: TimerId


status: StatusReport = {
    "single": False,
    "inline": 0,
    "capture": "",
    "repeat": 0,
    "single_cross_canister": bytes(),
    "repeat_cross_canister": bytes(),
}


@update
def clear_timer(timer_id: TimerId) -> void:
    ic.clear_timer(timer_id)
    ic.print(f"timer {timer_id} cancelled")


@update
def set_timers(delay: Duration, interval: Duration) -> TimerIds:
    captured_value = "🚩"

    single_id = ic.set_timer(delay, one_time_timer_callback)

    # Note: You cannot set global variables from within a lambda but you can
    # call a function that sets a global variable. So we've moved the "setting"
    # functionality out into helper functions while the printing is kept here in
    # the lambda.

    inline_id = ic.set_timer(
        delay,
        lambda: update_inline_status() or ic.print("Inline timer called"),
    )

    capture_id = ic.set_timer(
        delay,
        lambda: update_capture_status(captured_value)
        or ic.print(f"Timer captured value: {captured_value}"),
    )

    repeat_id = ic.set_timer_interval(interval, repeat_timer_callback)

    single_cross_canister_id = ic.set_timer(delay, single_cross_canister_timer_callback)

    repeat_cross_canister_id = ic.set_timer_interval(
        interval, repeat_cross_canister_timer_callback
    )

    return {
        "single": single_id,
        "inline": inline_id,
        "capture": capture_id,
        "repeat": repeat_id,
        "single_cross_canister": single_cross_canister_id,
        "repeat_cross_canister": repeat_cross_canister_id,
    }


@query
def status_report() -> StatusReport:
    return status


def one_time_timer_callback():
    status["single"] = True
    ic.print("one_time_timer_callback called")


def repeat_timer_callback():
    status["repeat"] += 1
    ic.print(f"Repeating timer. Call {status['repeat']}")


def update_inline_status():
    status["inline"] = 1


def update_capture_status(value: str):
    status["capture"] = value


def single_cross_canister_timer_callback() -> Async[void]:
    ic.print("single_cross_canister_timer_callback")

    result: CallResult[blob] = yield management_canister.raw_rand()

    def handle_ok(ok: blob):
        status["single_cross_canister"] = ok

    match(result, {"Ok": handle_ok, "Err": lambda err: ic.print(err)})


def repeat_cross_canister_timer_callback() -> Async[void]:
    ic.print("repeat_cross_canister_timer_callback")

    result: CallResult[blob] = yield management_canister.raw_rand()

    def handle_ok(ok: blob):
        status["repeat_cross_canister"] += ok

    match(result, {"Ok": handle_ok, "Err": lambda err: ic.print(err)})

Timers library limitations

Despite its superiority over the heartbeats, the CDK timers library has a few known shortcomings:

• Canister upgrades: the library keeps a global list of multiple and periodic tasks inside the canister heap. During the canister upgrade, a fresh WebAssembly state is created, all the timers are deactivated and the list of timers is cleared. It is up to the canister developer to serialize the timers in the canister_pre_upgrade and reactivate them in the canister_post_upgrade method if needed.
• Self canister calls: to isolate the tasks from each other and from the scheduling logic, CDK timers library initiates a self canister call to execute each task. There are a few implications:
  • Normal inter-canister call costs apply to execute each task. Note, timers are still more cost effective than the heartbeats.
  • The tasks to execute are added at the end of the canister input queue. Depending on the canister and subnet load, the actual execution might be delayed.
  • As the canister output queue is limited in size (500 messages at the moment), this implicitly limits the number of tasks which might be scheduled in one round.
• Advanced scheduling: the CDK timers library uses relative time to schedule tasks. To use an absolute time, canister developers should calculate the duration between now and the point in time, or use a third party library.

Heartbeats

Once an Internet Computer canister exports the canister_heartbeat function, it will be called every subnet heartbeat interval (see the Internet Computer interface specification).

The only way to disable the heartbeats is to upgrade the canister to a version which does not export the canister_heartbeat method. Also, the heartbeat interval is implementation-defined, and there is no way to adjust it.

Because of those limitations, in most cases CDK timers library for Rust or Motoko is a better option to schedule periodic tasks.

Frequently asked questions

• Do timers support deterministic time slicing (DTS)?

  Yes, as the CDK timers library initiates a self canister call to execute each task, normal update message instruction limits apply with DTS enabled.

• What happens if a timer handler awaits for a call?

  Normal await point rules apply: any new execution can start at the await point: a new message, another timer handler or a heartbeat. Once that new execution is finished or reached its await point, the execution of the current timer handler might be resumed.

• What happens if a 1s periodic timer is executing for 2s?

  If there are no await points in the timer handler, the periodic timer will be rescheduled at the end of the execution. If there are await points, it's implementation-defined when the periodic timer is rescheduled.

Tutorials and examples

Tutorials and examples using Motoko

• Motoko developer guide: Timers.
• Motoko developer guide: Heartbeats.

Tutorials and examples using Rust

• Backend tutorial: Using timers.
• Example: Periodic tasks and timers (compares the costs of timers and heartbeats).