Implementation of the best NFT marketplace with locked balances.
- node 16.x
- docker
Example commands below use make(1). Please, have a look at commands in Makefile if your platform doesn't support it. On Windows we recommend to use WSL.
# 1. Install dependencies
npm ci
# 2. Compile typescript files
make build
# 3. Start target Postgres database and detach
make up
# 4. Apply database migrations from db/migrations
make migrate
# 5. Start the processor
make process
# 6. The command above will block the terminal
# being busy with fetching the chain data,
# transforming and storing it in the target database.
#
# To start the graphql server open the separate terminal
# and run
make serve
# 7. Now you can see the results by visiting the localhost:4350/graphql
Start development by defining the schema of the target database via schema.graphql
.
Schema definition consists of regular graphql type declarations annotated with custom directives.
Full description of schema.graphql
dialect is available here.
Mapping developers use TypeORM entities
to interact with the target database during data processing. All necessary entity classes are
generated by the squid framework from schema.graphql
. This is done by running npx squid-typeorm-codegen
command.
All database changes are applied through migration files located at db/migrations
.
squid-typeorm-migration(1)
tool provides several commands to drive the process.
It is all TypeORM under the hood.
# Connect to database, analyze its state and generate migration to match the target schema.
# The target schema is derived from entity classes generated earlier.
# Don't forget to compile your entity classes beforehand!
npx squid-typeorm-migration generate
# Create template file for custom database changes
npx squid-typeorm-migration create
# Apply database migrations from `db/migrations`
npx squid-typeorm-migration apply
# Revert the last performed migration
npx squid-typeorm-migration revert
This is an optional part, but it is very advisable.
Event, call and runtime storage data come to mapping handlers as raw untyped json. While it is possible to work with raw untyped json data, it's extremely error-prone and the json structure may change over time due to runtime upgrades.
Squid framework provides tools for generating type-safe wrappers around events, calls and runtime storage items for each historical change in the spec version.
The end result looks like this:
/**
* Normalized `balances.Transfer` event data
*/
interface TransferEvent {
from: Uint8Array
to: Uint8Array
amount: bigint
}
function getTransferEvent(ctx: EventHandlerContext): TransferEvent {
// instanciate type-safe facade around event data
let event = new BalancesTransferEvent(ctx)
if (event.isV1020) {
let [from, to, amount, fee] = event.asV1020
return {from, to, amount}
} else if (event.isV1050) {
let [from, to, amount] = event.asV1050
return {from, to, amount}
} else {
// This cast will assert,
// that the type of a given event matches
// the type of generated facade.
return event.asLatest
}
}
Generation of type-safe wrappers for events and calls is currently a two-step process.
First, you need to explore the chain to find blocks which introduce new spec version and fetch corresponding metadata.
npx squid-substrate-metadata-explorer \
--chain wss://kusama-rpc.polkadot.io \
--archive https://kusama.indexer.gc.subsquid.io/v4/graphql \
--out kusamaVersions.json
In the above command --archive
parameter is optional, but it speeds up the process
significantly. From scratch exploration of kusama network without archive takes 20-30 minutes.
You can pass the result of previous exploration to --out
parameter. In that case exploration will
start from the last known block and thus will take much less time.
After chain exploration is complete you can use squid-substrate-typegen(1)
to generate
required wrappers.
npx squid-substrate-typegen typegen.json
Where typegen.json
config file has the following structure:
{
"outDir": "src/types",
"chainVersions": "kusamaVersions.json", // the result of chain exploration
"typesBundle": "kusama", // see types bundle section below
"events": [ // list of events to generate
"balances.Transfer"
],
"calls": [ // list of calls to generate
"timestamp.set"
]
}
Squid tools assume a certain project layout.
- All compiled js files must reside in
lib
and all TypeScript sources insrc
. The layout oflib
must reflectsrc
. - All TypeORM classes must be exported by
src/model/index.ts
(lib/model
module). - Database schema must be defined in
schema.graphql
. - Database migrations must reside in
db/migrations
and must be plain js files. sqd(1)
andsquid-*(1)
executables consult.env
file for a number of environment variables.
Substrate chains which have blocks with metadata versions below 14 don't provide enough information to decode their data. For those chains external type definitions are required.
Type definitions (typesBundle
) can be given to squid tools in two forms:
- as a name of a known chain (currently only
kusama
) - as a json file of a structure described below.
{
"types": {
"AccountId": "[u8; 32]"
},
"typesAlias": {
"assets": {
"Balance": "u64"
}
},
"versions": [
{
"minmax": [0, 1000], // block range with inclusive boundaries
"types": {
"AccountId": "[u8; 16]"
},
"typesAlias": {
"assets": {
"Balance": "u32"
}
}
}
]
}
.types
- scale type definitions similar to polkadot.js types.typesAlias
- similar to polkadot.js type aliases.versions
- per-block range overrides/patches for above fields.
All fields in types bundle are optional and applied on top of a fixed set of well known frame types.
Polkadot.js provides lots of specialized classes for various types of data.
Even primitives like u32
are exposed through special classes.
In contrast, squid framework works only with plain js primitives and objects.
This allows to decrease coupling and also simply dictated by the fact, that
there is not enough information in substrate metadata to distinguish between
interesting cases.
Account addresses is one example where such difference shows up. From substrate metadata (and squid framework) point of view account address is simply a fixed length sequence of bytes. On other hand, polkadot.js creates special wrapper for account addresses which aware not only of address value, but also of its ss58 formatting rules. Mapping developers should handle such cases themselves.
It is possible to extend squid-graphql-server(1)
with custom
type-graphql resolvers and to add request validation.
More details will be added later.
Because subsquid requires an archival indexer to be fast, there are currently 3 options how to do it:
- Leave it as it is
there is already indexer for base basilisk
- using archival node for Koda BSX Sandbox 🐍
ARCHIVE_URL=https://basilisk-test.indexer.gc.subsquid.io/v4/graphql
- running your own
git clone [email protected]:subsquid/squid-archive-setup.git;
cd squid-archive-setup/basilisk
in docker-compose.yml
set url for the chain
- - WS_PROVIDER_ENDPOINT_URI=wss://basilisk.api.onfinality.io/public-ws
+ - WS_PROVIDER_ENDPOINT_URI=wss://basilisk-kodadot.hydration.cloud
then just
docker compose up
and set
ARCHIVE_URL=http://localhost:4010/v1/graphql