blockchain-etl/ethereum-etl-postgres

README

Ethereum ETL for PostgreSQL

The steps below will allow you to bootstrap a PostgreSQL database in GCP with full historical and real-time Ethereum data: blocks, transactions, logs, token_transfers, and traces.

The whole process will take between 24 and 72 hours.

Prerequisites:

Python 3.6+
gcloud
psql
Cloud SQL Proxy

1. Export Ethereum data from BigQuery to CSV files in GCS

Install gcloud and run gcloud auth login
Run

pip install -r requirements.txt
export BUCKET=<your_gcs_bucket>
bash ethereum_bigquery_to_gcs.sh $BUCKET

Optionally provide start and end dates: bash ethereum_bigquery_to_gcs.sh $BUCKET 2020-01-01 2020-01-31

Exporting to CSV files is going to take about 10 minutes.

2. Import data from CSV files to PostgreSQL database in Cloud SQL

Create a new Cloud SQL instance

export CLOUD_SQL_INSTANCE_ID=ethereum-0
export ROOT_PASSWORD=<your_password>
gcloud sql instances create $CLOUD_SQL_INSTANCE_ID --database-version=POSTGRES_11 --root-password=$ROOT_PASSWORD \
    --storage-type=SSD --storage-size=100 --cpu=4 --memory=6 \
    --database-flags=temp_file_limit=2147483647

Notice the storage size is set to 100 GB. It will scale up automatically to around 1.5 TB when we load in the data.

Add Cloud SQL service account to GCS bucket as objectViewer. Run gcloud sql instances describe $CLOUD_SQL_INSTANCE_ID, then copy serviceAccountEmailAddress from the output and add it to the bucket.
Create the database and the tables:

gcloud sql databases create ethereum --instance=$CLOUD_SQL_INSTANCE_ID

# Install Cloud SQL Proxy following the instructions here https://cloud.google.com/sql/docs/mysql/sql-proxy#install
./cloud_sql_proxy -instances=myProject:us-central1:${CLOUD_SQL_INSTANCE_ID}=tcp:5433

cat schema/*.sql | psql -U postgres -d ethereum -h 127.0.0.1  --port 5433 -a

Run import from GCS to Cloud SQL:

bash ethereum_gcs_to_cloud_sql.sh $BUCKET $CLOUD_SQL_INSTANCE_ID

Importing to Cloud SQL is going to take between 12 and 24 hours.

A few performance optimization tips for initial loading of the data:

Turn off fsync https://www.postgresql.org/docs/11/runtime-config-wal.html.
Use UNLOGGED tables.
Turn OFF auto backups and vacuum on Google Cloud SQL instance.

3. Apply indexes to the tables

NOTE: indexes won't work for the contracts table due to the issue described here #11 (comment)

Run:

cat indexes/*.sql | psql -U postgres -d ethereum -h 127.0.0.1  --port 5433 -a

Creating indexes is going to take between 12 and 24 hours. Depending on the queries you're going to run you may need to create more indexes or partition the tables.

Cloud SQL instance will cost you between $200 and $500 per month depending on whether you use HDD or SSD and on the machine type.

4. Streaming

Use ethereumetl stream command to continually pull data from an Ethereum node and insert it to Postgres tables: https://github.com/blockchain-etl/ethereum-etl/tree/develop/docs/commands.md#stream.

Follow the instructions here to deploy it to Kubernetes: https://github.com/blockchain-etl/blockchain-etl-streaming.