use SavePolicy in multi pod will cause data lost

[王宇]

i use redis-watcher in my service, but i found SavePolicy in multi pod will cause data lost in concurrent request。(i use gorm as a adpter)

because i found NewEnforcer will select exist policy into memry, SavePolicy will truncate casbin_rule table and insert policy from memry, in three pod with concurrent request first, I found same case will cause casbin_rule data lost.

[casbin-bot]

@tangyang9464 @JalinWang @imp2002

[Zhou SiLe]

Thanks for your report, can you give a code sample that simulates the error? This helps us understand and try to solve the problem.

[王宇]

Thanks for your report, can you give a code sample that simulates the error? This helps us understand and try to solve the problem.

sorry i can't give the sample code ,because it not sample, the data lost case must trigger in multi pod with concurrent request.

but the SavePolicy code show , when i use db as adpter(gorm, use mysql as db), casbin.NewEnforcer will select policy from db, and SavePolicy will truncate casbin_rule and use policy from memry insert into db.

on concurrent request there has same time will truncate data, like this

pod1: select policy from db -> truncate casbin table -> insert into db with memry policy
pod2:                                                                               select policy from db -> policy is empty

after pod1 truncate casbin table , pod2 casbin.NewEnforcer with a empty policy from db

[Zhou SiLe]

I roughly understand your situation, do you expect (truncate casbin table -> insert into db with memory policy) to be an atomic operation? Then, other pods will not encounter an intermediate state with an empty policy

[王宇]

I roughly understand your situation, do you expect (truncate casbin table -> insert into db with memory policy) to be an atomic operation? Then, other pods will not encounter an intermediate state with an empty policy

yes!

[Zhou SiLe]

Ok, I'll try to fix this in the next few days.

[Zhou SiLe]

@WRAllen
By the way, do you have this problem with gorm-adapter or redis-adapter, or both? Can you give a link to the Github repository of the adapter you are using, as we may have multiple adapters of the same type.

According to your description, you use gorm-adapter (mysql database), but submitted an issue in redis-watcher, is there any relationship between them?

I think this issue is better submitted to the adapter's GitHub repository.

[王宇]

@WRAllen By the way, do you have this problem with gorm-adapter or redis-adapter, or both? Can you give a link to the Github repository of the adapter you are using, as we may have multiple adapters of the same type.

According to your description, you use gorm-adapter (mysql database), but submitted an issue in redis-watcher, is there any relationship between them?

I think this issue is better submitted to the adapter's GitHub repository.

OK～thinks you，I will