Analytics Pipeline

Analytics Pipeline

High-performance analytics with Redis counters and periodic database flush.

When to Use This Skill

• Need high-throughput event tracking (thousands/second)
• Want real-time counters without database bottlenecks
• Building dashboards with time-series data
• Tracking user activity, feature usage, or page views

Core Concepts

Write to Redis for speed, flush to PostgreSQL for persistence. Redis handles high write throughput, periodic workers batch-flush to the database.

Events → Redis Counters → Periodic Flush Worker → PostgreSQL → Dashboard Queries

Implementation

Python

from enum import Enum
from dataclasses import dataclass
from datetime import datetime, timezone, timedelta
from typing import Optional, Dict, List
import redis.asyncio as redis


class AnalyticsEventType(str, Enum):
    GENERATION_COMPLETED = "generation_completed"
    USER_SIGNUP = "user_signup"
    FEATURE_USED = "feature_used"
    PAGE_VIEW = "page_view"


@dataclass
class AnalyticsEvent:
    event_type: AnalyticsEventType
    user_id: Optional[str] = None
    properties: Optional[Dict] = None
    timestamp: Optional[datetime] = None

    def __post_init__(self):
        if self.timestamp is None:
            self.timestamp = datetime.now(timezone.utc)


class AnalyticsKeys:
    """Redis key patterns for analytics counters."""
    PREFIX = "analytics"

    @staticmethod
    def daily_counter(event_type: str, date: datetime = None) -> str:
        d = date or datetime.now(timezone.utc)
        return f"analytics:counter:{event_type}:{d.strftime('%Y-%m-%d')}"

    @staticmethod
    def hourly_counter(event_type: str, date: datetime = None) -> str:
        d = date or datetime.now(timezone.utc)
        return f"analytics:counter:{event_type}:{d.strftime('%Y-%m-%d:%H')}"

    @staticmethod
    def user_daily_counter(user_id: str, event_type: str, date: datetime = None) -> str:
        d = date or datetime.now(timezone.utc)
        return f"analytics:user:{user_id}:{event_type}:{d.strftime('%Y-%m-%d')}"

    @staticmethod
    def pending_flush_set() -> str:
        return "analytics:pending_flush"


class AnalyticsService:
    """High-performance analytics using Redis counters."""
    COUNTER_TTL = 7 * 24 * 60 * 60  # 7 days

    def __init__(self, redis_client: redis.Redis):
        self.redis = redis_client

    async def track_event(self, event: AnalyticsEvent) -> None:
        pipe = self.redis.pipeline()

        # Daily counter
        daily_key = AnalyticsKeys.daily_counter(event.event_type.value, event.timestamp)
        pipe.incr(daily_key)
        pipe.expire(daily_key, self.COUNTER_TTL)

        # Hourly counter
        hourly_key = AnalyticsKeys.hourly_counter(event.event_type.value, event.timestamp)
        pipe.incr(hourly_key)
        pipe.expire(hourly_key, self.COUNTER_TTL)

        # Per-user counter
        if event.user_id:
            user_key = AnalyticsKeys.user_daily_counter(event.user_id, event.event_type.value, event.timestamp)
            pipe.incr(user_key)
            pipe.expire(user_key, self.COUNTER_TTL)

        # Track for flush
        pipe.sadd(AnalyticsKeys.pending_flush_set(), 
                  f"{event.event_type.value}:{event.timestamp.strftime('%Y-%m-%d')}")

        await pipe.execute()

    async def get_daily_count(self, event_type: AnalyticsEventType, date: datetime = None) -> int:
        key = AnalyticsKeys.daily_counter(event_type.value, date)
        count = await self.redis.get(key)
        return int(count) if count else 0

    async def get_hourly_counts(self, event_type: AnalyticsEventType, date: datetime = None) -> Dict[int, int]:
        d = date or datetime.now(timezone.utc)
        pipe = self.redis.pipeline()
        for hour in range(24):
            hour_dt = d.replace(hour=hour, minute=0, second=0, microsecond=0)
            pipe.get(AnalyticsKeys.hourly_counter(event_type.value, hour_dt))
        results = await pipe.execute()
        return {hour: int(count) if count else 0 for hour, count in enumerate(results)}

class AnalyticsFlushWorker:
    """Periodically flushes Redis counters to PostgreSQL."""
    FLUSH_INTERVAL = 300  # 5 minutes
    BATCH_SIZE = 100

    def __init__(self, redis_client: redis.Redis, pg_pool):
        self.redis = redis_client
        self.pg = pg_pool
        self._running = False

    async def start(self) -> None:
        self._running = True
        while self._running:
            try:
                await self.flush()
            except Exception as e:
                logger.error(f"Flush error: {e}")
            await asyncio.sleep(self.FLUSH_INTERVAL)

    async def flush(self) -> int:
        pending = await self.redis.smembers(AnalyticsKeys.pending_flush_set())
        if not pending:
            return 0

        flushed = 0
        pending_list = list(pending)

        for i in range(0, len(pending_list), self.BATCH_SIZE):
            batch = pending_list[i:i + self.BATCH_SIZE]
            counters = await self._collect_counters(batch)

            if counters:
                await self._write_to_postgres(counters)
                flushed += len(counters)
                await self.redis.srem(AnalyticsKeys.pending_flush_set(), *batch)

        return flushed

    async def _collect_counters(self, pending_keys: List[str]) -> List[tuple]:
        counters = []
        pipe = self.redis.pipeline()

        for pending in pending_keys:
            parts = pending.split(":", 1)
            if len(parts) != 2:
                continue
            event_type, date = parts
            key = AnalyticsKeys.daily_counter(event_type, datetime.fromisoformat(date))
            pipe.getdel(key)  # Atomic get-and-delete

        results = await pipe.execute()

        for pending, count in zip(pending_keys, results):
            if count:
                parts = pending.split(":", 1)
                counters.append((parts[0], parts[1], int(count)))

        return counters

    async def _write_to_postgres(self, counters: List[tuple]) -> None:
        async with self.pg.acquire() as conn:
            await conn.executemany("""
                INSERT INTO analytics_daily (event_type, date, count, updated_at)
                VALUES ($1, $2, $3, NOW())
                ON CONFLICT (event_type, date)
                DO UPDATE SET count = analytics_daily.count + EXCLUDED.count, updated_at = NOW()
            """, counters)

Usage Examples

# Track events
analytics = AnalyticsService(redis_client)

await analytics.track_event(AnalyticsEvent(
    event_type=AnalyticsEventType.GENERATION_COMPLETED,
    user_id="user_123",
    properties={"model": "gpt-4"},
))

# Query real-time counts
today_count = await analytics.get_daily_count(AnalyticsEventType.GENERATION_COMPLETED)
hourly = await analytics.get_hourly_counts(AnalyticsEventType.GENERATION_COMPLETED)

# Start flush worker
worker = AnalyticsFlushWorker(redis_client, pg_pool)
asyncio.create_task(worker.start())

Best Practices

1. Use Redis pipelines for batched counter updates
2. Set TTL on counters to prevent memory growth
3. Use GETDEL for atomic flush to prevent double-counting
4. Upsert on flush to handle duplicate dates gracefully
5. Separate user vs global analytics tables for query efficiency

Common Mistakes

• Not setting TTL on Redis keys (memory leak)
• Using GET then DEL instead of GETDEL (race condition)
• Flushing too frequently (database load)
• Not batching flush operations

Related Patterns

• metrics-collection (system metrics)
• intelligent-cache (caching strategies)