Swap ioredis for iovalkey

[bcomnes]

Swap to a default redis client with a higher chance of getting maintained. iovalkey is a friendly fork maintained by @mcollina. I'm guessing this would be a breaking/major version bump if this lands.

Checklist

• [x] run npm run test and npm run benchmark
• [x] tests and/or benchmarks are included
• [x] documentation is changed or added
• [x] commit message and code follows the Developer's Certification of Origin and the Code of conduct

[simoneb]

Just curious, the issue where this is being discussed in the ioredis repo mentions that they recommend migrating to the official client: https://github.com/redis/ioredis/issues/1870#issuecomment-2029351413.

Have we considered that, versus another community-maintained package which may end up having the same maintenance issues?

[gurgunday]

Just curious, the issue where this is being discussed in the ioredis repo mentions that they recommend migrating to the official client: https://github.com/redis/ioredis/issues/1870#issuecomment-2029351413.

Yeah honestly I had the same question in mind

It also applies to fastify-rate-limit, currently it accepts an ioredis instance

[climba03003]

I would go for the official one if it is a long term choice. Using ioredis was a decision on 2018 #19, something might change now.

I also notice redis provide a benchmark script that we might want to try. https://github.com/redis/node-redis/tree/master/benchmark

[simoneb]

Oh I'm not challenging the original choice of ioredis by any means, the official client may not have even existed back then as far as I know. If we're to revise that decision now I think the official client might be the sensible option to choose, unless there are obvious reasons to avoid it (e.g. licensing or feature-completeness)

[climba03003]

Oh I'm not challenging the original choice of ioredis by any means, the official client may not have even existed back then as far as I know.

I means the decision shown on the issue about using ioredis instead of redis is based on performance and promise support. promise should not be a problem nowadays. performance can be checked via https://github.com/redis/node-redis/tree/master/benchmark

It is a good time to re-visit the decision made in the past.

redis seems to provide more advance and up-to-date feature according to the issue you provide.

[mcollina]

I've forked ioredis to iovalkey because:

1. I want to avoid being in the situation of being unable to fix ioredis in case the need arises. ioredis is essentially dead, but it's massively used.
2. I expect the server to break the protocol backward compatibility. It would be the best business move for them to make the server incompatible, making the official client to talk to the official database.
3. Ioredis seem to significantly outperform the official client.

None of the options looks good right now. I would likely hold for now, and possibly switch if the need arises.

[climba03003]

I don't like post with last updated date only, it seems to give a fault image to the reader the content is up-to-date and latest.

It's last update on Jan 2, 2024 and all the others scripts (including result image) is written or provided 3 years ago. I judge the conclusion in the article is still up-to-date.

[climba03003]

Here the result of using https://github.com/poppinlp/node_redis-vs-ioredis

Operation	redis(ms)	redis with multi(ms)	redis with batch(ms)
set	1489.4949	87.3262	83.2496
get	1425.5875	79.2133	55.7291
hmset	1447.7094	73.6554	71.0114
hgetall	1505.95	91.0191	83.9859
incr	1437.0455	62.8249	48.9377
keys	1447.0564	88.7329	79.2056

Operation	ioredis(ms)	ioredis with multi(ms)	ioredis with pipeline(ms)
set	1482.0976	82.4503	74.1322
get	1465.7855	70.7754	58.4931
hmset	1522.3452	96.9624	93.7096
hgetall	1559.0312	92.6324	83.0306
incr	1420.0616	57.512	97.9488
keys	1499.0053	111.735	102.1956

redis had an average time of 536.541ms ioredis had an average time of 553.884ms

The benchmark may differ between platform, but it does not provide an extreme performance boost between redis and ioredis.

cc @kibertoad I see you open issue on that repo 2 days ago, you may take the above graph as reference. I tested with redis@4 and ioredis@5

[simoneb]

Well at least we know that performance is not an issue then

[mcollina]

@climba03003 redis has autopipelining enabled by default, while ioredis doesn't. Can you check again with it enabled in ioredis?

[climba03003]

Updated at 2024-04-02 23:00 (UTC+8)

I am a bit surprise on the result, after the second run with enableAutoPipelining. I do not expect redis is even faster, but ioredis becomes slower.

Running with redis before ioredis with enableAutoPipelining: true	Operation	redis(ms)	redis with multi(ms)	redis with batch(ms)
set	1432.3858	81.8358	69.8106
get	1328.0434	78.1827	73.8312
hmset	1491.0987	85.4957	67.8254
hgetall	1643.6342	81.7741	76.3429
incr	1528.9046	73.3113	50.5294
keys	1479.9891	101.0535	78.4952

Operation	ioredis(ms)	ioredis with multi(ms)	ioredis with pipeline(ms)
set	1799.9244	100.9509	78.3812
get	1960.0755	79.4042	114.0666
hmset	1858.4125	114.2729	111.7964
hgetall	1944.1496	101.6841	127.9608
incr	1814.5	70.3818	67.1832
keys	1866.9967	117.0985	102.4569

redis had an average time of 545.697ms ioredis had an average time of 690.539ms

Running with ioredis with enableAutoPipelining: true before redis

Operation	redis(ms)	redis with multi(ms)	redis with batch(ms)
set	1365.6306	82.9341	66.0203
get	1362.5358	63.0452	62.4757
hmset	1471.8367	79.0067	70.4595
hgetall	1504.3269	88.4178	86.7542
incr	1381.3638	56.7824	51.427
keys	1303.7898	100.6238	77.3041

Operation	ioredis(ms)	ioredis with multi(ms)	ioredis with pipeline(ms)
set	1743.7271	86.7177	73.0347
get	1960.0755	79.4042	114.0666
hmset	2019.5258	114.2276	112.0642
hgetall	1926.7789	112.3453	87.6865
incr	1791.4514	67.3637	67.4776
keys	1793.8793	113.9202	98.0459

redis had an average time of 515.263ms ioredis had an average time of 687.855ms

[bcomnes]

I don't have a strong opinion on the matter, but it does seem like if we go the ioredis route, iovalkey might be the better long term solution (if @mcollina agrees). Is the official redis client a drop in replacement as well? I can make another PR with that when I have time.

[jsumners]

I don't know if we should land this to fastify-redis or create sth like fastify-iovalkey

Unless the API and protocol is different, I don't see a reason to create a completely new plugin with a different name.

[mcollina]

I 100% expect the protocol to change in Redis 8.

[bcomnes]

Closing until its more clear how to approach the issue.