Skip to content

Provides basic APIs to retrieve block, transaction, and event data for the most recent 50 blocks from Ethereum mainnet

License

Notifications You must be signed in to change notification settings

srinathln7/ethereum-data-service

Repository files navigation

VC-Ethereum Data-API-Service

VC-Ethereum Data-API-Service stores block info, transaction hashes, and events for the latest 50 blocks and provides API endpoints for querying events by address, blocks by number, and transactions by hash. Please refer to the project objective here which outlines the goal of the project and how the codebase is organized.

Prerequisites

Design and Architecture

Before diving into the codebase, it is highly recommended to review the design and architecture document. This provides a high-level explanation of how various services, components, and modules interact.

For a deeper dive into each service, refer to the README files in the respective directories:

Get Started

To build and start all required services, rename sample.env to .env and run:

# If Redis is already running on your localhost, stop it before running Redis in the container
sudo systemctl stop redis

make buildup

To stop all running services, run:

make builddown

To view logs in real-time for each service:


docker logs -f vc-api-server

docker logs -f vc-bootstrapper

docker logs -f vc-blocknotifier

docker logs -f vc-blocksubscriber

API Endpoints

ID Route Description Avg. Resp Time
VC-00 GET / List all routes 40.44 µs
VC-01 GET /v1/blocks List all block numbers currently available in local data store 9.13 ms
VC-02 GET /v1/block?block_number=<block_number> Get block info associated with a given block number 48.17 ms
VC-03 GET /v1/tx?tx_hash=<tx_hash> Get transaction info associated with a given transaction hash 631.97 µs
VC-04 GET /v1/events?address=<address> Get all events associated with a particular address 187.51 ms

VC-02, VC-03, VC-04 all get their info from the local data store.

Please note: When querying VC-04 with a widely used contract address such as 0xA0b86991c6218b36c1d19D4a2e9Eb0cE3606eB48 for Circle USDC Token, which can potentially involve fetching thousands of events, there may be a slight delay in response time, approaching close to a second. However, despite occasional delays, the average response time for VC-04 remains around 200ms.

Test

You can test the service using the following curl commands:

# VC-00: List all routes 
curl -X GET http://localhost:8080/ | jq 

# VC-01: List all block numbers currently available in local data store
curl -X GET http://localhost:8080/v1/blocks | jq

# VC-02: Get block info associated with a given block number 
curl -X GET http://localhost:8080/v1/block?block_number=<$BLOCK_NUMBER> | jq

# VC-03: Get transaction info associated with a given transaction hash
curl -X GET http://localhost:8080/v1/tx?tx_hash=<$TX_HASH> | jq

# VC-04: Get all events associated with a particular address
curl -X GET "http://localhost:8080/v1/events?address=<$ADDR>" | jq

Alternatively, you can test the service in your browser.

Observation and Analysis

To view real-time analytics of our local data store, navigate to http://localhost:5540 in your browser to access the Redis Insight GUI. Then, manually connect the database by clicking Add connection details manually and using the following details

host: redis
port: 6379

After continuously running the application for several hours, we monitored Redis using the Redis-Insight tool and observed that Redis peak memory consumption reached approximately 50MB. The number of indexed keys maintained averaged between 30,000 to 40,000, with Redis utilizing less than 1% of CPU resources. This performance is well within Redis's capabilities, as outlined in the official Redis FAQ, which states Redis can manage up to 2^32 keys and has been tested to handle at least 250 million keys per instance.

This validates that Redis is an optimal choice for our current project requirements.

Limitations and Future Improvements

In our current architecture, each service operates with a single instance, making each service vulnerable to being a single point of failure. This becomes particularly critical because if Redis experiences downtime, it impacts the entire application. To address this, we can implement strategies such as deploying Redis in a high-availability configuration using Redis Cluster. Additionally, we can deploy these service on Kubernetes and enable automatic scaling and resilience by managing multiple instances of each service. Implementing load balancing across these instances can further enhance availability and fault tolerance and incorporating monitoring and alerting mechanisms helps in promptly identifying and mitigating issues before they impact the entire system. These approaches collectively aim to enhance the reliability and availability of our application architecture.

Easy-to-Query APIs

To expose the stored data to customers in an easy-to-query API, I would consider implementing a GraphQL API on top of the existing API service. GraphQL provides a flexible and efficient way to query and retrieve data, allowing clients to request only the data they need.

API Security

To secure the API, I would implement the following measures:

  • Authentication and Authorization: Implement token-based authentication (e.g., JWT) and role-based access control (RBAC) to restrict access to the API.

  • Rate Limiting: Implement rate limiting to prevent abuse and protect the service from being overwhelmed by too many requests.

  • HTTPS: Use HTTPS to encrypt communication between clients and our API service.

References

About

Provides basic APIs to retrieve block, transaction, and event data for the most recent 50 blocks from Ethereum mainnet

Topics

Resources

License

Stars

Watchers

Forks