Object cache on a small Woo shop, measured

Golden_Thistle

Put Redis on a client's wall art shop, on and off, same pages, logged in.

Cart and checkout: clearly faster!

Product pages barely moved.

The https://developer.wordpress.org/reference/classes/wp_object_cache/ page says that's expected.

Worth the RAM on a small VPS or do you skip it?

Axel.Gaudin

If the host does it for you, you don't even notice. OVHcloud switched a cache on for the managed plans in spring, I got a mail about it. Nothing to do on my side.

I never measured. I wouldn't know how. But the shop felt quicker in the back office : the sizes, the variations, all of that.

aLLengoMEZ

Sounds like a page cache. Did the mail say object cache anywhere?

Axel.Gaudin

If I look in the panel, it just says Memcached on !

Hidden_Panda

Felt quicker isn't data: it would be the page cache doing the work. The club shop runs at shared hosting without object cache and it is fine - if the shop is small you wouldn't notice.

evan

Your numbers are what the thing is supposed to do.

Anonymous product pages never hit PHP once a page cache is on so there's nothing for Redis to speed up there.

Cart and checkout are different. Woo sets a session cookie as soon as something is in the cart and the page cache steps aside so each of those requests runs the full query stack and that's where an object cache earns its RAM.

Logged-in users, same story.

How much RAM is Redis holding on that box?

Golden_Thistle

That's it, mate.

Logged in, TTFB halved, 900 ms to 450. Anonymous didn't move.

Redis is at 60 MB. Does the cookie bypass the page cache for the whole visit?

dustyfalcon

I'm since 2008 on plain shared hosting. But does this Redis thing work there at all? Thanks.

aLLengoMEZ

Only if your host runs Redis or Memcached for you. The object cache is a drop-in file, object-cache.php in wp-content. And it talks to a service on the server. So if there's no service there's nothing to connect to and you're back to the normal per-request cache you have now. Check your control panel for a Redis or Memcached toggle and if it isn't there. No plugin brings it.

The client site I look after is on WP Engine, their panel has Object Cache under Utilities, flip it on and Query Monitor's Object Cache panel shows hit and miss counts right away, https://wordpress.org/plugins/query-monitor/ if you don't have it.

Turkan.Turgut

I turn it off and on - like my printers! thx

ewoods

Redis on a small box gets worse once memory fills because it starts swapping and a cache hit ends up slower than the query it replaced so set maxmemory plus an eviction policy or don't bother. What's maxmemory on yours?

Golden_Thistle

Fair, that's on me. The client box has maxmemory 256mb and maxmemory-policy allkeys-lru in redis.conf, and it's there because another site filled the whole VPS last year and the log was one long row of 'Can't allocate memory' until MySQL fell over too. Should've said so. Still worth it for cart and checkout, but without the cap it's a time bomb. The names are in the redis.conf example at https://github.com/redis/redis

gabrielmunro

How do you measure TTFB without tools?

I just have a browser.

Thanks again

aLLengoMEZ

You've got the tool already, it's in the browser.

Press F12 on the shop, open the Network tab, tick Disable cache then reload the page. The first row is the page itself. Click it and open the Timing tab. The green bar, Waiting for server response, is your TTFB.

Do it twice: once in a private window, that's your anonymous number and once logged in with something in the cart, that's the one an object cache changes.

Reload three times each and take the middle one. The first hit is always slow.

https://web.dev/articles/ttfb explains what counts as good.

gabrielmunro

Found it, thanks, 1.4 seconds.

patrickkramer

I tried the same setup on the helmets shop last week, maxmemory 256mb and allkeys-lru... Checkout logged in went from about 1.9 s to 1.1 s, measured the dev tools way - Anonymous pages didn't move. I have left it on.

Golden_Thistle

Same ratio here, roughly half.

martinmastrangelo

Bookmarking this for when our host gets Redis.

evan

Pointed two people at this thread this week so for whoever lands here next: measure first, dev tools, logged in and out. If the logged-in number is the slow one an object cache is the fix, otherwise it isn't... Cap the memory, no cap, no Redis.