Codex Marketplace and its interactions are defined by a smart contract deployed on an EVM-compatible blockchain. This specification describes these interactions for the various roles within the network.
The document is intended for implementors of Codex nodes.
The keywords “MUST”, “MUST NOT”, “REQUIRED”, “SHALL”, “SHALL NOT”, “SHOULD”, “SHOULD NOT”, “RECOMMENDED”, “MAY”, and “OPTIONAL” in this document are to be interpreted as described in[2119](https://www.ietf.org/rfc/rfc2119.txt).
The Codex network aims to create a peer-to-peer storage engine with robust data durability, data persistence guarantees, and a comprehensive incentive structure.
The marketplace is a critical component of the Codex network, serving as a platform where all involved parties interact to ensure data persistence. It also provides mechanisms to enforce agreements and facilitate data repair when Storage Nodes fail to fulfill their duties.
Implemented as a smart contract on an EVM-compatible blockchain, the marketplace enables various scenarios where nodes assume one or more roles to maintain a reliable persistence layer for users. This specification details these interactions.
The marketplace contract manages storage requests, maintains the state of allocated storage slots, and orchestrates SP rewards, collaterals, and storage proofs.
An SP is a long-lived node providing storage for clients in exchange for profit. To ensure a reliable, robust service for clients, SPs are required to periodically provide proofs that they are persisting the data.
A validator ensures that SPs have submitted valid proofs each period where the smart contract required a proof to be submitted for slots filled by the SP.
When a user prompts the client node to create a storage request, the client node SHOULD receive the input parameters for the storage request from the user.
To create a request to persist a dataset on the Codex network, client nodes MUST split the dataset into data chunks, $(c_1, c_2, c_3, \ldots, c_{n})$. Using the erasure coding method and the provided input parameters, the data chunks are encoded and distributed over a number of slots. The applied erasure coding method MUST use the [Reed-Soloman algorithm](https://hackmd.io/FB58eZQoTNm-dnhu0Y1XnA). The final slot roots and other metadata MUST be placed into a `Manifest` (TODO: Manifest RFC). The CID for the `Manifest` MUST then be used as the `cid` for the stored dataset.
After the dataset is prepared, a client node MUST call the smart contract function `requestStorage(request)`, providing the desired request parameters in the `request` parameter. The `request` parameter is of type `Request`:
| `content` | `Content` | The dataset that will be hosted with the storage request. |
| `expiry` | `uint256` | Timeout in seconds during which all the slots have to be filled, otherwise Request will get cancelled. The final deadline timestamp is calculated at the moment the transaction is mined. |
| `nonce` | `byte32` | Random value to differentiate from other requests of same parameters. It SHOULD be a random byte array. |
| `reward` | `uint256` | Amount of tokens that will be awarded to SPs for finishing the storage request. It MUST be an amount of Tokens offered per slot per second. The Ethereum address that submits the `requestStorage()` transaction MUST have [approval](https://docs.openzeppelin.com/contracts/2.x/api/token/erc20#IERC20-approve-address-uint256-) for the transfer of at least an equivalent amount in Tokens. |
| `collateral` | `uint256` | The amount of tokens that SPs submit when they fill slots. Collateral is then slashed or forfeited if SPs fail to provide the service requested by the storage request (more information in the [Slashing](#Slashing) section). |
| `proofProbability` | `uint256` | Determines the average frequency that a proof is required within a period: $\frac{1}{proofProbability}$. SPs are required to provide proofs of storage to the marketplace smart contract when challenged by the smart contract. To prevent hosts from only coming online when proofs are required, the frequency at which proofs are requested from SPs is stochastic and is influenced by the `proofProbability` parameter. |
| `duration` | `uint256` | Total duration of the storage request in seconds. |
| `slots` | `uint64` | The number of requested slots. The slots will all have the same size. |
| `slotSize` | `uint256` | Amount of storage per slot in bytes. |
| `maxSlotLoss` | `uint64` | Max slots that can be lost without data considered to be lost. |
| `cid` | `string` | An identifier used to locate the Manifest representing the dataset. It MUST be a [CIDv1](https://github.com/multiformats/cid#cidv1), SHA-256 [multihash](https://github.com/multiformats/multihash) and the data it represents SHOULD be discoverable in the network, otherwise the request will be eventually canceled. |
| `merkleRoot` | `byte32` | Merkle root of the dataset, used to verify storage proofs |
It should be noted that the marketplace does not support extending requests. It is REQUIRED that if the user wants to extend the duration of a request, a new request with the same CID must be [created](#Creating-storage-requests) **before the original request completes**. This ensures that the data will continue to persist in the network at the time when the new (or existing) SPs need to retrieve the complete dataset to fill the slots of the new request.
The client node SHOULD monitor the status of the requests it created. When a storage request enters the `Cancelled` state (this occurs when not all slots have been filled before the `expiry` timeout), the client node SHOULD initiate the withdrawal of the remaining funds from the smart contract using the `withdrawFunds(requestId)` function.
- The request is considered `Cancelled` if no `requestFulfilled(requestId)` event is observed during the timeout specified by the value returned from the `requestExpiresAt(requestId)` function.
- The request is considered `Failed` when the `RequestFailed(requestId)` event is observed.
- The request is considered `Finished` after the interval specified by the value returned from the `getRequestEnd(requestId)` function.
-`ask` - The specification of the request parameters. For details, see the definition of the `Request` type in the _Creating Storage Requests_ section above.
-`expiry` - A Unix timestamp specifying when the request will be canceled if all slots are not filled by then.
It is then up to the SP node to decide, based on the parameters provided by the node operator, whether it wants to participate in the request and attempt to fill its slot(s) (note that one SP can fill more than one slot). If the SP node decides to ignore the request, no further action is required. However, if the SP decides to fill a slot, and succeeds, it MUST follow the remaining steps described below.
The node acting as an SP MUST decide which slot, specified by the slot index, it wants to fill. The SP MAY attempt to fill more than one slot. To fill a slot, the SP MUST first download the slot data using the CID of the manifest (**TODO: Manifest RFC**) and the slot index. The CID is specified in `request.content.cid`, which can be retrieved from the smart contract using `getRequest(requestId)`. Then, the node MUST generate a proof over the downloaded data (**TODO: Proving RFC**).
- The Ethereum address of the node from which the transaction originates MUST have [approval](https://docs.openzeppelin.com/contracts/2.x/api/token/erc20#IERC20-approve-address-uint256-) for the transfer of at least the amount of Tokens required as collateral for the request.
If the proof delivered by the SP is invalid or the slot was already filled by another SP, then the transaction will be reverted. Otherwise, a `SlotFilled(requestId, slotIndex)` event is emitted. If the transaction is successful, the SP SHOULD transition into the __proving__ state, where it will need to submit proof of data possession when prompted by the smart contract.
It should be noted that if the SP node observes a `SlotFilled` event for the slot it is currently downloading the dataset for or generating the proof for, it means that the slot has been filled by another node in the meantime. In response, the SP SHOULD stop its current operation and attempt to fill a different, unfilled slot.
Once an SP successfully fills a slot, it MUST periodically, though non-deterministically, provide proofs to the smart contract that it is storing the data it committed to store. An SP node SHOULD detect whether a proof is required using the `isProofRequired(slotId)` smart contract function, or anticipate that a proof will be required using `willProofBeRequired(slotId)` in case the node is in [downtime](https://github.com/codex-storage/codex-research/blob/41c4b4409d2092d0a5475aca0f28995034e58d14/design/storage-proof-timing.md).
Once the SP knows it must provide a proof, it MUST retrieve the proof challenge using `getChallenge(slotId)`, which then NEEDS to be incorporated into the proof generation as described in the Proving RFC (**TODO: Proving RFC**).
There is a slashing scheme orchestrated by the smart contract to incentivize correct behavior and proper proof submissions by SPs. This scheme is configured at the smart contract level and applies uniformly to all participants in the network. The configuration of the slashing scheme can be obtained via the `getConfig()` contract call.
- It is then slashed by `config.collateral.slashPercentage`**of the originally required collateral** (the slashing amount is always consistent for a given request).
- If the number of slashes exceeds `config.collateral.maxNumberOfSlashes`, the slot is freed, the remaining collateral is burned, and the slot is offered to other nodes for repair. The smart contract also emits the `SlotFreed(requestId, slotIndex)` event.
If, at any time, the number of freed slots exceeds the value specified by the `request.ask.maxSlotLoss` parameter, the dataset is considered lost, and the request is deemed _failed_. The collateral of all SPs that hosted the slots associated with the request is burned, and the `RequestFailed(requestId)` event is emitted.
When a slot is freed due to too many missed proofs, which SHOULD be detected by listening to the `SlotFreed(requestId, slotIndex)` event, an SP node can decide whether to participate in repairing the slot. Similar to filling a slot, the node SHOULD consider the operator's configuration when making this decision. The SP that originally hosted the slot but failed to comply with proving requirements MAY also participate in the repair. However, by refilling the slot, the SP **will not** recover its original collateral and must submit new collateral using the `fillSlot()` call.
The repair process is similar to filling slots. If the original slot dataset is no longer present in the network, the SP MAY use Erasure Coding to reconstruct the dataset. Reconstructing the original slot dataset requires retrieving other pieces of the dataset stored in other slots belonging to the request. For this reason, the node that successfully repairs a slot is entitled to an additional reward. (**TODO: Implementation**)
2. The SP MUST download the chunks of data required to reconstruct the freed slot's data. The node MUST use the [Reed-Solomon algorithm](https://hackmd.io/FB58eZQoTNm-dnhu0Y1XnA) to reconstruct the missing data.
3. The SP MUST generate proof over the reconstructed data.
4. The SP MUST call the `fillSlot()` smart contract function with the same parameters and collateral allowance as described in the [Filling Slots](#filling-slot) section.
- A storage request is considered `Cancelled` if no `RequestFulfilled(requestId)` event is observed within the time indicated by the `expiry` request parameter. Note that a `RequestCancelled` event may also be emitted, but the node SHOULD NOT rely on this event to assert the request expiration, as the `RequestCancelled` event is not guaranteed to be emitted at the time of expiry.
- A storage request is considered `Finished` when the time indicated by the value returned from the `getRequestEnd(requestId)` function has elapsed.
- A node concludes that a storage request has `Failed` upon observing the `RequestFailed(requestId)` event.
- In the `Cancelled` state, the collateral is returned along with a proportional payout based on the time the node actually hosted the dataset before the expiry was reached.
- In the `Finished` state, the full reward for hosting the slot, along with the collateral, is collected.
- In the `Failed` state, no funds are collected. The reward is returned to the client, and the collateral is burned. The slot is removed from the list of slots and is no longer included in the list of slots returned by the `mySlots()` function.
In a blockchain, it is impossible to act on events that **do not happen** since every action results from a transaction. Therefore, our smart contract requires an external trigger to periodically check and confirm that a storage proof has been delivered by the SP. This is where the validator role is essential.
It is the smart contract that checks if the proof requested from an SP has been delivered. The validator's job is to trigger this check on the smart contract for SPs "observed" by the validator. To incentivize validators, they receive a reward each time they help identify a missing proof from an SP.
Each time a validator observes the `SlotFilled` event, it adds the slot reported in the `SlotFilled` event to its list of watched slots. Then, at the end of each period, a validator has up to `config.proofs.timeout` seconds (a configuration parameter retrievable with `getConfig()`) to request proof validation from the smart contract for each slot in its list. If a slot lacks the required proof, the validator SHOULD call the `markProofAsMissing(slotId, period)` function on the smart contract. After confirming the missing proof for the slot with ID `slotId` in the given `period`, the `markProofAsMissing(slotId, period)` function will reward the validator.
If validating all the slots observed by the validator is not feasible within the specified `timeout`, the validator MAY choose to validate only a subset of the observed slots.
