I would also like to get more technical details about this.

I noticed increased latency (>300ms) when the JS of the worker gets very big, but the CPU time in the dashboard was still very low (~5ms).

https://developers.cloudflare.com/workers/platform/limits#cpu-runtime According to the docs, latency should not be correlated with cpu runtime. But if you actually face the situation, I think there is a bug.

Oh wow they increased the Cache Max object size. I’m sure it was 50 MB, I wonder when this was changed?

http://web.archive.org/web/20200821164009/https://developers.cloudflare.com/workers/platform/limits#cache-api-limits At least since August

no earlier history is available

Are you referring to the CPU time vs real-time under the CPU runtime section of the linked document?

yep. Latency only refer to so-called realtime as I understand

Yes, but doesn’t mean having a connection open for longer doesn’t require more CPU time

right, so I also wanna know the answer to your question

👍

I also wonder that besides javascript execution time, what else affect cpu time

But why this problem bother you? I think the latency is the biggest problem when serving to mainland China. If cf doesnt abort your cpu usage, then cpu time can be ignore.

Doesn’t bother me, I just want clarification. After noticing wild swings in CPU time I eventually found a correlation between China users and world users that I tested be copying a script and using a different domain for one set if users to another. I’d just like clarification around this so I can better understand it. If we had a clearer understanding from the beginning then I would have known the answer right way instead of having to guess and test the “issue” over the course of a month.

Again, I’ll mention this is only observer in my case with a non-enterprise account where i’m running personal projects.

There's no reasonable way to use HTMLRewriter as a DOM parser, right? I want to extract the inner text of some (specific types of) nodes, but I need to use some other library, right?

Looking at the docs it seems there is only a way to set the inner content and not get it, unlike attributes. Not sure if it's a missing thing in the docs or actually not there.

Yeah, I was slightly surprised that it didn't seem possible (hence me double-checking in here, in case I was just being an idiot)

Can't find in the docs. Not sure if it's quicker to check the code or mention someone like @User or @User 🙂

@User reading more, I believe it's handled in the "text chucnks" section... It does't exist as it's not guaranteeing getting all the text in one chunk.

Yeah, but I can't figure out how I get the element that that text came from though.

So if there's multiple elements, I can't tell which piece of text goes in which element.

Could it be that you do a

.on()

for a specific element it launches the handler for the element and then text in the same call? Never totally used HTMLRewriter for this use case...

It’s been a while since I used HTMLRewriter but I believe I did something similar by just looking at *, storing the current element as it recurses through, and then looking at the last seen element.

Ah, does it do everything in order?

And then doing my filtering in the element function directly, looking at tagName, etc. A bit inefficient, but functional.

As far as I recall, yes. Someone feel free to correct me though.

I officially concede. I can't make HTMLRewriter work as a parser. If you've got that code somewhere @User , I'd love to see it at some point (no worries if not). But I'm just going to find a library to do it for me.