Introduction

Transactional Outbox Pattern in Go. Accompanies the blog post at https://www.gmhafiz.com/blog/transactional-outbox-pattern/

Quick start

docker-compose up -d redis postgres mailpit asynqmon

Run main api server

go run cmd/api/main.go

Run workers (as many as you want)

go run cmd/worker/main.go

asynq: pid=209309 2023/02/20 11:32:46.998734 INFO: Starting processing
asynq: pid=209309 2023/02/20 11:32:46.998761 INFO: Send signal TSTP to stop processing new tasks
asynq: pid=209309 2023/02/20 11:32:46.998768 INFO: Send signal TERM or INT to terminate the process

Send email

curl -v http://localhost:3080/api/mail/send -H 'Content-type: application/json' -d '{"from":"[email protected]","to":["[email protected]"],"subject":"Test Subject","content":"some content"}'

Monitor tasks

Go to http://localhost:8080

Show emails

Go to http://localhost:8025

Simulate Errors

Simulate email sending error by turning off the email server and send an email

docker-compose stop mailpit

curl -v http://localhost:3080/api/mail/send -H 'Content-type: application/json' -d '{"from":"[email protected]","to":["[email protected]"],"subject":"Test Mail Server Is DOWN","content":"some content"}'

Watch the task is retrying for 25 times and its mail_deliveries records has both a Failed status and error message saved.

To resume, re-start the mail server

docker-compose up -d mailpit

Shut down all containers

docker-compose down

Monitor Database

Using pg_top, we can see it is doing 1 transaction per second

sudo apt install pgtop

Run with

pgtop pg_top -h localhost  -p 5432 -d outbox_pattern -U user

last pid: 432794;  load avg  2.10,  2.17,  2.31;       up 1+14:17:12                                                                                                                                                10:01:27
5 processes: 5 sleeping
CPU states:  0.0% user,  3.7% nice,  1.0% system, 87.3% idle,  8.0% iowait
Memory: 54G used, 9089M free, 6822M buffers, 9511M cached
DB activity:   1 tps,  0 rollbs/s,   0 buffer r/s, 100 hit%,     66 row r/s,    0 row w/s 
DB I/O:     3 reads/s,   384 KB/s,   142 writes/s,   985 KB/s  
DB disk: 0.0 GB total, 0.0 GB free (100% used)
Swap: 385M used, 3710M free, 99M cached

  PID USERNAME PRI NICE  SIZE   RES STATE   TIME   WCPU    CPU COMMAND
   70 root      20    0    0K    0K sleep   0:01  0.02%  0.00% ksoftirqd/9
   27 root     -99    0    0K    0K sleep   0:00  0.00%  0.00% migration/2
   69 root     -99    0    0K    0K sleep   0:00  0.00%  0.00% migration/9
   31 root      20    0    0K    0K sleep   0:00  0.00%  0.00% cpuhp/3
   26 root     -51    0    0K    0K sleep   0:00  0.00%  0.00% idle_inject/2

Graceful Shutdown

To prevent tasks from being dropped prematurely, we gracefully shut down each service properly. They both listen to OS signals so if these were deployed in kubernetes, it will be handled automatically. Otherwise, follow instructions in each section below.

API

We know it is running at port 3080

kill -SIGTERM $(lsof -t -i :3080) 

Workers

Each worker has its own PID. Send a TSTP signal to stop processing new tasks, then send a TERM signal to shut down the worker

kill -SIGTSTP 209309
kill -SIGTERM 209309