Redis cache using a socket

[nodle]

My host uses a socket for both Memcache and Redis. Apparently it's faster than using an IP. My question is I can't seem to configure it properly. It is enabled under my PHP setting and is active. This is the defualt setting for Redis according to the manual.

$config['cache']['enabled'] = true;
$config['cache']['provider'] = 'Redis';
$config['cache']['config'] = [
'host' => '127.0.0.1',
'password' => 'password'
];

I have tried to paste it under the host, then with no password since I don't have one. Along with changing host to database, and port. Still will not work. Any idea or has anyone used a socket with Redis? Thanks!

 

[nodle]

I may have gotten it to work by using this below in case others run into the same problem. If someone could tell me if this looks ok that would be helpful:

$config['cache']['enabled'] = true;
$config['cache']['provider'] = 'Redis';
$config['cache']['config'] = [
'host' => '/home/pathtoyoursocket.sock',
'port' => 0,
];

Also is there a way to tell if Redis is actually working?

 

[m1ne]

Also is there a way to tell if Redis is actually working?

$ redis-cli monitor

Or you'll see in the admin panel if you use Xon's addon.

 

[nodle]

$ redis-cli monitor

Where would I add this at? I am just using Xenforo code, not Xon's.

 

[shimspedy]

Can you send link to that xon Addon please

 

[imno007]

Can you send link to that xon Addon please

Redis Cache By Xon

This add-on uses Credis with a custom cache provider for Redis. For best performance, install the php extension: phpredis You must have a Redis instance installed, this is likely not possible with shared hosting While XenForo 2 has a redis...

xenforo.com

 

[masmo]

I may have gotten it to work by using this below in case others run into the same problem.

Thank you, you helped me a lot!
I hadn't added 'port' => 0, and it didn't work.

 

[VoklLLC]

Thank you, you helped me a lot!
I hadn't added 'port' => 0, and it didn't work.

You have to use something like this:

'server' => 'unix:///var/run/redis.sock'

You need to make sure that Redis has permissions to the sock file or it will not work.

Xon also has an example here for the socket setup: https://xenforo.com/community/resources/redis-cache-by-xon.5562/field?field=faq

 

[masmo]

You have to use something like this:

'server' => 'unix:///var/run/redis.sock'

You need to make sure that Redis has permissions to the sock file or it will not work.

Xon also has an example here for the socket setup: https://xenforo.com/community/resources/redis-cache-by-xon.5562/field?field=faq

My redis server path is /server/redis/redis.sock
If I set 'server' => '/server/redis/redis.sock' or 'server' => 'unix:///server/redis/redis.sock' the site does not load.

I am not using plugin and it seems that 'server' option is not recognized.

The only way to make it work is with:

$config['cache']['enabled'] = true;
$config['cache']['provider'] = 'Redis';
$config['cache']['config'] = [
'host' => '/home/pathtoyoursocket.sock',
'port' => 0,
];

 

[MySiteGuy]

I tried running it through a socket for 3 "big boards" on a single host, and there really wasn't a difference in performance worth noting. There was larger improvement going from Redis to Keydb (which is Redis compatible).

 

[tristan9]

Host-local redis (ie 127.0.0.1) should generally be as fast as a socket (though sockets are technically superior, yes).

What is really important however is to have redis colocated with XF. Even in a low-latency private network (think ~1ms lat to redis), the difference between local and remote is significant under load.

If impossible to do so, it’s 100% worth it to colocate a redis replica node and have only writes go to the remote redis node. At least in our experience.

 

[eva2000]

I tried running it through a socket for 3 "big boards" on a single host, and there really wasn't a difference in performance worth noting. There was larger improvement going from Redis to Keydb (which is Redis compatible).

How many cpu cores/threads did the server have? How are you measuring the improvement? In terms of latency or ops/sec or Xenforo performance?

I just tested Redis vs KeyDB vs Dragonfly - all three Redis compatible with memtier benchmark with 1:15 SET:GET ratio and very interesting results

Tested on Intel Xeon E-2276G 6C/12T, 32GB, 2x 960GB NVMe software raid 1 AlmaLinux 8 Centmin Mod LEMP stack server

Ops/sec


Average latency

99% Percentile latency

 

[Xon]

I've been keeping my eye out for Dragonfly, but it is disappointing to see such terrible 99% percentile latency statistics. But it is a fairly young project still, but the operational improvements are something that redis badly needs.

There are some outstanding bug reports around latency, so maybe this chart will change soon

Dragonfly is about 10x slower than Redis when used by JuiceFS · Issue #1285 · dragonflydb/dragonfly

JuiceFS supports using redis as the metadata engine, and we had feedback from the community that they wanted to use dragonfly to replace redis, but after testing, I found that dragonfly is very uns...

github.com

 

[MySiteGuy]

I don't have the specific details any longer, it must be at least 2 years ago, on a 24 core server. I looked at latency since it is something Googlebot scores against.

I've been keeping a watchful eye at Skytable's progress. It looks to be promising under high concurrent loads, blowing everything else out of the water.

 

[Xon]

How many cpu cores/threads did the server have? How are you measuring the improvement? In terms of latency or ops/sec or Xenforo performance?

I just tested Redis vs KeyDB vs Dragonfly - all three Redis compatible with memtier benchmark with 1:15 SET:GET ratio and very interesting results

Ops/sec

View attachment 290021
Average latency
View attachment 290022
99% Percentile latency

View attachment 290023

Did you use --tcp_nodelay with dragonfly (redis sets this by default) ? Apparently this can help the latency figures

 

[eva2000]

I've been keeping my eye out for Dragonfly, but it is disappointing to see such terrible 99% percentile latency statistics. But it is a fairly young project still, but the operational improvements are something that redis badly needs.

It's by design straight from Dragonfly devs - they designed it for high CPU core counts - most of their benchmarks are with up to 30+ CPU cores IIRC. See https://github.com/dragonflydb/dragonfly/discussions/1688. Basically if you have single or low CPU core counts keep using Redis

not a bug - but a design choice.

we had many questions like these before. @Niennienzz we should probably add a documentation page on how to benchmark Dragonfly so we could just reply with a single link.

TLDR: if you intend to run your memory store locally or using a single thread - please continue using Redis. It has lower latency for a single connection. Also, there are not many performance advantages when running Dragonfly in a single thread. So if this is your use-case - then do not use Dragonfly.

Every implementation has its own best practices.
Dragonfly can not parallelize efficiently workloads when serving a single connection because it needs to respect the order for the incoming requests on that connection. Dragonfly shines when there are multiple client connections. Dozens, hundreds - the more the merrier. Dragonfly also has limits internally of how much pipelining it allows. AFAIK, the factor is 10, meaning pipelining more than the limit won't produce more throughput. The reason for this is that pipelining adds more latency and more memory pressure and again, the best practice with Dragonfly is to parallelize using multiple connections.

I don't have the specific details any longer, it must be at least 2 years ago, on a 24 core server. I looked at latency since it is something Googlebot scores against.

I've been keeping a watchful eye at Skytable's progress. It looks to be promising under high concurrent loads, blowing everything else out of the water.

Yes saw Skytable's results too - another interesting option

Did you use --tcp_nodelay with dragonfly (redis sets this by default) ? Apparently this can help the latency figures

Ah I read about that but didn't set it. Will have to retest to see. But Dragonfly results are inline with their devs claims - they've optimised for high CPU core count servers