Simple Bank

Introduction

• This is a simple bank service built with Golang (gin), PostgreSQL, sqlc, PASETO.
• Easy to maintain, preferment also type-safe code for query generated from sqlc.
• Structured and dependency injection with uber-go/fx
• Test-driven development style with high test coverage using golang/mock
• Token-based authentication using PASETO and auth middleware.
• Containerized service, easy to run with docker-compose
• Separated config into env var.
• RESTful API with auto generated api doc

Functions

• User can create a User based on unique username and email.
• A log-in User can create multiple accounts with different currencies.
• Record all account balance changes in Entry table. Whenever some money is added to or subtracted from the account, an account entry record will be created.
• /transfer api, provide a money transfer function between 2 accounts. This happen within a transaction and transfer is thread-safe operation.

How to Run the Service

• docker compose up

Database Schema

Setup local development

Install tools

• Docker desktop

• TablePlus

• Golang

• Homebrew

• Migrate

  brew install golang-migrate

• Sqlc

  brew install sqlc

• Gomock

  go install github.com/golang/mock/[email protected]

Setup Dev Infrastructure

docker-compose -f docker-compose.infra.yaml up -d

How to generate code

• Generate SQL CRUD with sqlc:

  make sqlc

• Generate DB mock with gomock:

  make mock

• Create a new db migration:

  migrate create -ext sql -dir db/migration -seq <migration_name>

Play Manually with Postman

1. install Postman
2. import postman-cmds.json
3. have fun
   1. create user
   2. login
   3. JWT header: after login, copy the access_token in response and update variable the auth header with bearer {access_token}
4. check http://localhost:8080/swagger/index.html for API doc

Run Production code

Run Production server with Local Infra

docker-compose -f docker-compose.infra.yaml up -d
docker-compose -f docker-compose.server-only.yaml up --force-recreate --build api

Run All together

docker-compose up --force-recreate --build api

Progress

1. Setup local environment

2. Design dbdiagram with https://dbdiagram.io/

• Foreign Key: ref: > A.id,
• Timestamp Type: timestamptz
• Generate sql 000001_init_schema.up.sql

3. Setup Postgres with Docker and DB Migration

docker-compose -f docker-compose.infra.yaml up -d
docker exec -it <CONTAINER_ID> psql -d simple_bank -U root

• now we should be able to see tables created by migration script
• we can also connect DB with TablePlus

4. Generate CRUD Golang code from SQL

• Write CRUD SQL query in db/query
• generate golang code with make sqlc
• init go module go mod init github.com/hhow09/simple_bank

5. Write Golang unit tests for database CRUD with random data

• Write tests
  • main_test.go: to make db connection
  • use testQueries to access functions in [query].sql.go
  • write following tests
  • account_test.go
  • entry_test.go
  • transfer_test.go
• make test
• go context: carries deadlines, cancellation signals, and other request-scoped values across API boundaries and between processes.

6. implement database transaction in Golang

• Create store.go
  • Store: provides all funcs to execute queries and transactions
  • execTx: define a private transaction function: begin -> Commit or Rollback
  • TransferTx: define a public transfer transaction function
    1. create transfer record
    2. create Entry of from_account
    3. create Entry of to_account
    4. update accounts' balance
• Write store_test.go
  • create 5 goroutine to test transaction
  • get the err and result with go channel

7. Handle Transaction Lock

• Now transfer transaction will not pass the test since
  • GetAccount SQL is SELECT and does not block each other
  • it will result in all concurrent GetAccount just return initial value
• Create a SQL that SELECT FOR UPDATE

  -- name: GetAccountForUpdate :one
  SELECT * FROM accounts
  WHERE id = $1 LIMIT 1;
  FOR UPDATE
  

Deadlock

BEGIN;
INSERT INTO transfers (from_account_id, to_account_id, amount) VALUES (1, 2, 10) RETURNING *: -- exclusive lock on accounts

INSERT INTO entries (account_id, amount) VALUES (1, -10) RETURNING
INSERT INTO entries (account_id, amount) VALUES (2, 10) RETURNING *;
SELECT * FROM accounts WHERE id = 1 FOR UPDATE;
UPDATE accounts SET balance = 90 WHERE id = 1 RETURNING *;

SELECT * FROM accounts WHERE id = 2 FOR UPDATE; -- exclusive lock on accounts
UPDATE accounts SET balance = 110 WHERE id = 2 RETURNING *;
COMMIT;

• since transfers Table has foreign key from_account_id and to_account_id referencing accounts Table
• INSERT INTO transfers will acquire a RowExclusiveLock on accounts Table to ensure that ID of accounts are not consistent.
• SELECT * FROM accounts ... FOR UPDATE will also acquire a lock on accounts

Deadlock happens

TX1	TX2
BEWGIN
	BEGIN
	INSERT INTO TRANSFERS	TX2 lock on account table
	INSERT INTO ENTRIES
	INSERT INTO ENTRIES
INSERT INTO TRANSFERS		TX1 lock on account table
	SELECT * FROM accounts ... FOR UPDATE	waiting lock from TX1
INSERT INTO ENTRIES
INSERT INTO ENTRIES
SELECT * FROM accounts ... FOR UPDATE		waiting lock from TX2 deadlock!

Solution

• we are only update the balance of account. The lock is unneeded.
• change: FOR UPDATE -> FOR NO KEY UPDATE
• Refactor
  • getAccountForUpdate+UpdateAccount = AddAccountBalance

8. Avoid DeadLock

• We will encounter deadlock when 2 transactions: acc1 -> acc2 and acc2 -> acc1 are running concurrently.

-- gorutine 1: transfer from id=1 to id=2
BEGIN;
UPDATE accounts SET balance = balance - 10 WHERE id = 1 RETURNING *;
UPDATE accounts SET balance = balance + 10 WHERE id = 2 RETURNING *; 
COMMIT; 

-- gorutine 2: transfer from id=2 to id=1
BEGIN;
UPDATE accounts SET balance = balance - 10 WHERE id = 2 RETURNING *; 
UPDATE accounts SET balance = balance + 10 WHERE id = 1 RETURNING *;
COMMIT; 

• However if we switch the order so that transactions always acquire locks in a consistent order

if arg.FromAccountID < arg.ToAccountID {
	result.FromAccount, result.ToAccount, err = addMoney(ctx, q, arg.FromAccountID, -arg.Amount, arg.ToAccountID, arg.Amount)
} else {
	result.ToAccount, result.FromAccount, err = addMoney(ctx, q, arg.ToAccountID, arg.Amount, arg.FromAccountID, -arg.Amount)
}

the deadlock will not happen.

• we can test with TestTransferTxDeadlock

9. Isolation levels & read phenomena in MySQL & PostgreSQL

MySQL

SELECT @@transaction_isolation --isolation level of current session
SELECT @@global.transaction_isolation --isolation level of global session

PostgreSQL

• only has 3 isolation level since read uncommitted is actually read committed
• transaction isolation can only be set in one transaction.

show transaction isolation level;

repeatable read is different in MySQL and PostgreSQL

• in isolation level repeatable read

Table accounts
 id  | owner  | balance | currency |          
-----+--------+---------+----------+
   1 | tom    |     100 | USD      |
   2 | mary   |     100 | USD      |

Steps:
1. process A, select * from accounts: tom's balance = 100
2. process B, select * from accounts where id=1: tom's balance = 100
3. process A, update accounts set balance = balance - 10 where id=1 returning *; tom's balance = 90
4. process B, select * from accounts where balance >=100: tom will appear since tom's balance = 100, (repeatable read)
5. process A, commit; 
6. process B, update accounts set balance = balance - 10 where id=1 returning *; 
7. process B, commit;

• running those steps in MySQL:
  • after step 7 we will get tom's balance: 80
  • however it does not make sense since we expect tom's balance to be 100
• running those steps in MySQL:
  • after step 6 we will get ERROR: could not serialize access due to concurrent update

MySQL v.s. PostgreSQL

	MySQL	PostgreSQL
isolation levels	4	3
mechanism	lock	dependency detection
default isolation level	read commited	repeatable read

Reference

• MySQL - 15.7.2.1 Transaction Isolation Levels
• PostgreSQL - 13.2. Transaction Isolation

10. Github Action For unit test

• .github/workflows/test.yml

11. Implement RESTful HTTP API in Go using Gin

• install gin-gonic/gin for http router and implement REST API
• in api
  • define routing and handlers
  • use Model binding and validation to do param validation.

12. Load config from file & environment variables in Golang with Viper

• default config app.env
• env var can be further override with environment in docker-compose.yaml >

13. Mock DB for testing HTTP API in Go and achieve 100% coverage

• We need a mockDB

Currently Server is using store of real db

type Server struct {
	store  *db.Store
    ...
}

• we can generate Querier interface by settingemit_interface: true
  • var _ Querier = (*Queries)(nil) means Queries struct must implement Querier interface
• we can make a higher level abstraction Store interface
  • we can embed Querier interface in to Store to ensure it has all
  • SQLStore must implements Store
• make mock
  • package name: mockdb
  • destination db/mock/store.go
  • Interface to mock: Store
• write tests in account_test.go using store *mockdb.MockStore
  • use httptest.NewRecorder and router.ServeHTTP to write test and record them.
  • achieve 100%

14. Implement transfer money API with a custom params validator

• implement POST /transfers API
  • instead of hard code the available currency in param binding, register a custom currency validator to gin
  • use ShouldBindJSON in *gin.Context to parse JSON
  • validate requested transfer currency should match two accounts

15. Add users table with unique & foreign key constraints in PostgreSQL

• now we still missing user authentication and authorization, we need user Tables.
• modify dbdiagram to add user Table, some db constraints:
  • email should be unique
  • a user can have multiple account: accounts.owner should be the foreign key of username.
  • an owner cannot own two account with same currency:
    • (owner, currency) [unique]
    • ALTER TABLE "accounts" ADD CONSTRAINT "owner_currency_key" UNIQUE ("owner", "currency");
• generate postgreSQL script from https://dbdiagram.io/
• migrate create -ext sql -dir db/migration -seq add_users to add a new migration script, 000002_add_users.up.sql
• make migratedown

16. How to handle DB errors in Golang correctly

• make sqlc to generate updated model and *.sql.go
• write unit test user_test.go
• handle db constraints violation in CreateAccount
  • foreign_key_violation: an account should only be created when owner is an existing user
  • unique_violation: an owner-currency index should be unique
  • both of them should return http.StatusForbidden (403)

17. How to securely store passwords? Hash password in Go with Bcrypt!

• use bcrypt to do password hashing and checking in password.go
  • bcrypt will generate hashed password with cost random salt
  • if we hash same password twice, output should be different
• implement createUser handler in user.go.
  • check go-playground/validator - Baked-in Validations.

18. How to write stronger unit tests with a custom gomock matcher

• user_test.go
  • first randomUser() to generate a random user as input
  • compare the input user with API result, the password part should be checked with custom matcher since hashing 2 times will generate different result.

19. PASETO better than JWT

JWT

• composed of: base64(Header) + base64(Payload) + base64(Signature)
• for local use
  • symmetric digital signature algorithm: use same secrete key to sign and verify
• for public use
  • asymmetric digital signature algorithm: use private key to sign token and public key to verify token.
• cons: some algorithms are known to be vulnerable
• cons: prone to trivial token forgery:
  • Send a header that specifies the none algorithm be used
  • Send a header that specifies the HS256 algorithm when the application normally signs messages with an RSA public key.

PASETO

• A Thorough Introduction to PASETO
• similar to JWT
• for local use
  • symmetric encryption: use same secrete key to sign and verify
  • composed of: Version + local + encrypted(Payload) + base64(Footer)
  • Payload: data + expiration date + Nonce + Auth Tag
• for Public use
  • symmetric encryption: use same secrete key to sign and verify
  • composed of: Version + public + base64(signed string)
• Pros: Stronger algorithms than JWT.
• Pros: No trivial forgery since you don't need to choose algorithm.
• Pros: local Payload is encrypted, there won’t be any way for attackers to see any of your payload data without your secret key.

20. How to create and verify JWT & PASETO token in Golang

• token package

21. Implement login user API that returns PASETO or JWT access token in Go

• add config of TOKEN_SYMMETRIC_KEY and ACCESS_TOKEN_DURATION as server config
• setup loginUser API
• write test cases in user_test.go

22. Implement authentication middleware and authorization rules in Golang using Gin

• setup middleware.go and middleware_test.go to setup token verification for routes
• setup router.Group("/").Use(authMiddleware(server.tokenMaker)) in server.go
• update the auth rules for apis
  • createUser: public
  • CreateAccount: auth middleware, a user can only create accounts it owns.
  • getAccount: auth middleware, can only see accounts created by the request user itself.
  • listAccounts: auth middleware, can only see accounts created by the request user itself.
    • add WHERE owner = $1 for listAccount Query
  • CreateTransfer: auth middleware, fromAccount should be the account owned by user itself.
• update unit test with setupAuth

23. Build a minimal Golang Docker image with a multistage Dockerfile

test the built simple_bank_api container in dev environment

docker-compose -f docker-compose.infra.yaml up -d
docker-compose -f docker-compose.server-only.yaml up --force-recreate --build api

24. Use docker network to connect 2 stand-alone containers

• solved in docker-compose.server-only.yaml

25. write docker-compose.yaml

1. start db
2. run migration
3. start service

26. Generate API Doc with swaggo/swag

• make swagger then swagger files will be in docs
• run the app
• check http://localhost:8080/swagger/index.html

26. Dependency Injection using uber-go/fx

• reference: dipeshdulal/clean-gin