Jedis-Mock

Java reimplementation of Redis/Valkey for white box testing

Ivan Ponomarev

Team Lead @ Synthesized.io
Teaching CS @ Northeastern University London

2. What is Redis/Valkey?

Remote Dictionary Server
In-memory key–value database
Different kinds of abstract data structures
Not open source since 2024 with OOS forks available, but it’s a separate story…

https://db-engines.com/en/ranking

https://db-engines.com/en/ranking/key-value+store

5. Scale of Users

Twitter, AirBnB, Tinder, Yahoo, Adobe, Hulu, Amazon…

(https://en.wikipedia.org/wiki/Redis#Users)

6. What makes Redis so popular?

Really easy to use
Really high performant: 10000 max clients default value (compare with the default maximum for 100 for PostgreSQL)
Swiss army knife: caches, queues, atomic counters, distributed locks, semaphores, Bloom filters, etc etc etc

Price we pay: no ACID (only isolation)

7. So how do we test?

Testcontainers, of course!

(In fact, Redis is one of the first demo examples for Testcontainers ever)

8. Redis mocks in different programming languages

Python: https://github.com/cunla/fakeredis-py
NodeJS: https://github.com/yeahoffline/redis-mock
Java: https://github.com/fppt/jedis-mock

9. Why mock, if we have Testcontainers?

Still, Testcontainers require Docker and some time to download and run the container.
Jedis-Mock requires Java only.
Sometimes we want to manage the reply:
- imitate a system or network failure
- give a deterministic answer for non-deterministic query

10. Why mock, if we have Testcontainers?

Sometimes we need to check the actual queries being sent

11. JedisMock history

2017: Filipe Peliz Pinto Teixeira forked an abandoned POC repository.
2019: I tried to use it in my project and soon gave up
2021-now: actively developed by MIPT students under my supervision
2023: I have full rights for maintaing & publishig new versions

12. JedisMock: who uses it?

Known OSS dependents: Slack API client, dropwizard.io

13. JedisMock: who uses it?

Stable flow of incoming issues and requests.
180+ stars on GitHub — please give it a star!

14. Some technical details of importance

Redis is written in C (155 KLOC of C code)

15. Some technical details of importance

Redis/Valkey heavily relies on system calls and runs on Unix-like systems only.
Redis/Valkey uses a single thread to process all the requests (thus isolation).

16. JedisMock: the idea

sak

Redis data type

Java data type

HMAP

HashMap

LIST

ArrayList

SET

HashSet

ZSET

TreeSet+HashMap

Bit operations

BitSet

17. JedisMock: networking&threading

java.net.ServerSocket
Each incoming client occupies a thread in CachedThreadPool
All the threads synchronize on a single lock before making changes
Blocking operations (like BLPOP) use wait/notify

18. So how do we make it similar to real Redis?

Redis: 246 commands of varying popularity (Pareto 80/20 principle applies)
Jedis-mock: 166 commands supported (67% and the number is growing)

19. So how do we make it similar to real Redis?

Redis: 246 commands of varying popularity (Pareto 80/20 princple applies)
Jedis-mock: 166 commands supported (67% and the number is growing)

20. How do we count it?

Redis COMMAND command lists all the supported commands (currently 225 of them)
Classes annotated @Command(<COMMAND_NAME>) implement commands in Jedis-mock

21. An automatically built fit-gap report

22. What makes us sure that commands work correctly?

Comparison testing

@TestTemplate
public void whenHSettingOnTheSameKeys_EnsureReturnTypeIs1WhenKeysAreNew(Jedis jedis) {
    assertEquals(1L, jedis.hset(HASH, FIELD_1, VALUE_1));
    assertEquals(0L, jedis.hset(HASH, FIELD_1, VALUE_1));
}

23. What makes you sure that commands work correctly?

Comparison testing

24. Native Redis E2E tests

Written in Tcl

test {INCR over 32bit value} {
    r set novar 17179869184
    r incr novar
} {17179869185}

3932 tests in 200 Tcl scripts
we only run a subset of them, gradually increasing the number of successful "native" tests

25. Errors found by running "native" tests

SET  key1 stringvalue     // puts a string to key1
HSET key2 subkey1 avalue  // puts a hashmap to key2

GET  key1                 // returns stringvalue
GET  key2                 // error: WRONGTYPE

MGET key1 key2            // ????

26. Errors found by running "native" tests

LPOP  mylist     // returns the first element
                 // or null, if the list is empty

BLPOP mylist 1   // returns the first element
                 // or waits for it for 1 second maximum

BLPOP mylist 0   // ????

27. What next / open questions?

How closely a test mock should mock 'the real thing'?
Are those "bugs" Jedis-mock problems or Redis spec problems / possible source of bad practices?
Some day we will successfully run 100% of the tests. What happens when somebody finds a bug in Jedis-mock?

28. Thanks for listening!

Give jedis-mock a !

https://github.com/fppt/jedis-mock

ivan@synthesized.io

@inponomarev