Managing cycles
Usage of the Internet Computer is measured, and paid for, in cycles. The Internet Computer maintains a balance of cycles per canister smart contract. In addition, cycles can be transferred between canisters.
In Motoko programs targeting the Internet Computer, each actor represents an Internet Computer canister, and has an associated balance of cycles. The ownership of cycles can be transferred between actors. Cycles are selectively sent and received through messages, that is, shared function calls. A caller can choose to transfer cycles with a call, and a callee can choose to accept cycles that are made available by the caller. Unless explicitly instructed, no cycles are transferred by callers or accepted by callees.
Callees can accept all, some or none of the available cycles up to limit determined by their actor’s current balance. Any remaining cycles are refunded to the caller. If a call traps, all its accompanying cycles are automatically refunded to the caller, without loss.
In future, we may see Motoko adopt dedicated syntax and types to support safer programming with cycles. For now, we provide a temporary way to manage cycles through a low-level imperative API provided by the ExperimentalCycles library in package base
.
This library is subject to change and likely to be replaced by more high-level support for cycles in later versions of Motoko.
The ExperimentalCycles
Library
The ExperimentalCycles
library provides imperative operations for observing an actor’s current balance of cycles, transferring cycles and observing refunds.
The library provides the following operations:
func balance() : (amount : Nat)
func available() : (amount : Nat)
func accept(amount : Nat) : (accepted : Nat)
func add(amount : Nat) : ()
func refunded() : (amount : Nat)
Function balance()
returns the actor’s current balance of cycles as amount
. Function balance()
is stateful and may return different values after calls to accept(n)
, calling a function after add
ing cycles, or resuming from await (reflecting a refund).
Since cycles measure computational resources spent, the value of balance()
generally decreases from one shared function call to the next.
Function available()
, returns the currently available amount
of cycles. This is the amount received from the current caller, minus the cumulative amount accept
ed so far by this call. On exit from the current shared function or async
expression via return
or throw
any remaining available amount is automatically refunded to the caller.
Function accept
transfers amount
from available()
to balance()
. It returns the amount actually transferred, which may be less than requested, for example, if less is available, or if canister balance limits are reached.
Function add(amount)
indicates the additional amount of cycles to be transferred in the next remote call, i.e. evaluation of a shared function call or async
expression. Upon the call, but not before, the total amount of units add
ed since the last call is deducted from balance()
. If this total exceeds balance()
, the caller traps, aborting the call.
the implicit register of added amounts, incremented on each add
, is reset to zero on entry to a shared function, and after each shared function call or on resume from an await.
Function refunded()
reports the amount
of cycles refunded in the last await
of the current context, or zero if no await has occurred yet. Calling refunded()
is solely informational and does not affect balance()
. Instead, refunds are automatically added to the current balance, whether or not refunded
is used to observe them.
Example
To illustrate, we will now use the ExperimentalCycles
library to implement a toy piggy bank for saving cycles.
Our piggy bank has an implicit owner, a benefit
callback and a fixed capacity
, all supplied at time of construction. The callback is used to transfer withdrawn amounts.
import Cycles "mo:base/ExperimentalCycles";
shared(msg) actor class PiggyBank(
benefit : shared () -> async (),
capacity: Nat
) {
let owner = msg.caller;
var savings = 0;
public shared(msg) func getSavings() : async Nat {
assert (msg.caller == owner);
return savings;
};
public func deposit() : async () {
let amount = Cycles.available();
let limit : Nat = capacity - savings;
let acceptable =
if (amount <= limit) amount
else limit;
let accepted = Cycles.accept<system>(acceptable);
assert (accepted == acceptable);
savings += acceptable;
};
public shared(msg) func withdraw(amount : Nat)
: async () {
assert (msg.caller == owner);
assert (amount <= savings);
Cycles.add<system>(amount);
await benefit();
let refund = Cycles.refunded();
savings -= amount - refund;
};
}
The owner of the bank is identified with the (implicit) caller of constructor PiggyBank()
, using the shared pattern, shared(msg)
. Field msg.caller
is a Principal
and is stored in private variable owner
(for future reference). See Principals and caller identification for more explanation of this syntax.
The piggy bank is initially empty, with zero current savings
.
Only calls from owner
may:
query the current
savings
of the piggy bank (functiongetSavings()
), orwithdraw amounts from the savings (function
withdraw(amount)
).
The restriction on the caller is enforced by the statements assert (msg.caller ==
owner)
, whose failure causes the enclosing function to trap, without revealing the balance or moving any cycles.
Any caller may deposit
an amount of cycles, provided the savings will not exceed capacity
, breaking the piggy bank. Because the deposit function only accepts a portion of the available amount, a caller whose deposit exceeds the limit will receive an implicit refund of any unaccepted cycles. Refunds are automatic and ensured by the Internet Computer infrastructure.
Since transfer of cycles is one-directional (from caller to callee), retrieving cycles requires the use of an explicit callback (the benefit
function, taken by the constructor as an argument). Here, benefit
is called by the withdraw
function, but only after authenticating the caller as owner
. Invoking benefit
in withdraw
inverts the caller/caller relationship, allowing cycles to flow "upstream".
Note that the owner of the PiggyBank
could, in fact, supply a callback that rewards a beneficiary distinct from owner
.
Here’s how an owner, Alice
, might use an instance of PiggyBank
:
import Cycles = "mo:base/ExperimentalCycles";
import Lib = "PiggyBank";
actor Alice {
public func test() : async () {
Cycles.add<system>(10_000_000_000_000);
let porky = await Lib.PiggyBank(Alice.credit, 1_000_000_000);
assert (0 == (await porky.getSavings()));
Cycles.add<system>(1_000_000);
await porky.deposit();
assert (1_000_000 == (await porky.getSavings()));
await porky.withdraw(500_000);
assert (500_000 == (await porky.getSavings()));
await porky.withdraw(500_000);
assert (0 == (await porky.getSavings()));
Cycles.add<system>(2_000_000_000);
await porky.deposit();
let refund = Cycles.refunded();
assert (1_000_000_000 == refund);
assert (1_000_000_000 == (await porky.getSavings()));
};
// Callback for accepting cycles from PiggyBank
public func credit() : async () {
let available = Cycles.available();
let accepted = Cycles.accept<system>(available);
assert (accepted == available);
}
}
Let’s dissect Alice
's code.
Alice
imports the PiggyBank
actor class as a library, so she can create a new PiggyBank
actor on demand.
Most of the action occurs in Alice
's test()
function:
Alice dedicates 10_000_000_000_000
of her own cycles for running the piggy bank, by calling Cycles.add(10_000_000_000_000)
just before creating a new instance, porky
, of the PiggyBank
, passing callback Alice.credit
and capacity (1_000_000_000
). Passing Alice.credit
nominates Alice
as the beneficiary of withdrawals. The 10_000_000_000_000
cycles, minus a small installation fee, are credited to porky
's balance without any further action by porky
initialization code. You can think of this as an electric piggy bank, that consumes its own resources as its used. Since constructing a PiggyBank
is asynchronous, Alice
needs to await
the result.
After creating porky
, she first verifies that the porky.getSavings()
is zero using an assert
.
Alice
dedicates 1_000_000
of her cycles (Cycles.add(1_000_000)
) to transfer to porky
with the next call to porky.deposit()
. The cycles are only consumed from Alice’s balance if the call to porky.deposit()
succeeds (which it should).
Alice
now withdraws half the amount, 500_000
, and verifies that porky
's savings have halved. Alice
eventually receives the cycles via a callback to Alice.credit()
, initiated in porky.withdraw()
. Note the received cycles are precisely the cycles add
ed in porky.withdraw()
, before it invokes its benefit
callback, that is, Alice.credit
.
Alice
withdraws another 500_000
cycles to wipe out her savings.
Alice
vainly tries to deposit 2_000_000_000
cycles into porky
but this exceeds porky
's capacity by half, so porky
accepts 1_000_000_000
and refunds the remaining 1_000_000_000
to Alice
. Alice
verifies the refund amount (Cycles.refunded()
), which has (already) been automatically restored to her balance. She also verifies porky
's adjusted savings.
Alice
's credit()
function simply accepts all available cycles by calling Cycles.accept(available)
, checking the actually accepted
amount with an assert.
For this example, Alice is using her (readily available) cycles, that she already owns.
Because porky
consumes cycles in its operation, it is possible for porky
to spend some or even all of Alice’s cycle savings before she has a chance to retrieve them.