Version: Canary 🚧

Oracles and Oracle Networks

Summary

Oracles are services that provide external data to a blockchain network
There are many Oracle providers on Solana.
You can build your own Oracle to create a custom data feed
You have to be careful when choosing your data feed providers

Lesson

Oracles are services that provide external data to a blockchain network. Blockchains by nature are siloed environments that do not know the outside world. This constraint inherently puts a limit on the use cases for decentralized applications (dApps). Oracles provide a solution to this limitation by creating a decentralized way to get real-world data onchain.

Oracles can provide just about any type of data onchain. Examples include:

Results of sporting events
Weather data
Political election results
Market data
Randomness

While the exact implementation may differ from blockchain to blockchain, generally Oracles work as follows:

Data is sourced off-chain.
That data is published onchain via a transaction, and stored in an account.
Programs can read the data stored in the account and use that data in the program's logic.

This lesson will go over the basics of how oracles work, the state of oracles on Solana, and how to effectively use oracles in your Solana development.

Trust and Oracle Networks

The primary hurdle oracles need to overcome is one of trust. Since blockchains execute irreversible financial transactions, developers and users alike need to know they can trust the validity and accuracy of oracle data. The first step in trusting an oracle is understanding how it's implemented.

Broadly speaking, there are three implementation types:

Single, centralized oracle publishes data onchain.
1. Pro: It’s simple; there's one source of truth.
2. Con: nothing is stopping the oracle provider from providing inaccurate data.
Network of oracles publish data and a consensus mechanism is used to determine the final result.
1. Pro: Consensus makes it less likely that bad data is pushed onchain.
2. Con: There is no way to disincentivize bad actors from publishing bad data and trying to sway the consensus.
Oracle network with some kind of proof of stake mechanism. I.e. require oracles to stake tokens to participate in the consensus mechanism. On every response, if an oracle deviates by some threshold from the accepted range of results, their stake is taken by the protocol and they can no longer report.
1. Pro: Ensures no single oracle can influence the final result too drastically, while also incentivizing honest and accurate actions.
2. Con: Building decentralized networks is challenging, incentives need to be set up properly and be sufficient to get participation, etc.

Depending on the use case of an oracle, any of the above solutions could be the right approach. For example, you might be perfectly willing to participate in a blockchain-based game that utilizes centralized oracles to publish gameplay information onchain.

On the other hand, you may be less willing to trust a centralized oracle providing price information for trading applications.

You may end up creating many standalone oracles for your own applications simply as a way to get access to off-chain information that you need. However, those oracles are unlikely to be used by the broader community where decentralization is a core tenet. You should also be hesitant to use centralized, third-party oracles yourself.

In a perfect world, all important and/or valuable data would be provided onchain through a highly efficient oracle network through a trustworthy proof of stake consensus mechanism. By introducing a staking mechanism, it’s in the oracle providers' best interest to ensure their data is accurate to keep their staked funds.

Even when an oracle network claims to have such a consensus mechanism, be sure to know the risks involved with using the network. If the total value involved of the downstream applications is greater than the oracle's allocated stake, oracles still may have sufficient incentive to collude.

It is your job to know how the oracle network is configured and judge if it can be trusted. Generally, Oracles should only be used for non-mission-critical functions and worst-case scenarios should be accounted for.

Oracles on Solana

There are many Oracle providers on Solana. Two of the most well known are Pyth and Switchboard. They’re each unique and follow slightly different design choices.

Pyth is primarily focused on financial data published from top-tier financial institutions. Pyth’s data providers publish the market data updates. These updates are then aggregated and published onchain by the Pyth program. The data sourced from Pyth is not completely decentralized as only approved data providers can publish data. The selling point of Pyth is that its data is vetted directly by the platform and sourced from financial institutions, ensuring higher quality.

Switchboard is a completely decentralized oracle network and has data of all kinds available. Check out all of the feeds on their website Additionally, anyone can run a Switchboard oracle and anyone can consume their data. This means you'll have to be diligent about researching feeds. We'll talk more about what to look for later in the lesson.

Switchboard follows a variation of the stake weighted oracle network described in the third option of the previous section. It does so by introducing what are called TEEs (Trusted Execution Environments). TEEs are secure environments isolated from the rest of the system where sensitive code can be executed. In simple terms, given a program and an input, TEEs can execute and generate an output along with a proof. If you’d like to learn more about TEEs, please read Switchboard’s documentation.

By introducing TEEs on top of stake weighted oracles, Switchboard is able to verify each oracle’s software to allow participation in the network. If an oracle operator acts maliciously and attempts to change the operation of the approved code, a data quote verification will fail. This allows Switchboard oracles to operate beyond quantitative value reporting, such as functions -- running off-chain custom and confidential computations.

Switchboard Oracles

Switchboard oracles store data on Solana using data feeds. These data feeds, also called aggregators, are each a collection of jobs that get aggregated to produce a single result. These aggregators are represented onchain as a regular Solana account managed by the Switchboard program. When an oracle updates, it writes the data directly to these accounts. Let's go over a few terms to understand how Switchboard works:

Aggregator (Data Feed) - Contains the data feed configuration, dictating how data feed updates get requested, updated, and resolved onchain from its assigned source. The Aggregator is the account owned by the Switchboard Solana program and is where the data is published onchain.
Job - Each data source should correspond to a job account. The job account is a collection of Switchboard tasks used to instruct the oracles on how to fetch and transform data. In other words, it stores the blueprints for how data is fetched off-chain for a particular data source.
Oracle - A separate program that sits between the internet and the blockchain and facilitates the flow of information. An oracle reads a feed’s job definitions, calculates the result, and submits its response onchain.
Oracle Queue - A group of oracles that get assigned to update requests in a round-robin fashion. The oracles in the queue must be actively heartbeating onchain to provide updates. Data and configurations for this queue are stored onchain in an account owned by the Switchboard program.
Oracle Consensus - Determines how oracles come to agreement on the accepted onchain result. Switchboard oracles use the median oracle response as the accepted result. A feed authority can control how many oracles are requested and how many must respond to influence its security.

Switchboard oracles are incentivized to update data feeds because they are rewarded for doing so accurately. Each data feed has a LeaseContract account. The lease contract is a pre-funded escrow account to reward oracles for fulfilling update requests. Only the predefined leaseAuthority can withdraw funds from the contract, but anyone can contribute to it. When a new round of updates is requested for a data feed, the user who requested the update is rewarded from the escrow. This is to incentivize users and crank turners (anyone who runs software to systematically send update requests to Oracles) to keep feeds updating based on a feed’s configurations. Once an update request has been successfully fulfilled and submitted onchain by the oracles in the queue, the oracles are transferred rewards from the escrow as well. These payments ensure active participation.

Additionally, oracles have to stake tokens before they can service update requests and submit responses onchain. If an oracle submits a result onchain that falls outside the queue’s configured parameters, their stake will be slashed (if the queue has slashingEnabled). This helps ensure that oracles are responding in good faith with accurate information.

Now that you understand the terminology and economics, let’s take a look at how data is published onchain:

Oracle queue setup - When an update is requested from a queue, the next N oracles are assigned to the update request and cycled to the back of the queue. Each oracle queue in the Switchboard network is independent and maintains its own configuration. The configuration influences its level of security. This design choice enables users to tailor the oracle queue's behavior to match their specific use case. An Oracle queue is stored onchain as an account and contains metadata about the queue. A queue is created by invoking the oracleQueueInit instruction on the Switchboard Solana program.
1. Some relevant Oracle Queue configurations:
  1. oracle_timeout - Interval when stale oracles will be removed if they fail to heartbeat.
  2. reward - Rewards to provide oracles and round openers on this queue.
  3. min_stake - The minimum amount of stake that oracles must provide to remain on the queue.
  4. size - The current number of oracles on a queue.
  5. max_size - The maximum number of oracles a queue can support.
Aggregator/data feed setup - The aggregator/feed account gets created. A feed belongs to a single oracle queue. The feed’s configuration dictates how update requests are invoked and routed through the network.
Job account setup - In addition to the feed, a job account for each data source must be set up. This defines how oracles can fulfill the feed’s update requests. This includes defining where the oracles should fetch the data the feed is requesting.
Request assignment - Once an update has been requested with the feed account, the oracle queue assigns the request to different oracles/nodes in the queue to fulfill. The oracles will fetch the data from the data source defined in each of the feed’s job accounts. Each job account has a weight associated with it. The oracle will calculate the weighted median of the results from across all the jobs.
After minOracleResults responses are received, the onchain program calculates the result using the median of the oracle responses. Oracles who respond within the queue’s configured parameters are rewarded, while the oracles who respond outside this threshold are slashed (if the queue has slashingEnabled).
The updated result is stored in the data feed account so it can be read/consumed onchain.

How to use Switchboard Oracles

To use Switchboard oracles and incorporate off-chain data into a Solana program, you first have to find a feed that provides the data you need. Switchboard feeds are public and there are many already available that you can choose from. When looking for a feed, you have to decide how accurate/reliable you want the feed, where you want to source the data from, as well as the feed’s update cadence. When consuming a publicly available feed, you have no control over these things, so choose carefully!

For example, there is a Switchboard-sponsored BTC_USD feed. This feed is available on Solana devnet/mainnet with pubkey 8SXvChNYFhRq4EZuZvnhjrB3jJRQCv4k3P4W6hesH3Ee. It provides the current price of Bitcoin in USD onchain.

The actual onchain data for a Switchboard feed account looks a little like this:

// from the switchboard solana program
// https://github.com/switchboard-xyz/sbv2-solana/blob/0b5e0911a1851f9ca37042e6ff88db4cd840067b/rust/switchboard-solana/src/oracle_program/accounts/aggregator.rs#L60

pub struct AggregatorAccountData {
    /// Name of the aggregator to store onchain.
    pub name: [u8; 32],
    ...
		...
    /// Pubkey of the queue the aggregator belongs to.
    pub queue_pubkey: Pubkey,
    ...
    /// Minimum number of oracle responses required before a round is validated.
    pub min_oracle_results: u32,
    /// Minimum number of job results before an oracle accepts a result.
    pub min_job_results: u32,
    /// Minimum number of seconds required between aggregator rounds.
    pub min_update_delay_seconds: u32,
    ...
    /// Change percentage required between a previous round and the current round. If variance percentage is not met, reject new oracle responses.
    pub variance_threshold: SwitchboardDecimal,
    ...
		/// Latest confirmed update request result that has been accepted as valid. This is where you will find the data you are requesting in latest_confirmed_round.result
	  pub latest_confirmed_round: AggregatorRound,
		...
    /// The previous confirmed round result.
    pub previous_confirmed_round_result: SwitchboardDecimal,
    /// The slot when the previous confirmed round was opened.
    pub previous_confirmed_round_slot: u64,
		...
}

You can view the full code for this data structure in the Switchboard program here.

Some relevant fields and configurations on the AggregatorAccountData type are:

min_oracle_results - Minimum number of oracle responses required before a round is validated.
min_job_results - Minimum number of job results before an oracle accepts a result.
variance_threshold - Change percentage required between a previous round and the current round. If variance percentage is not met, reject new oracle responses.
latest_confirmed_round - Latest confirmed update request result that has been accepted as valid. This is where you will find the data of the feed in latest_confirmed_round.result
min_update_delay_seconds - Minimum number of seconds required between aggregator rounds.

The first three configs listed above are directly related to the accuracy and reliability of a data feed.

The min_job_results field represents the minimum amount of successful responses from data sources an oracle must receive before it can submit its response onchain. Meaning if min_job_results is three, each oracle has to pull from three job sources. The higher this number, the more reliable and accurate the data on the feed will be. This also limits the impact that a single data source can have on the result.

The min_oracle_results field is the minimum amount of oracle responses required for a round to be successful. Remember, each oracle in a queue pulls data from each source defined as a job. The oracle then takes the weighted median of the responses from the sources and submits that median onchain. The program then waits for min_oracle_results of weighted medians and takes the median of that, which is the final result stored in the data feed account.

The min_update_delay_seconds field is directly related to a feed’s update cadence. min_update_delay_seconds must have passed between one round of updates and the next one before the Switchboard program will accept results.

It can help to look at the jobs tab of a feed in Switchboard's explorer. For example, you can look at the BTC_USD feed in the explorer. Each job listed defines the source the oracles will fetch data from and the weighting of each source. You can view the actual API endpoints that provide the data for this specific feed. When determining what data feed to use in your program, things like this are very important to consider.

Below is a two of the jobs related to the BTC_USD feed. It shows two sources of data: MEXC and Coinbase.

Oracle Jobs

Once you’ve chosen a feed to use, you can start reading the data in that feed. You do this by simply deserializing and reading the state stored in the account. The easiest way to do that is by making use of the AggregatorAccountData struct we defined above from the switchboard_v2 crate in your program.

// import anchor and switchboard crates
use {
    anchor_lang::prelude::*,
    switchboard_v2::AggregatorAccountData,
};

...

#[derive(Accounts)]
pub struct ConsumeDataAccounts<'info> {
	// pass in data feed account and deserialize to AggregatorAccountData
	pub feed_aggregator: AccountLoader<'info, AggregatorAccountData>,
	...
}

Notice that we use the AccountLoader type here instead of the normal Account type to deserialize the aggregator account. Due to the size of AggregatorAccountData, the account uses what's called zero copy. This in combination with AccountLoader prevents the account from being loaded into memory and gives our program direct access to the data instead. When using AccountLoader we can access the data stored in the account in one of three ways:

load_init after initializing an account (this will ignore the missing account discriminator that gets added only after the user’s instruction code)
load when the account is not mutable
load_mut when the account is mutable

If you’d like to learn more, check out the Advance Program Architecture lesson where we touch on Zero-Copy and AccountLoader.

With the aggregator account passed into your program, you can use it to get the latest oracle result. Specifically, you can use the type's get_result() method:

// inside an Anchor program
...

let feed = &ctx.accounts.feed_aggregator.load()?;
// get result
let val: f64 = feed.get_result()?.try_into()?;

The get_result() method defined on the AggregatorAccountData struct is safer than fetching the data with latest_confirmed_round.result because Switchboard has implemented some nifty safety checks.

// from switchboard program
// https://github.com/switchboard-xyz/sbv2-solana/blob/0b5e0911a1851f9ca37042e6ff88db4cd840067b/rust/switchboard-solana/src/oracle_program/accounts/aggregator.rs#L195

pub fn get_result(&self) -> anchor_lang::Result<SwitchboardDecimal> {
    if self.resolution_mode == AggregatorResolutionMode::ModeSlidingResolution {
        return Ok(self.latest_confirmed_round.result);
    }
    let min_oracle_results = self.min_oracle_results;
    let latest_confirmed_round_num_success = self.latest_confirmed_round.num_success;
    if min_oracle_results > latest_confirmed_round_num_success {
        return Err(SwitchboardError::InvalidAggregatorRound.into());
    }
    Ok(self.latest_confirmed_round.result)
}

You can also view the current value stored in an AggregatorAccountData account client-side in Typescript.

import { AggregatorAccount, SwitchboardProgram} from '@switchboard-xyz/solana.js'

...
...
// create keypair for test user
let user = new anchor.web3.Keypair()

// fetch switchboard devnet program object
switchboardProgram = await SwitchboardProgram.load(
  "devnet",
  new anchor.web3.Connection("https://api.devnet.solana.com"),
  user
)

// pass switchboard program object and feed pubkey into AggregatorAccount constructor
aggregatorAccount = new AggregatorAccount(switchboardProgram, solUsedSwitchboardFeed)

// fetch latest SOL price
const solPrice: Big | null = await aggregatorAccount.fetchLatestValue()
if (solPrice === null) {
  throw new Error('Aggregator holds no value')
}

Remember, Switchboard data feeds are just accounts that are updated by third parties (oracles). Given that, you can do anything with the account that you can typically do with accounts external to your program.

Best Practices and Common Pitfalls

When incorporating Switchboard feeds into your programs, there are two groups of concerns to consider: choosing a feed and actually consuming the data from that feed.

Always audit the configurations of a feed before deciding to incorporate it into a program. Configurations like Min Update Delay, Min job Results, and Min Oracle Results can directly affect the data that is eventually persisted onchain to the aggregator account. For example, looking at the config section of the BTC_USD feed you can see its relevant configurations.

Oracle Configs

The BTC_USD feed has Min Update Delay = 6 seconds. This means that the price of BTC is only updated at a minimum of every 6 seconds on this feed. This is probably fine for most use cases, but if you wanted to use this feed for something latency sensitive, it’s probably not a good choice.

It’s also worthwhile to audit a feed's sources in the Jobs section of the oracle explorer. Since the value that is persisted onchain is the weighted median result the oracles pull from each source, the sources directly influence what is stored in the feed. Check for shady links and potentially run the APIs yourself for a time to gain confidence in them.

Once you have found a feed that fits your needs, you still need to make sure you're using the feed appropriately. For example, you should still implement necessary security checks on the account passed into your instruction. Any account can be passed into your program's instructions, so you should verify it’s the account you expect it to be.

In Anchor, if you deserialize the account to the AggregatorAccountData type from the switchboard_v2 crate, Anchor checks that the account is owned by the Switchboard program. If your program expects that only a specific data feed will be passed in the instruction, then you can also verify that the public key of the account passed in matches what it should be. One way to do this is to hard code the address in the program somewhere and use account constraints to verify the address passed in matches what is expected.

use {
  anchor_lang::prelude::*,
  solana_program::{pubkey, pubkey::Pubkey},
	switchboard_v2::{AggregatorAccountData},
};

pub static BTC_USDC_FEED: Pubkey = pubkey!("8SXvChNYFhRq4EZuZvnhjrB3jJRQCv4k3P4W6hesH3Ee");

...
...

#[derive(Accounts)]
pub struct TestInstruction<'info> {
	// Switchboard SOL feed aggregator
	#[account(
	    address = BTC_USDC_FEED
	)]
	pub feed_aggregator: AccountLoader<'info, AggregatorAccountData>,
}

On top of ensuring the feed account is the one you expect, you can also do some checks on the data stored in the feed in your program's instruction logic. Two common things to check for are data staleness and the confidence interval.

Each data feed updates the current value stored in it when triggered by the oracles. This means the updates are dependent on the oracles in the queue that it’s assigned to. Depending on what you intend to use the data feed for, it may be beneficial to verify that the value stored in the account was updated recently. For example, a lending protocol that needs to determine if a loan’s collateral has fallen below a certain level may need the data to be no older than a few seconds. You can have your code check the timestamp of the most recent update stored in the aggregator account. The following code snippet checks that the timestamp of the most recent update on the data feed was no more than 30 seconds ago.

use {
    anchor_lang::prelude::*,
    anchor_lang::solana_program::clock,
    switchboard_v2::{AggregatorAccountData, SwitchboardDecimal},
};

...
...

let feed = &ctx.accounts.feed_aggregator.load()?;
if (clock::Clock::get().unwrap().unix_timestamp - feed.latest_confirmed_round.round_open_timestamp) <= 30{
      valid_transfer = true;
  }

The latest_confirmed_round field on the AggregatorAccountData struct is of type AggregatorRound defined as:

// https://github.com/switchboard-xyz/sbv2-solana/blob/0b5e0911a1851f9ca37042e6ff88db4cd840067b/rust/switchboard-solana/src/oracle_program/accounts/aggregator.rs#L17

pub struct AggregatorRound {
    /// Maintains the number of successful responses received from nodes.
    /// Nodes can submit one successful response per round.
    pub num_success: u32,
    /// Number of error responses.
    pub num_error: u32,
    /// Whether an update request round has ended.
    pub is_closed: bool,
    /// Maintains the `solana_program::clock::Slot` that the round was opened at.
    pub round_open_slot: u64,
    /// Maintains the `solana_program::clock::UnixTimestamp;` the round was opened at.
    pub round_open_timestamp: i64,
    /// Maintains the current median of all successful round responses.
    pub result: SwitchboardDecimal,
    /// Standard deviation of the accepted results in the round.
    pub std_deviation: SwitchboardDecimal,
    /// Maintains the minimum node response this round.
    pub min_response: SwitchboardDecimal,
    /// Maintains the maximum node response this round.
    pub max_response: SwitchboardDecimal,
    /// Pubkeys of the oracles fulfilling this round.
    pub oracle_pubkeys_data: [Pubkey; 16],
    /// Represents all successful node responses this round. `NaN` if empty.
    pub medians_data: [SwitchboardDecimal; 16],
    /// Current rewards/slashes oracles have received this round.
    pub current_payout: [i64; 16],
    /// Keep track of which responses are fulfilled here.
    pub medians_fulfilled: [bool; 16],
    /// Keeps track of which errors are fulfilled here.
    pub errors_fulfilled: [bool; 16],
}

There are some other relevant fields that may be of interest to you in the Aggregator account like num_success, medians_data, std_deviation, etc. num_success is the number of successful responses received from oracles in this round of updates. medians_data is an array of all of the successful responses received from oracles this round. This is the dataset that is used to derive the median and the final result. std_deviation is the standard deviation of the accepted results in this round. You might want to check for a low standard deviation, meaning that all of the oracle responses were similar. The switchboard program is in charge of updating the relevant fields on this struct every time it receives an update from an oracle.

The AggregatorAccountData also has a check_confidence_interval() method that you can use as another verification on the data stored in the feed. The method allows you to pass in a max_confidence_interval. If the standard deviation of the results received from the oracle is greater than the given max_confidence_interval, it returns an error.

// https://github.com/switchboard-xyz/sbv2-solana/blob/0b5e0911a1851f9ca37042e6ff88db4cd840067b/rust/switchboard-solana/src/oracle_program/accounts/aggregator.rs#L228

pub fn check_confidence_interval(
    &self,
    max_confidence_interval: SwitchboardDecimal,
) -> anchor_lang::Result<()> {
    if self.latest_confirmed_round.std_deviation > max_confidence_interval {
        return Err(SwitchboardError::ConfidenceIntervalExceeded.into());
    }
    Ok(())
}

You can incorporate this into your program like so:

use {
    crate::{errors::*},
    anchor_lang::prelude::*,
    std::convert::TryInto,
    switchboard_v2::{AggregatorAccountData, SwitchboardDecimal},
};

...
...

let feed = &ctx.accounts.feed_aggregator.load()?;

// check feed does not exceed max_confidence_interval
feed.check_confidence_interval(SwitchboardDecimal::from_f64(max_confidence_interval))
    .map_err(|_| error!(ErrorCode::ConfidenceIntervalExceeded))?;

Lastly, it’s important to plan for worst-case scenarios in your programs. Plan for feeds going stale and plan for feed accounts closing.

Conclusion

If you want functional programs that can perform actions based on real-world data, you’re going to have to use oracles. Fortunately, there are some trustworthy oracle networks, like Switchboard, that make using oracles easier than they would otherwise be. However, make sure to do your due diligence on the oracles you use. You are ultimately responsible for your program's behavior!

Lab

Let's practice using oracles! We'll be building a "Michael Burry Escrow" program that locks SOL in an escrow account until SOL is above a certain USD value. This is named after the investor Michael Burry who's famous for predicting the 2008 housing market crash.

We will be using the devnet SOL_USD oracle from switchboard. The program will have two main instructions:

Deposit - Lock up the SOL and set a USD price to unlock it at.
Withdraw - Check the USD price and withdraw the SOL if the price is met.

1. Program Setup

To get started, let’s create the program with

anchor init burry-escrow

Next, replace the program ID in lib.rs and Anchor.toml with the program ID shown when you run anchor keys list.

Next, add the following to the bottom of your Anchor.toml file. This will tell Anchor how to configure our local testing environment. This will allow us to test our program locally without having to deploy and send transactions to devnet.

At the bottom of Anchor.toml:

[test.validator]
url="https://api.devnet.solana.com"

[test]
startup_wait = 10000

[[test.validator.clone]] # sbv2 devnet programID
address = "SW1TCH7qEPTdLsDHRgPuMQjbQxKdH2aBStViMFnt64f"

[[test.validator.clone]] # sbv2 devnet IDL
address = "Fi8vncGpNKbq62gPo56G4toCehWNy77GgqGkTaAF5Lkk"

[[test.validator.clone]] # sbv2 SOL/USD Feed
address="GvDMxPzN1sCj7L26YDK2HnMRXEQmQ2aemov8YBtPS7vR"

Additionally, we want to import the switchboard-v2 crate in our Cargo.toml file. Make sure your dependencies look as follows:

[dependencies]
anchor-lang = "0.28.0"
switchboard-v2 = "0.4.0"

Before we get started with the logic, let’s go over the structure of our program. With small programs, it’s very easy to add all of the smart contract code to a single lib.rs file and call it a day. To keep it more organized though, it’s helpful to break it up across different files. Our program will have the following files within the programs/src directory:

/instructions/deposit.rs

/instructions/withdraw.rs

/instructions/mod.rs

errors.rs

state.rs

lib.rs

The lib.rs file will still serve as the entry point to our program, but the logic for each instruction will be contained in their own separate file. Go ahead and create the program architecture described above and we’ll get started.

2. `lib.rs`

Before we write any logic, we are going to set up all of our boilerplate information. Starting with lib.rs. Our actual logic will live in the /instructions directory.

The lib.rs file will serve as the entrypoint to our program. It will define the API endpoints that all transactions must go through.

use anchor_lang::prelude::*;
use instructions::deposit::*;
use instructions::withdraw::*;
use state::*;

pub mod instructions;
pub mod state;
pub mod errors;

declare_id!("YOUR_PROGRAM_KEY_HERE");

#[program]
mod burry_oracle_program {

    use super::*;

    pub fn deposit(ctx: Context<Deposit>, escrow_amt: u64, unlock_price: u64) -> Result<()> {
        deposit_handler(ctx, escrow_amt, unlock_price)
    }

    pub fn withdraw(ctx: Context<Withdraw>) -> Result<()> {
        withdraw_handler(ctx)
    }
}

3. `state.rs`

Next, let's define our data account for this program: EscrowState. Our data account will store two pieces of info:

unlock_price - The price of SOL in USD at which point you can withdraw; you can hard-code it to whatever you want (e.g. $21.53)
escrow_amount - Keeps track of how many lamports are stored in the escrow account

We will also be defining our PDA seed of "MICHAEL BURRY" and our hardcoded SOL_USD oracle pubkey SOL_USDC_FEED.

// in state.rs
use anchor_lang::prelude::*;

pub const ESCROW_SEED: &[u8] = b"MICHAEL BURRY";
pub const SOL_USDC_FEED: &str = "GvDMxPzN1sCj7L26YDK2HnMRXEQmQ2aemov8YBtPS7vR";

#[account]
pub struct EscrowState {
    pub unlock_price: f64,
    pub escrow_amount: u64,
}

4. Errors

Let’s define the custom errors we’ll use throughout the program. Inside the errors.rs file, paste the following:

use anchor_lang::prelude::*;

#[error_code]
#[derive(Eq, PartialEq)]
pub enum EscrowErrorCode {
    #[msg("Not a valid Switchboard account")]
    InvalidSwitchboardAccount,
    #[msg("Switchboard feed has not been updated in 5 minutes")]
    StaleFeed,
    #[msg("Switchboard feed exceeded provided confidence interval")]
    ConfidenceIntervalExceeded,
    #[msg("Current SOL price is not above Escrow unlock price.")]
    SolPriceAboveUnlockPrice,
}

5. `mod.rs`

Let's set up our instructions/mod.rs file.

// inside mod.rs
pub mod deposit;
pub mod withdraw;

6. Deposit

Now that we have all of the boilerplate out of the way, lets move onto our Deposit instruction. This will live in the /src/instructions/deposit.rs file. When a user deposits, a PDA should be created with the “MICHAEL BURRY” string and the user’s pubkey as seeds. This inherently means a user can only open one escrow account at a time. The instruction should initialize an account at this PDA and send the amount of SOL that the user wants to lock up to it. The user will need to be a signer.

Let’s build the Deposit Context struct first. To do that, we need to think about what accounts will be necessary for this instruction. We start with the following:

//inside deposit.rs
use crate::state::*;
use anchor_lang::prelude::*;
use anchor_lang::solana_program::{
    system_instruction::transfer,
    program::invoke
};

#[derive(Accounts)]
pub struct Deposit<'info> {
    // user account
    #[account(mut)]
    pub user: Signer<'info>,
    #[account(
      init,
      seeds = [ESCROW_SEED, user.key().as_ref()],
      bump,
      payer = user,
      space = std::mem::size_of::<EscrowState>() + 8
    )]
    pub escrow_account: Account<'info, EscrowState>,
		// system program
    pub system_program: Program<'info, System>,
}

Notice the constraints we added to the accounts:

Because we'll be transferring SOL from the User account to the escrow_state account, they both need to be mutable.
We know the escrow_account is supposed to be a PDA derived with the “MICHAEL BURRY” string and the user’s pubkey. We can use Anchor account constraints to guarantee that the address passed in actually meets that requirement.
We also know that we have to initialize an account at this PDA to store some state for the program. We use the init constraint here.

Let’s move onto the actual logic. All we need to do is to initialize the state of the escrow_state account and transfer the SOL. We expect the user to pass in the amount of SOL they want to lock up in escrow and the price to unlock it at. We will store these values in the escrow_state account.

After that, the method should execute the transfer. This program will be locking up native SOL. Because of this, we don’t need to use token accounts or the Solana token program. We’ll have to use the system_program to transfer the lamports the user wants to lock up in escrow and invoke the transfer instruction.

pub fn deposit_handler(ctx: Context<Deposit>, escrow_amt: u64, unlock_price: u64) -> Result<()> {
		msg!("Depositing funds in escrow...");

    let escrow_state = &mut ctx.accounts.escrow_account;
    escrow_state.unlock_price = unlock_price;
    escrow_state.escrow_amount = escrow_amount;

    let transfer_ix = transfer(
      &ctx.accounts.user.key(),
      &escrow_state.key(),
      escrow_amount
    );

    invoke(
        &transfer_ix,
        &[
            ctx.accounts.user.to_account_info(),
            ctx.accounts.escrow_account.to_account_info(),
            ctx.accounts.system_program.to_account_info()
        ]
    )?;

    msg!("Transfer complete. Escrow will unlock SOL at {}", &ctx.accounts.escrow_account.unlock_price);
}

That’s is the gist of the deposit instruction! The final result of the deposit.rs file should look as follows:

use crate::state::*;
use anchor_lang::prelude::*;
use anchor_lang::solana_program::{
    system_instruction::transfer,
    program::invoke
};

pub fn deposit_handler(ctx: Context<Deposit>, escrow_amount: u64, unlock_price: f64) -> Result<()> {
    msg!("Depositing funds in escrow...");

    let escrow_state = &mut ctx.accounts.escrow_account;
    escrow_state.unlock_price = unlock_price;
    escrow_state.escrow_amount = escrow_amount;

    let transfer_ix = transfer(
        &ctx.accounts.user.key(),
        &escrow_state.key(),
        escrow_amount
    );

    invoke(
        &transfer_ix,
        &[
            ctx.accounts.user.to_account_info(),
            ctx.accounts.escrow_account.to_account_info(),
            ctx.accounts.system_program.to_account_info()
        ]
    )?;

    msg!("Transfer complete. Escrow will unlock SOL at {}", &ctx.accounts.escrow_account.unlock_price);

    Ok(())
}

#[derive(Accounts)]
pub struct Deposit<'info> {
    // user account
    #[account(mut)]
    pub user: Signer<'info>,
    // account to store SOL in escrow
    #[account(
        init,
        seeds = [ESCROW_SEED, user.key().as_ref()],
        bump,
        payer = user,
        space = std::mem::size_of::<EscrowState>() + 8
    )]
    pub escrow_account: Account<'info, EscrowState>,

    pub system_program: Program<'info, System>,
}

Withdraw

The withdraw instruction will require the same three accounts as the deposit instruction plus the SOL_USDC Switchboard feed account. This code will go in the withdraw.rs file.

use crate::state::*;
use crate::errors::*;
use std::str::FromStr;
use anchor_lang::prelude::*;
use switchboard_v2::AggregatorAccountData;
use anchor_lang::solana_program::clock::Clock;

#[derive(Accounts)]
pub struct Withdraw<'info> {
    // user account
    #[account(mut)]
    pub user: Signer<'info>,
    // escrow account
    #[account(
        mut,
        seeds = [ESCROW_SEED, user.key().as_ref()],
        bump,
        close = user
    )]
    pub escrow_account: Account<'info, EscrowState>,
    // Switchboard SOL feed aggregator
    #[account(
        address = Pubkey::from_str(SOL_USDC_FEED).unwrap()
    )]
    pub feed_aggregator: AccountLoader<'info, AggregatorAccountData>,
    pub system_program: Program<'info, System>,
}

Notice we’re using the close constraint because once the transaction completes, we want to close the escrow_account. The SOL used as rent in the account will be transferred to the user account.

We also use the address constraints to verify that the feed account passed in is actually the usdc_sol feed and not some other feed (we have the SOL_USDC_FEED address hard coded). In addition, the AggregatorAccountData struct that we deserialize comes from the Switchboard rust crate. It verifies that the given account is owned by the switchboard program and allows us to easily look at its values. You’ll notice it’s wrapped in a AccountLoader. This is because the feed is actually a fairly large account and it needs to be zero copied.

Now let's implement the withdraw instruction's logic. First, we check if the feed is stale. Then we fetch the current price of SOL stored in the feed_aggregator account. Lastly, we want to check that the current price is above the escrow unlock_price. If it is, then we transfer the SOL from the escrow account back to the user and close the account. If it isn’t, then the instruction should finish and return an error.

pub fn withdraw_handler(ctx: Context<Withdraw>, params: WithdrawParams) -> Result<()> {
    let feed = &ctx.accounts.feed_aggregator.load()?;
    let escrow_state = &ctx.accounts.escrow_account;

    // get result
    let val: f64 = feed.get_result()?.try_into()?;

    // check whether the feed has been updated in the last 300 seconds
    feed.check_staleness(Clock::get().unwrap().unix_timestamp, 300)
    .map_err(|_| error!(EscrowErrorCode::StaleFeed))?;

    msg!("Current feed result is {}!", val);
    msg!("Unlock price is {}", escrow_state.unlock_price);

    if val < escrow_state.unlock_price as f64 {
        return Err(EscrowErrorCode::SolPriceAboveUnlockPrice.into())
    }

	....
}

To finish the logic off, we will execute the transfer, this time we will have to transfer the funds in a different way. Because we are transferring from an account that also holds data we cannot use the system_program::transfer method like before. If we try to, the instruction will fail to execute with the following error.

'Transfer: `from` must not carry data'

To account for this, we’ll use try_borrow_mut_lamports() on each account and add/subtract the amount of lamports stored in each account.

// 'Transfer: `from` must not carry data'
  **escrow_state.to_account_info().try_borrow_mut_lamports()? = escrow_state
      .to_account_info()
      .lamports()
      .checked_sub(escrow_state.escrow_amount)
      .ok_or(ProgramError::InvalidArgument)?;

  **ctx.accounts.user.to_account_info().try_borrow_mut_lamports()? = ctx.accounts.user
      .to_account_info()
      .lamports()
      .checked_add(escrow_state.escrow_amount)
      .ok_or(ProgramError::InvalidArgument)?;

The final withdraw method in the withdraw.rs file should look like this:

use crate::state::*;
use crate::errors::*;
use std::str::FromStr;
use anchor_lang::prelude::*;
use switchboard_v2::AggregatorAccountData;
use anchor_lang::solana_program::clock::Clock;

pub fn withdraw_handler(ctx: Context<Withdraw>) -> Result<()> {
    let feed = &ctx.accounts.feed_aggregator.load()?;
    let escrow_state = &ctx.accounts.escrow_account;

    // get result
    let val: f64 = feed.get_result()?.try_into()?;

    // check whether the feed has been updated in the last 300 seconds
    feed.check_staleness(Clock::get().unwrap().unix_timestamp, 300)
    .map_err(|_| error!(EscrowErrorCode::StaleFeed))?;

    msg!("Current feed result is {}!", val);
    msg!("Unlock price is {}", escrow_state.unlock_price);

    if val < escrow_state.unlock_price as f64 {
        return Err(EscrowErrorCode::SolPriceAboveUnlockPrice.into())
    }

    // 'Transfer: `from` must not carry data'
    **escrow_state.to_account_info().try_borrow_mut_lamports()? = escrow_state
        .to_account_info()
        .lamports()
        .checked_sub(escrow_state.escrow_amount)
        .ok_or(ProgramError::InvalidArgument)?;

    **ctx.accounts.user.to_account_info().try_borrow_mut_lamports()? = ctx.accounts.user
        .to_account_info()
        .lamports()
        .checked_add(escrow_state.escrow_amount)
        .ok_or(ProgramError::InvalidArgument)?;

    Ok(())
}

#[derive(Accounts)]
pub struct Withdraw<'info> {
    // user account
    #[account(mut)]
    pub user: Signer<'info>,
    // escrow account
    #[account(
        mut,
        seeds = [ESCROW_SEED, user.key().as_ref()],
        bump,
        close = user
    )]
    pub escrow_account: Account<'info, EscrowState>,
    // Switchboard SOL feed aggregator
    #[account(
        address = Pubkey::from_str(SOL_USDC_FEED).unwrap()
    )]
    pub feed_aggregator: AccountLoader<'info, AggregatorAccountData>,
    pub system_program: Program<'info, System>,
}

And that’s it for the program! At this point, you should be able to run anchor build without any errors.

If you see an error like the one presented below, you can safely ignore it.

Compiling switchboard-v2 v0.4.0
Error: Function _ZN86_$LT$switchboard_v2..aggregator..AggregatorAccountData$u20$as$u20$core..fmt..Debug$GT$3fmt17hea9f7644392c2647E Stack offset of 4128 exceeded max offset of 4096 by 32 bytes, please minimize large stack variables

7. Testing

Let's write some tests. We should have four of them:

Creating an Escrow with the unlock price below the current SOL price so we can test withdrawing it
Withdrawing and closing from the above escrow
Creating an Escrow with the unlock price above the current SOL price so we can test withdrawing it
Withdrawing and failing from the above escrow

Note that there can only be one escrow per user, so the above order matters.

We'll provide all the testing code in one snippet. Take a look through to make sure you understand it before running anchor test.

// tests/burry-escrow.ts

import * as anchor from "@coral-xyz/anchor";
import { Program } from "@coral-xyz/anchor";
import { BurryEscrow } from "../target/types/burry_escrow";
import { Big } from "@switchboard-xyz/common";
import {
  AggregatorAccount,
  AnchorWallet,
  SwitchboardProgram,
} from "@switchboard-xyz/solana.js";
import { assert } from "chai";

export const solUsedSwitchboardFeed = new anchor.web3.PublicKey(
  "GvDMxPzN1sCj7L26YDK2HnMRXEQmQ2aemov8YBtPS7vR",
);

describe("burry-escrow", () => {
  // Configure the client to use the local cluster.
  anchor.setProvider(anchor.AnchorProvider.env());
  const provider = anchor.AnchorProvider.env();
  const program = anchor.workspace.BurryEscrow as Program<BurryEscrow>;
  const payer = (provider.wallet as AnchorWallet).payer;

  it("Create Burry Escrow Below Price", async () => {
    // fetch switchboard devnet program object
    const switchboardProgram = await SwitchboardProgram.load(
      "devnet",
      new anchor.web3.Connection("https://api.devnet.solana.com"),
      payer,
    );
    const aggregatorAccount = new AggregatorAccount(
      switchboardProgram,
      solUsedSwitchboardFeed,
    );

    // derive escrow state account
    const [escrowState] = await anchor.web3.PublicKey.findProgramAddressSync(
      [Buffer.from("MICHAEL BURRY"), payer.publicKey.toBuffer()],
      program.programId,
    );

    // fetch latest SOL price
    const solPrice: Big | null = await aggregatorAccount.fetchLatestValue();
    if (solPrice === null) {
      throw new Error("Aggregator holds no value");
    }
    const failUnlockPrice = solPrice.minus(10).toNumber();
    const amountToLockUp = new anchor.BN(100);

    // Send transaction
    try {
      const tx = await program.methods
        .deposit(amountToLockUp, failUnlockPrice)
        .accounts({
          user: payer.publicKey,
          escrowAccount: escrowState,
          systemProgram: anchor.web3.SystemProgram.programId,
        })
        .signers([payer])
        .rpc();

      await provider.connection.confirmTransaction(tx, "confirmed");

      // Fetch the created account
      const newAccount = await program.account.escrowState.fetch(escrowState);

      const escrowBalance = await provider.connection.getBalance(
        escrowState,
        "confirmed",
      );
      console.log("Onchain unlock price:", newAccount.unlockPrice);
      console.log("Amount in escrow:", escrowBalance);

      // Check whether the data onchain is equal to local 'data'
      assert(failUnlockPrice == newAccount.unlockPrice);
      assert(escrowBalance > 0);
    } catch (e) {
      console.log(e);
      assert.fail(e);
    }
  });

  it("Withdraw from escrow", async () => {
    // derive escrow address
    const [escrowState] = await anchor.web3.PublicKey.findProgramAddressSync(
      [Buffer.from("MICHAEL BURRY"), payer.publicKey.toBuffer()],
      program.programId,
    );

    // send tx
    const tx = await program.methods
      .withdraw()
      .accounts({
        user: payer.publicKey,
        escrowAccount: escrowState,
        feedAggregator: solUsedSwitchboardFeed,
        systemProgram: anchor.web3.SystemProgram.programId,
      })
      .signers([payer])
      .rpc();

    await provider.connection.confirmTransaction(tx, "confirmed");

    // assert that the escrow account has been closed
    let accountFetchDidFail = false;
    try {
      await program.account.escrowState.fetch(escrowState);
    } catch (e) {
      accountFetchDidFail = true;
    }

    assert(accountFetchDidFail);
  });

  it("Create Burry Escrow Above Price", async () => {
    // fetch switchboard devnet program object
    const switchboardProgram = await SwitchboardProgram.load(
      "devnet",
      new anchor.web3.Connection("https://api.devnet.solana.com"),
      payer,
    );
    const aggregatorAccount = new AggregatorAccount(
      switchboardProgram,
      solUsedSwitchboardFeed,
    );

    // derive escrow state account
    const [escrowState] = await anchor.web3.PublicKey.findProgramAddressSync(
      [Buffer.from("MICHAEL BURRY"), payer.publicKey.toBuffer()],
      program.programId,
    );
    console.log("Escrow Account: ", escrowState.toBase58());

    // fetch latest SOL price
    const solPrice: Big | null = await aggregatorAccount.fetchLatestValue();
    if (solPrice === null) {
      throw new Error("Aggregator holds no value");
    }
    const failUnlockPrice = solPrice.plus(10).toNumber();
    const amountToLockUp = new anchor.BN(100);

    // Send transaction
    try {
      const tx = await program.methods
        .deposit(amountToLockUp, failUnlockPrice)
        .accounts({
          user: payer.publicKey,
          escrowAccount: escrowState,
          systemProgram: anchor.web3.SystemProgram.programId,
        })
        .signers([payer])
        .rpc();

      await provider.connection.confirmTransaction(tx, "confirmed");
      console.log("Your transaction signature", tx);

      // Fetch the created account
      const newAccount = await program.account.escrowState.fetch(escrowState);

      const escrowBalance = await provider.connection.getBalance(
        escrowState,
        "confirmed",
      );
      console.log("Onchain unlock price:", newAccount.unlockPrice);
      console.log("Amount in escrow:", escrowBalance);

      // Check whether the data onchain is equal to local 'data'
      assert(failUnlockPrice == newAccount.unlockPrice);
      assert(escrowBalance > 0);
    } catch (e) {
      console.log(e);
      assert.fail(e);
    }
  });

  it("Attempt to withdraw while price is below UnlockPrice", async () => {
    let didFail = false;

    // derive escrow address
    const [escrowState] = await anchor.web3.PublicKey.findProgramAddressSync(
      [Buffer.from("MICHAEL BURRY"), payer.publicKey.toBuffer()],
      program.programId,
    );

    // send tx
    try {
      const tx = await program.methods
        .withdraw()
        .accounts({
          user: payer.publicKey,
          escrowAccount: escrowState,
          feedAggregator: solUsedSwitchboardFeed,
          systemProgram: anchor.web3.SystemProgram.programId,
        })
        .signers([payer])
        .rpc();

      await provider.connection.confirmTransaction(tx, "confirmed");
      console.log("Your transaction signature", tx);
    } catch (e) {
      // verify tx returns expected error
      didFail = true;
      console.log(e.error.errorMessage);
      assert(
        e.error.errorMessage ==
          "Current SOL price is not above Escrow unlock price.",
      );
    }

    assert(didFail);
  });
});

If you feel confident in the testing logic, go ahead and run anchor test in your shell of choice. You should get four passing tests.

If something went wrong, go back through the lab and make sure you got everything right. Pay close attention to the intent behind the code rather than just copy/pasting. Also feel free to review the working code on the main branch of its Github repository.

Challenge

As an independent challenge, create a fallback plan if the data feed ever goes down. If the Oracle queue has not updated the aggregator account in X time or if the data feed account does not exist anymore, withdraw the user’s escrowed funds.

A potential solution to this challenge can be found in the Github repository on the challenge-solution branch.

Push your code to GitHub and [tell us what you thought of this lesson](https://form.typeform.com/to/IPH0UGz7#answers-lesson=1a5d266c-f4c1-4c45-b986-2afd4be59991)!

Summary​

Lesson​

Trust and Oracle Networks​

Oracles on Solana​

Switchboard Oracles​

How to use Switchboard Oracles​

Best Practices and Common Pitfalls​

Conclusion​

Lab​

1. Program Setup​

2. lib.rs​

3. state.rs​

4. Errors​

5. mod.rs​

6. Deposit​

7. Testing​

Challenge​