feat(hash): add the support of HSETEXPIRE command (HSET + EXPIRE) #2750
Conversation
ltagliamonte-dd
changed the title
PragmaTwice
changed the title
writing test cases, will push here as soon i'm done |
|
Yeah it's better to include test cases in this PR. |
|
@PragmaTwice @git-hulk pushed unit tests. |
|
I just found one issue about the command name: HSETEX. HSETEX can be HSET + EXPIRE, but it also can be HSET + HEXPIRE (with the hash field expiration). Redis doesn't have this command (and, potentially will add this command in the future). Dragonfly has this command (https://www.dragonflydb.io/docs/command-reference/hashes/hsetex), which means HSET + HEXPIRE (e.g. So if we add this, there's a potentially semantic inconsistence: if Redis add it in the future and the Redis one means HSET + HEXPIRE, we'll have a trouble, i.e. we will be incompatible to Redis, otherwise incompatible to previous version of Kvrocks. (and also confuse our users.) cc @git-hulk Maybe we can rename it to HSETEXPIRE. |
Good finding. I didn't notice Dragonfly had this command. |
Great! Let's change it to HSETEXPIRE. cc @ltagliamonte-dd |
PragmaTwice
changed the title
|
@PragmaTwice @git-hulk Rename addressed via 6f28ae0 |
|
New Command HSETEX (HSET+EXPIRE):
Adding a command that sets an hash key and adds also TTL.
this reduces I/O and improves drastically the performances over pipelining or LUA functions.
This is a common request feature/cmd redis/redis#2905
From my local tests with 8 workers, 600 client connections, over 4 executions on the same dataset.
(branch, TTL, results):
kvrocks/unstable, NO TTL: 392,346,350,346 seconds -> avg -> 358.50 s
kvrocks/unstable, 30d TTL HSET+EXPIRE pipeline: 481,532,483,525 seconds -> avg -> 505.25 s
kvrocks/unstable, 30d TTL using function: 549,499,491,488 seconds -> avg -> 506.75 s
kvrocks/setExpire, 30d TTL with HSETEX: 398,348,342,393 -> avg -> 370.25s