Designing Notification Systems at Scale: Push, Email, SMS, and In-App

Notifications are how your application stays in touch with users. Done well, they drive engagement and retention. Done poorly, they drive uninstalls. The technical challenge is building a system that's reliable, scalable, personalized, and respects user preferences across multiple channels.

Notification Architecture

A production notification system has three layers: (1) Event Layer—events trigger notification intents (article published, comment received). (2) Decision Layer—determine which users to notify, via which channels, based on preferences. (3) Delivery Layer—send via the appropriate provider (FCM for push, SES for email, Twilio for SMS) with retry and rate limiting.

Preference Management

class NotificationPreference(db.Model):
    user_id = db.Column(db.Integer, ...)
    notification_type = db.Column(db.String)  # "new_article", "comment_reply"
    channel = db.Column(db.String)            # "email", "push", "sms", "in_app"
    enabled = db.Column(db.Boolean, default=True)
    # Frequency limits
    max_per_hour = db.Column(db.Integer, default=5)
    max_per_day = db.Column(db.Integer, default=20)

Deduplication and Rate Limiting

Use Redis to track notification counts per user per channel per time window. Before sending, check if the user has exceeded their preference limits. Deduplicate—if a user receives 100 upvotes in 5 minutes, send one "You received many upvotes" digest, not 100 individual notifications.

Delivery Providers and Fallback

Never depend on a single delivery provider. Implement a fallback chain: primary provider fails → retry 3 times → failover to secondary provider. Track delivery rates per provider and automatically switch when delivery drops below 95%.

Do-Not-Disturb Windows

Respect user timezone and DND settings. Queue non-urgent notifications and deliver them at the user's preferred time window. Use Celery's ETA feature: send_notification.apply_async(args=[...], eta=next_delivery_time).

Production Event Sourcing & CQRS Configuration Example

Here is an enterprise-grade implementation snippet representing a command dispatcher and read-model projector pattern to enforce clean architectural boundaries:

from typing import Dict, List, Callable, Any

class Command:
    pass

class Event:
    pass

class CommandBus:
    def __init__(self) -> None:
        self._handlers: Dict[type, Callable] = {}

    def register(self, command_type: type, handler: Callable) -> None:
        self._handlers[command_type] = handler

    def dispatch(self, command: Command) -> Any:
        handler = self._handlers.get(type(command))
        if not handler:
            raise ValueError(f"No handler registered for {type(command)}")
        return handler(command)

# Read model projection example
class ReadModelProjector:
    def __init__(self) -> None:
        self.views: Dict[str, Any] = {}

    def project(self, event: Event) -> None:
        """Update read-only projections dynamically in response to domain events."""
        event_name = type(event).__name__
        handler_name = f"handle_{event_name.lower()}"
        handler = getattr(self, handler_name, None)
        if handler:
            handler(event)

    def handle_ordercreated(self, event: Event) -> None:
        # Simulate projection update
        self.views[event.order_id] = {"status": "created", "total": event.total}

Production Trade-offs & Implementation Decisions

Deploying this solution in production environments requires a careful analysis of the trade-offs involved. For instance, focusing purely on consistency (such as ACID compliance) can limit network throughput and horizontal scalability. On the other hand, adopting an eventual consistency model can lead to dirty reads and requires complex conflict resolution strategies in the application layer.

At MirahLabs, our engineering teams balance these architectural constraints by separating critical transaction paths from analytics workloads. We apply message-driven architectures with idempotent consumer systems to guarantee that network failures or retries do not result in double processing or state contamination.

Real-World Benchmarks & Resource Planning

Below is a typical performance comparison profile compiled by our engineering team in staging environments under simulated loads (10k concurrent virtual users):

When capacity planning, we recommend scaling out horizontally using containerized workloads rather than vertically upgrading underlying instance models. This maximizes uptime and provides cost efficiency through dynamic scaling policies.

Security Considerations & Vulnerability Mitigations

No production blueprint is complete without addressing security. Ensure that all data paths utilize encryption in transit (TLS 1.3) and at rest (using AES-256). Furthermore, implement strict Role-Based Access Control (RBAC) to limit operations. For APIs, always enforce rate limits (e.g. using token bucket algorithms in Redis) and run continuous static application security testing (SAST) in your CI pipeline.

How MirahLabs Applies This in Practice

Our experience building high-volume solutions like MirahCare.ai and Ayurveda.ai has taught us that early optimization is often a trap, but ignoring structural security and data design early leads to fatal development blocks. We design all client products from day one to support modular extensions, robust query indexing, and standard schema definitions, ensuring rapid iteration without technical debt growth.

Production Event Sourcing & CQRS Configuration Example

Here is an enterprise-grade implementation snippet representing a command dispatcher and read-model projector pattern to enforce clean architectural boundaries:

from typing import Dict, List, Callable, Any

class Command:
    pass

class Event:
    pass

class CommandBus:
    def __init__(self) -> None:
        self._handlers: Dict[type, Callable] = {}

    def register(self, command_type: type, handler: Callable) -> None:
        self._handlers[command_type] = handler

    def dispatch(self, command: Command) -> Any:
        handler = self._handlers.get(type(command))
        if not handler:
            raise ValueError(f"No handler registered for {type(command)}")
        return handler(command)

# Read model projection example
class ReadModelProjector:
    def __init__(self) -> None:
        self.views: Dict[str, Any] = {}

    def project(self, event: Event) -> None:
        """Update read-only projections dynamically in response to domain events."""
        event_name = type(event).__name__
        handler_name = f"handle_{event_name.lower()}"
        handler = getattr(self, handler_name, None)
        if handler:
            handler(event)

    def handle_ordercreated(self, event: Event) -> None:
        # Simulate projection update
        self.views[event.order_id] = {"status": "created", "total": event.total}

Production Trade-offs & Implementation Decisions

Deploying this solution in production environments requires a careful analysis of the trade-offs involved. For instance, focusing purely on consistency (such as ACID compliance) can limit network throughput and horizontal scalability. On the other hand, adopting an eventual consistency model can lead to dirty reads and requires complex conflict resolution strategies in the application layer.

At MirahLabs, our engineering teams balance these architectural constraints by separating critical transaction paths from analytics workloads. We apply message-driven architectures with idempotent consumer systems to guarantee that network failures or retries do not result in double processing or state contamination.

Real-World Benchmarks & Resource Planning

Below is a typical performance comparison profile compiled by our engineering team in staging environments under simulated loads (10k concurrent virtual users):

When capacity planning, we recommend scaling out horizontally using containerized workloads rather than vertically upgrading underlying instance models. This maximizes uptime and provides cost efficiency through dynamic scaling policies.

Security Considerations & Vulnerability Mitigations

No production blueprint is complete without addressing security. Ensure that all data paths utilize encryption in transit (TLS 1.3) and at rest (using AES-256). Furthermore, implement strict Role-Based Access Control (RBAC) to limit operations. For APIs, always enforce rate limits (e.g. using token bucket algorithms in Redis) and run continuous static application security testing (SAST) in your CI pipeline.

How MirahLabs Applies This in Practice

Our experience building high-volume solutions like MirahCare.ai and Ayurveda.ai has taught us that early optimization is often a trap, but ignoring structural security and data design early leads to fatal development blocks. We design all client products from day one to support modular extensions, robust query indexing, and standard schema definitions, ensuring rapid iteration without technical debt growth.