LiveVote: Real-Time Polling Engine

Standard polling applications often fail during live events due to massive database write contention. In a typical relational or simple document setup, having thousands of users vote within a short window causes single-document write locks (e.g., Firestore's strict limit of 1 write/second per document).

This contention depletes thread pools, causes connection timeouts, and drops critical vote transactions, leading to a frustrating user experience and inaccurate metrics during live interactive broadcasts.

I designed a decoupled buffer-and-process architecture to separate incoming HTTP traffic from database write cycles:

• In-Memory Buffer (Redis): Incoming votes are immediately ingested into a memory-efficient Redis list acting as a FIFO queue, returning a 202 Accepted status to the user in <15ms.
• Firestore Sharding: Instead of incrementing a single counter, I implemented a Distributed Counter Sharding pattern. Every poll counter is split across 20 distinct document shards.
• Asynchronous Worker Nodes: A group of lightweight Node.js consumer daemons pull vote batches from the Redis queue and write updates to random shards in Firestore, avoiding database write contention.

The toughest roadblock was dealing with uneven shard distribution (hotkey clustering). Under random shard assignment, statistical clustering led to certain shards receiving concurrent writes, re-triggering contention warnings.

To solve this, I replaced random selection with a consistent hashing algorithm. Using a custom distribution ring based on the voter's salted session ID, I ensured votes from different sessions mapped evenly to distinct shards.

Additionally, I built a scheduling service that runs every 5 seconds to query the shards, aggregate the numbers, and update a single materialized read-view document, ensuring clients can fetch clean poll totals with a single read operation instead of reading all shards on every page load.

Related Reading: For a deeper comparative analysis of backend JavaScript engines and why Node.js or Bun are suited for high-concurrency tasks, read my article: Node.js vs Bun vs Deno: The Ultimate JavaScript Runtime Comparison.