Cardano RPC node - GraphQL API setup

Hello,

We started to discover how we can setup Cardano RPC to be able to get blockchain data via API.

Accordingly to the Cardano architecture, we need just some components. We can interact using GraphQL with data stored in PostgreSQL DB and collected by DB-Sync from the Full node.

App --> GraphQL API --> PostgreSQL <-- DB Sync <-- Node

Also, we found information about cardano-node-ogmios, but looks like it is not solution we need.

The easiest way to run all this is using docker compose.

We already started

  1. Postgres - works
  2. Node - in sync
  3. DB-Sync - follow node and sync data
  4. GraphQL - doesn’t work (Connection refused)
  5. Hashura - started and we can access console but not clear why do we need it
  6. Ogmios - not use it because it is not specified on the Architecture diagram

Issue

GraphQL container is up and running and we also enabled debug logs

graphql  | {"name":"cardano-graphql","hostname":"65f028378a05","pid":1,"level":20,"module":"Db","msg":"pgSubscriber: Connected","time":"2022-05-25T11:30:23.065Z","v":0}
graphql  | {"name":"cardano-graphql","hostname":"65f028378a05","pid":1,"level":30,"module":"Server","msg":"Initializing","time":"2022-05-25T11:30:23.066Z","v":0}
graphql  | {"name":"cardano-graphql","hostname":"65f028378a05","pid":1,"level":30,"module":"HasuraClient","msg":"Initializing","time":"2022-05-25T11:30:23.466Z","v":0}
graphql  | {"name":"cardano-graphql","hostname":"65f028378a05","pid":1,"level":20,"module":"HasuraClient","msg":"{\"level\":\"info\",\"msg\":\"Help us improve Hasura! The cli collects anonymized usage stats which\\nallow us to keep improving Hasura at warp speed. To opt-out or read more,\\nvisit https://hasura.io/docs/1.0/graphql/manual/guides/telemetry.html\\n\",\"time\":\"2022-05-25T11:30:23Z\"}\n{\"level\":\"info\",\"msg\":\"Applying migrations...\",\"time\":\"2022-05-25T11:30:24Z\"}\n{\"level\":\"info\",\"msg\":\"nothing to apply\",\"time\":\"2022-05-25T11:30:24Z\"}\n","time":"2022-05-25T11:30:24.355Z","v":0}
graphql  | {"name":"cardano-graphql","hostname":"65f028378a05","pid":1,"level":40,"module":"Db","msg":"pgSubscriber: cardano-db-sync-extended starting, schema will be reset","time":"2022-05-25T11:31:12.982Z","v":0}
graphql  | {"name":"cardano-graphql","hostname":"65f028378a05","pid":1,"level":30,"module":"CardanoNodeClient","msg":"Initializing. This can take a few minutes...","time":"2022-05-25T11:31:13.469Z","v":0}
graphql  | {"name":"cardano-graphql","hostname":"65f028378a05","pid":1,"level":20,"msg":"Establishing connection to cardano-node: Attempt 1 of 101, retrying...","time":"2022-05-25T11:31:13.481Z","v":0}
graphql  | {"name":"cardano-graphql","hostname":"65f028378a05","pid":1,"level":20,"module":"DataFetcher","instance":"ServerHealth","msg":"Initial value fetched","time":"2022-05-25T11:31:14.483Z","v":0}
graphql  | {"name":"cardano-graphql","hostname":"65f028378a05","pid":1,"level":20,"msg":"Establishing connection to cardano-node: Attempt 2 of 101, retrying...","time":"2022-05-25T11:31:14.487Z","v":0}
graphql  | {"name":"cardano-graphql","hostname":"65f028378a05","pid":1,"level":20,"module":"DataFetcher","instance":"ServerHealth","msg":"Initial value fetched","time":"2022-05-25T11:31:15.687Z","v":0}
graphql  | {"name":"cardano-graphql","hostname":"65f028378a05","pid":1,"level":20,"msg":"Establishing connection to cardano-node: Attempt 3 of 101, retrying...","time":"2022-05-25T11:31:15.690Z","v":0}
graphql  | {"name":"cardano-graphql","hostname":"65f028378a05","pid":1,"level":20,"module":"DataFetcher","instance":"ServerHealth","msg":"Initial value fetched","time":"2022-05-25T11:31:17.132Z","v":0}

Questions

We can’t connect to the GraphQL playground and port is not opening and we have the following questions

  1. How long GraphQL will initialize?

  2. To which node it is trying to connect, because it needs connection only to Postgres accordingly to the Architecture?

  3. Accordingly to the Architecture diagram we need Node/DB-Sync/Postgres/GraphQL. What is about Hashura/Ogmios, do we need them and why?
    Their configuration is specified in the docker-compose for cardano-graphql.

  4. Is Cardano Node store all blockchain data?
    Analogy from Ethereum - Archive node, will we be able to discover all the transactions, starting from the genesis?

  5. If Node store all the data, is there a way to prune old data?

  6. DB-Sync started to sync data with PostgreSQL and we see 64G node-db / 60G postgres.
    It means that we have two copies of the data, one on the Node and second in the PostgreSQL?
    Is there a way to maintain just one copy of the data? It may refer to Q.2 - pruning.

  7. Should we wait Node to be sync before running DB-Sync or it will handle all by itself and handle all sync/pauses/delays and so on?

  8. DB-Sync extended is the same docker image and EXTENDED=true variable activate extended mode?
    If we initially started in a simple mode (without variable), can be change it and an epoch table will be added or we need to do a full re-sync?

1 Like

We discovered the following about Cardano RPC API configuration

  1. We should run Postgres / Ogmios / DB-Sync / Hasura / GraphQL
  2. GraphQL uses Postgres and Harusa and Ogmios
  3. Postgres configuration should be tuned
  4. When we start sync from the scratch or from snapshot we will have all the data from genesis block
  5. We will have a kind of duplicated data - on the Node and in the Postgres and it is not clear how to prune Node data (probably for now it is not possible)

We started our journey by running 4 different nodes with a different configuration and it was related to the disk performance and sync method.

Node Instance Performance Disk Sync type Sync duration
Cardano-ebs-gen r5a.xlarge 2 x vCPU / 16 GiB EBS - 400 GB / GP3/ 16,000 IOPS/ 1,000MiB/s Genesis > 180 days (estimate)
Cardano-ebs-snap r5a.2xlarge 4 x vCPU / 32 GiB EBS - 400 GB / GP3/ 16,000 IOPS/ 1,000MiB/s Snapshot ~ 11 days
Cardano-ssd-snap i4i.xlarge 4 x vCPU / 32 GiB SSD - 937 GB / AWS Nitro / 55,000 IOPS Snapshot ~ 4 days
Cardano-ssd-2x-gen i4i.2xlarge 8 x vCPU / 64 GiB SSD - 1,875 GB / AWS Nitro / 110,000 IOPS Genesis ~ 9 days

Note: Cardano-ebs-gen was stopped because of the too low sync speed. Maybe early experiment was not so correct and we will try to repeat it later.

Disk space

2022-06-20 - epoch 346

volumes/db-sync  -  12G
volumes/node-db  -  65G
volumes/node-ipc -   0G
volumes/postgres - 265G

Sync from snapshot

  1. OS partition should have enough space for snapshot downloading ~ 25 GB (archive size)
  2. Snapshot unpack is done to the db-sync container /tmp directory ~ 70 GB (snapshot size) and we can use volumes to mount it on the OS /data disk
  3. After snapshot restore is finished, we should restart whole container stack and then db-sync will continue to sync following the node blocks
  4. After db-sync will finish to sync from the node, we should restart whole stack and then graphql will open its interface

Sync from genesis

  1. We should use disk with a high IOPS ~ 60,000 IOPS, please see db-sync System Requirements
  2. After db-sync will finish to sync from the node, we should restart whole stack and then graphql will open its interface
  3. We should copy all data from SSD disk to EBS volume and then start whole stack from it

Guide for AWS

Strategy

  1. Run instance with instance store fast SSD - i4i.2xlarge
  2. Perform sync on SSD from genesis
  3. Mount EBS volume
  4. Copy data from SSD to EBS
  5. Change instance type to the one without instance store - r6i.xlarge
  6. Update postgres containet memory settings

Instance type: i4i.2xlarge
OS: Amazon Linux 2
Performance: 8 vCPU / 64 GiB
Disks: / = 20 GB / GP3 | /data = 1,875 GB AWS Nitro SSD / 200,000 Read IOPS / 110,000 Write IOPS | /data2 = 500 GB GP3 / 16,000 IOPS / 1,000 MiB/s

# Instance store

# List
lsblk

# Format
mkfs -t xfs /dev/nvme1n1

# Mount
mkdir /data
mount /dev/nvme1n1 /data
# Docker
sudo amazon-linux-extras install docker -y
sudo systemctl enable docker
sudo systemctl start docker
# Docker compose
sudo curl -L https://github.com/docker/compose/releases/latest/download/docker-compose-$(uname -s)-$(uname -m) -o /usr/local/bin/docker-compose
sudo ln -s /usr/local/bin/docker-compose /usr/bin/docker-compose
sudo chmod +x /usr/local/bin/docker-compose
# Volumes directories
directory="/data/cardano"

mkdir -p $directory/volumes/{db-sync,postgres,node-db,node-ipc}

cd $directory
# Get configs
yum install git

git clone \
  --single-branch \
  --branch 6.2.0 \
  --recurse-submodules \
  https://github.com/input-output-hk/cardano-graphql.git
# PostgreSQL secrets
mkdir secrets

echo -n cexplorer >secrets/postgres_db
echo -n postgres >secrets/postgres_user
grep -ao '[A-Za-z0-9]' < /dev/urandom | head -30 | tr -d '\n' >secrets/postgres_password
# Docker compose
vi docker-compose.yml
version: "3.9"

services:
  postgres:
    container_name: postgres
    image: postgres:11.5-alpine
    shm_size: 10g
    command: -c min_wal_size=2GB -c max_wal_size=8GB -c work_mem=4GB -c max_worker_processes=8 -c shared_buffers=16GB -c effective_cache_size=48GB
    environment:
      - POSTGRES_LOGGING=true
      - POSTGRES_DB_FILE=/run/secrets/postgres_db
      - POSTGRES_USER_FILE=/run/secrets/postgres_user
      - POSTGRES_PASSWORD_FILE=/run/secrets/postgres_password
    secrets:
      - postgres_db
      - postgres_password
      - postgres_user
    ports:
      - ${POSTGRES_PORT:-5432}:5432
    volumes:
      - postgres:/var/lib/postgresql/data
    restart: on-failure
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U postgres"]
      interval: 10s
      timeout: 5s
      retries: 5
    logging:
      driver: "json-file"
      options:
        max-size: "200k"
        max-file: "10"

  ogmios:
    container_name: ogmios
    image: cardanosolutions/cardano-node-ogmios:${CARDANO_NODE_OGMIOS_VERSION:-v5.4.0}-${NETWORK:-mainnet}
    environment:
      - NETWORK=${NETWORK:-mainnet}
    ports:
      - ${OGMIOS_PORT:-1337}:1337
    restart: on-failure
    healthcheck:
      test: ["CMD-SHELL", "curl -f 127.0.0.1:12788 || exit 1"]
      interval: 60s
      timeout: 10s
      retries: 5
    volumes:
      - node-db:/db
      - node-ipc:/ipc
    logging:
      driver: "json-file"
      options:
        max-size: "400k"
        max-file: "20"

  db-sync:
    container_name: db-sync
    image: inputoutput/cardano-db-sync:12.0.2
    command: [
      "--config", "/config/cardano-db-sync/config.json",
      "--socket-path", "/node-ipc/node.socket"
    ]
    environment:
      - EXTENDED=true
      - POSTGRES_HOST=postgres
      - POSTGRES_PORT=5432
      - RESTORE_SNAPSHOT=${RESTORE_SNAPSHOT:-}
      - RESTORE_RECREATE_DB=N
    secrets:
      - postgres_db
      - postgres_user
      - postgres_password
    depends_on:
      ogmios:
        condition: service_healthy
      postgres:
        condition: service_healthy
    volumes:
      - ./cardano-graphql/config/network/${NETWORK:-mainnet}:/config
      - db-sync:/var/lib/cdbsync
      - node-ipc:/node-ipc
    restart: on-failure
    logging:
      driver: "json-file"
      options:
        max-size: "200k"
        max-file: "10"

  hasura:
    container_name: hasura
    image: inputoutput/cardano-graphql-hasura:${CARDANO_GRAPHQL_VERSION:-6.2.0}
    ports:
      - ${HASURA_PORT:-8090}:8080
    depends_on:
      postgres:
        condition: service_healthy
    restart: on-failure
    environment:
      - HASURA_GRAPHQL_ENABLE_CONSOLE=true
      - HASURA_GRAPHQL_CORS_DOMAIN=http://localhost:9695
    secrets:
      - postgres_db
      - postgres_password
      - postgres_user
    logging:
      driver: "json-file"
      options:
        max-size: "200k"
        max-file: "10"

  graphql:
    container_name: graphql
    image: inputoutput/cardano-graphql:${CARDANO_GRAPHQL_VERSION:-6.2.0}-${NETWORK:-mainnet}
    environment:
      - OGMIOS_HOST=ogmios
      - ALLOW_INTROSPECTION=true
      - CACHE_ENABLED=true
      - LOGGER_MIN_SEVERITY=${LOGGER_MIN_SEVERITY:-debug}
    secrets:
      - postgres_db
      - postgres_password
      - postgres_user
    depends_on:
      postgres:
        condition: service_healthy
    expose:
      - ${API_PORT:-3100}
    ports:
      - ${API_PORT:-3100}:3100
    restart: on-failure
    logging:
      driver: "json-file"
      options:
        max-size: "200k"
        max-file: "10"

secrets:
  postgres_db:
    file: ./secrets/postgres_db
  postgres_password:
    file: ./secrets/postgres_password
  postgres_user:
    file: ./secrets/postgres_user

volumes:
  postgres:
    driver: local
    driver_opts:
        type: none
        device: /data/cardano/volumes/postgres
        o: bind
  node-db:
    driver: local
    driver_opts:
        type: none
        device: /data/cardano/volumes/node-db
        o: bind
  node-ipc:
    driver: local
    driver_opts:
        type: none
        device: /data/cardano/volumes/node-ipc
        o: bind
  db-sync:
    driver: local
    driver_opts:
        type: none
        device: /data/cardano/volumes/db-sync
        o: bind
# Run
docker-compose up -d

# Check logs
docker-compose logs -f

Migrate from SSD to EBS

  1. Mount EBS volume to /data2
  2. Rsync /data to /data2 ~ 4 hours
    rsync -av --delete --progress --stats /data/ /data2/
    
  3. Stop whole stack
  4. Perform latest rsync ~ 1 hours
  5. Umount SSD from /data and mount EBS to /data
  6. Start whole stack on /data
  7. Wait for node chunks validation and for db-sync and graphql start ~ 1 hour
  8. After some period of time, db-sync will finish to sync from the node and graphql will open the port
  9. Now we can change instance type to the one without SSD
1 Like

I’ve never set up the full Cardano API stack but for other application stacks on AWS I’ve seen an astonishing difference in reliability & performance after moving the DB part of the application to an instance of Amazon RDS:

Good point and probably should be tested. And we also can consider Amazon Aurora and maybe Serverless v2. We should pay attention to the price because of the hight I/O at least during the initial sync.

DB Sync best practices

Having all software on the same machine

The recommended configuration is to have node, DB Sync, and PostgreSQL servers on the same machine. During syncing (getting historical data from the blockchain), there is a large amount of data traffic between db-sync and the database. Traffic to a local database is significantly faster than traffic to a database on the LAN or other remote location.

1 Like