Connect an account
The account request follows the web path to Accounts API and the accounts table.
Interactive diagram · architecture
Design Mint.com
Extract transactions
Slow extraction moves to a queue, then the worker stores, categorizes, and aggregates the result.
Recommend and notify
Monthly spending drives Budget Service, which can send a notification.
Read summaries
The read path checks memory first and falls back to SQL read replicas.
Client
What it is
The user-facing Mint application.
Its job here
It sends account and read requests to the Web Server. It receives completed account data and summaries.
Trade-off
Users make transactions daily, but few visit the site daily.
In the primer
DNS
What it is
It resolves the Mint domain for the client.
Its job here
It receives the client lookup. It directs the client toward Mint's front door.
Trade-off
DNS is one of the components added to address scalability issues.
In the primer
CDN
What it is
A cache for static content.
Its job here
It serves static content from the Object Store to the client. It reduces work on the Web Server.
Trade-off
Static content can be served from the Object Store and cached on the CDN.
In the primer
Load Balancer
What it is
A replicated front door for application traffic.
Its job here
It receives client traffic after DNS. It forwards requests to the Web Server fleet.
Trade-off
Load balancers and multiple Web Servers address bottlenecks.
In the primer
Web Server
What it is
A reverse proxy for Mint requests.
Its job here
It receives client traffic from the Load Balancer. It forwards account work to Accounts API and reads to Read API.
Trade-off
Internal load balancers are not shown to reduce clutter.
In the primer
Category Service
What it is
A seller-to-category dictionary.
Its job here
Transaction Extraction Service sends it each transaction. It hands a category back for the Budget Service and SQL Database.
Trade-off
Manual category overrides can supply sellers not initially seeded in the map.
Budget Service
What it is
The service that calculates monthly spending by category.
Its job here
It receives categorized transactions from Transaction Extraction Service. It passes budget status to Notification Service and stores monthly spending.
Trade-off
A generic budget template means only user overrides need storage.
Notification Service
What it is
A service that tells users about budget and extraction events.
Its job here
It receives budget status from Budget Service and completion events from Transaction Extraction Service. It sends notices to the user asynchronously.
Trade-off
Budget notifications do not need to be instant.
In the primer
Transaction Extraction Service
What it is
A worker that extracts transactions for an account.
Its job here
It pulls jobs from Queue. It writes raw logs to Object Store, calls Category Service and Budget Service, then updates SQL Database.
Trade-off
Extraction can take awhile, so asynchronous work adds complexity.
In the primer
Queue
What it is
A holding area for slow account work.
Its job here
Accounts API puts extraction work on it. Transaction Extraction Service pulls the job and starts processing.
Examples
- Amazon SQS
- RabbitMQ
Trade-off
It lets the service extract transactions asynchronously, but adds complexity.
In the primer
Accounts API
What it is
The API that accepts financial account details.
Its job here
It receives account requests from the Web Server. It updates the SQL Database accounts table and queues extraction work.
Trade-off
Account additions and removals are relatively rare.
In the primer
Read API
What it is
The API for summaries and transactions.
Its job here
It receives read requests from the Web Server. It checks Memory Cache, then fetches from SQL Read Replicas on a miss.
Trade-off
Popular content should use Memory Cache instead of the database.
In the primer
Memory Cache
What it is
Memory for sessions, aggregate stats, and recent transactions.
Its job here
Read API checks it first. On a SQL miss, Read API updates it with fetched content.
Examples
- Redis
- Memcached
Trade-off
It helps with uneven traffic and traffic spikes.
In the primer
SQL Write Master–Slave
What it is
The SQL database that stores accounts, transactions, and monthly spending.
Its job here
Accounts API writes account details here. Transaction Extraction Service writes categorized transactions and monthly aggregates.
Trade-off
A single SQL writer might be tough for transaction writes.
In the primer
SQL Read Replicas
What it is
SQL copies that serve read cache misses.
Its job here
They receive cache-miss reads from Read API. They return summaries and transactions to Read API.
Trade-off
Replicas must not be bogged down with replicating writes.
In the primer
Object Store
What it is
Storage for raw transaction log files and static content.
Its job here
Transaction Extraction Service writes raw logs here. CDN can serve its static content to the client.
Examples
- Amazon S3
Trade-off
Running analyses on transaction files can reduce database load.
← → step · space play/pause · R replayStep 4 of 4
Detail · follows playback, or click any box or step to pin it
Close · follow playbackClick any box for an explanation, or use Next or Play to walk through the four steps.
1 · Connect an account
The client reaches a Web Server, which sends account information to Accounts API. Accounts API updates the SQL accounts table.
2 · Extract transactions
Accounts API queues the slow job. The worker extracts logs, categorizes transactions, aggregates spending, and updates storage.
3 · Recommend and notify
Budget Service calculates spending by category and asks Notification Service to warn users near or over budget.
4 · Read summaries
Read API checks Memory Cache first. On a miss it fetches SQL data and refreshes the cache.