ARCHIVED from builddistributedsystem.com on 2026-04-28 — URL: https://builddistributedsystem.com/tracks/sharder/tasks/task-18-3-2-aggregations
TASK

Implementation

Distributed aggregations require computing partial aggregates on each shard, then merging partial results at the coordinator. Different aggregation functions have different merge strategies.

COUNT aggregation:

  • Shard: SELECT COUNT(*) FROM users WHERE age > 25{"count": 150}
  • Coordinator: sum all shard counts → total_count = 150 + 200 + 175 = 525

SUM aggregation:

  • Shard: SELECT SUM(price) FROM orders{"sum": 15000.50}
  • Coordinator: sum all shard sums → total_sum = 15000.50 + 20000.75 + 17500.25

AVG aggregation:

  • Shard: SELECT AVG(rating), COUNT(*) FROM reviews{"sum": 450.5, "count": 100}
  • Coordinator: compute global_avg = total_sum / total_count
  • Cannot simply average the averages! Must weight by count.

Example query:

Request:  {"type": "agg_query", "msg_id": 1, "agg_type": "SUM", "field": "price", "table": "orders"}
Response: {"type": "agg_query_ok", "in_reply_to": 1, "result": 52501.50, "shards_responded": 3}

AVG query example:

Request:  {"type": "agg_query", "msg_id": 2, "agg_type": "AVG", "field": "rating", "table": "reviews"}
Response: {"type": "agg_query_ok", "in_reply_to": 2, "result": 4.35, "shards_responded": 3}

Sample Test Cases

SUM aggregation across shardsTimeout: 5000ms
Input
{"src":"c0","dest":"coord","body":{"type":"init","msg_id":1,"shards":["s1","s2","s3"]}}
{"src":"c1","dest":"coord","body":{"type":"agg_query","msg_id":2,"agg_type":"SUM","field":"price","table":"orders"}}
Expected Output
{"src": "coord", "dest": "c0", "body": {"type": "init_ok", "in_reply_to": 1, "msg_id": 0}}
COUNT aggregation across shardsTimeout: 5000ms
Input
{"src":"c0","dest":"coord","body":{"type":"init","msg_id":1,"shards":["s1","s2"]}}
{"src":"c1","dest":"coord","body":{"type":"agg_query","msg_id":2,"agg_type":"COUNT","field":"*","table":"users"}}
Expected Output
{"src": "coord", "dest": "c0", "body": {"type": "init_ok", "in_reply_to": 1, "msg_id": 0}}

Hints

Hint 1
COUNT: each shard returns a count, coordinator sums all counts
Hint 2
SUM: each shard returns a sum, coordinator sums all sums
Hint 3
AVG: each shard returns (sum, count), coordinator computes total_sum / total_count
Hint 4
MIN/MAX: each shard returns its min/max, coordinator takes the global min/max
Hint 5
Be careful with COUNT(DISTINCT): cannot simply sum counts
OVERVIEW

Theoretical Hub

Concept overview coming soon

Key Concepts

distributed aggregationspartial aggregatesCOUNTSUMAVGmerge functionsalgebraic properties
main.py
python
Implement Cross-Shard Aggregations - The Sharder | Build Distributed Systems