3d ago

[v1.4.29] Feed sometimes skips a whole load of entries

When scrolling through the feed it randomly happens to skip a bunch of entries. It suddenly jumps far down the feed for no apparent reason.

Some observations and possible cause thoughts:

I believe it skips one load batch (20) of entries.
Scrolling up those entries gets me back to where I was in the feed.
I think it happens when the last entry of a previous load batch expands an attached media item.
It's more apparent in media heavy feeds.
It's independent of the scroll speed.

14 comments

I have 1.4.30 stable enough for an alpha pre-release if you'd want to test that to see if the change I made to the feed fixed that "jump" behavior. It's an x64 build (I can't do arm builds locally), so hopefully that isn't an issue. There's still a lot of "to do" items, but it's stable enough to test that; I also daily drive the dev version.
Before switching, you'll want to export your Tesseract settings to a JSON file and keep those somewhere since there are some changes to the settings structure in this release, and I'm not sure how well those will downgrade if you switch back to 1.4.29 (I only have forward migrations in the code).
Export Settings: Settings -> Import/Export -> Export
Alternatively,since you're self-hosting, you could deploy the 1.4.30 pre-release to a different subdomain which would not affect your settings in your current deployment since local storage is based on the site origin.
Pre-Alpha Docker Tag: ghcr.io/asimons04/tesseract:1.4.30-pre-alpha-1
- I tried a few times but couldn't replicate the issue. I think I saw quite a bunch of times when it would have jumped but instead a different post quickly flashed for a fraction of a second but nothing else happened otherwise. I'll keep running this alpha and report back if such a jump should ever happen to me again.
  However, I did notice from the continued rather fast scrolling, that now with the v1.4.30-pre-alpha-1 one core was constantly at 100% utilization - even when switching to this post and doing other things for a bit. Only a reload of the tesseract tab let the process get back to normal CPU utilization. The process was pasta, the network component of podman I believe. Maybe there's some cleanup or cancellation still missing? I don't think I saw this extensive utilization over such a long time with the v1.4.29 version. Oh well, that's what an alpha is for :)
  
  Are you running it in podman on the same machine as you're running the browser?
  Not that there's an issue with that, just making sure I understand which aspect is causing a CPU spike (server side or client side). There are some server-side components, but they don't do a lot of heavy lifting: mostly ancillary, on-demand lookups (e.g. when you click on a Loops video, it does a server-side fetch to get the video URL since that's not available otherwise in the metadata).
  I have noticed that older FF versions (I run ESR 128.5 on my dev machine) does have that issue, but newer, non-ESR on my main machine does not. I don't recall when I first noticed that, but I think it was when I updated @sveltejs/adapter-node. I may try rolling that back to an older release (any newer, and it only works with Svelte 5 which I'm not ready to port to yet).

One possible cause is that in the last few versions, I changed the crosspost rollup behavior so that it checks for crossposts against the whole feed and not just the current batch So a post may come in, but there's already a matching crosspost in the feed above, so it folds into that and "jumps" a bit as that one briefly appears and disappears (I'm aware of that, but it's not been too frequent of an issue).
A whole batch seems too much for that to be the cause, but it is possible.
If you scroll back up and see that some posts suddenly have crossposts listed, then that could be the cause.
If so, then I guess I can revert that to just checking for crossposts within the current batch and deal with seeing more duplicate posts.
Another possible cause could be something interrupting the fetch but the page cursor gets out of sync?
Not sure what to ask to dive deeper into those without being able to see it, but maybe I can replicate the behavior somehow.
What instance are you browsing when that happens? feddit.org?
Are you using the hosted Tesseract instance, does feddit.org host an instance, or are you hosting one yourself?
Are you using "Compact" or "Card" view when that happens?
Do you have the "Enable Embedded Media in the Feed" option enabled?
I can probably test with the hosted instance, but just trying to get as much info as I can.
- If you scroll back up and see that some posts suddenly have crossposts listed, then that could be the cause.
  I just had a jump and checked: No crossposts on them. It was in the "all" feed with "scaled" in "hybrid" mode. Here is a video of all the jumped posts: https://imgur.com/a/PgSRhud The jump was from the PlayStation post to the Formula1/Ukraine post.
  What instance are you browsing when that happens? feddit.org?
  Yes.
  Are you using the hosted Tesseract instance, does feddit.org host an instance, or are you hosting one yourself?
  Although feddit.org has an instance, I'm using my own selfhosted instance though podman.
  Are you using "Compact" or "Card" view when that happens?
  I'm using Hybrid. I have the feeling it happens with unseen, hence expanded, posts though.
  Do you have the "Enable Embedded Media in the Feed" option enabled?
  Yes.
  
  Ok, thanks (to both replies). I'll match my settings and see if I can reproduce that.
  With the confirmation that it's probably not rolling them up as crossposts, the only thing I can think of is it may be missing a page cursor or something?
  How fluent are you with the browser dev tools? Would you be able to open those, put it on the network tab, and filter for "XHR", scroll through, and see if any API requests fail when you experience one of the jumps?
  Also, what about any of the feed options? Any of those set? I don't think they'd make a difference for this, but since I'm grasping at straws at the moment, it would at least be handy to know I guess.
  
  Other suggestions:
  Disable media embeds in the feed and see if that is the issue; you can still see the embeds in the feed, but you'll have to click the thumbnail to load the iframe with the media. Always showing iframes has been mostly deprecated for a while now, and in the next release (1.4.30), those are actually going away with everything being click to play. The reasoning is in the changelog, but in short, it makes a huge improvement in performance and memory utilization.

14 comments