Oh no, look out, it's another list of feed reader behaviors. This is based on my observations for those test keys which have been active in the past few days. Anything that stops polling eventually ages out of the report and won't show up here.
My thanks to those feed authors and contributors who have been cranking away to tweak things here and there. It's making a clear difference in the real world. I see more 304s and fewer 429s in general, and that makes me very happy.
Here is what I have to say for this time around, with some amount of aggregation applied. Some entries cover multiple test keys if the behavior is the same.
...
No UA. Sends bogus "" INM at startup, then makes up IMS values subsequently. Broken caching behavior that will miss updates in the real world.
Miniflux dev stuff. Seems a whole lot better now. Need to do some more pathological injections to see how they handle it.
Miniflux, older versions. Needs to upgrade. (Several different test keys).
SpaceCowboys Android RSS Reader 2.6.{31,32,33}. It's still doing unconditionals seemingly at random, and sometimes has bad timing. No idea what the hell is going on here.
FreshRSS dev stuff. Seems decent of late. This one also seems to be honoring Cache-Control "max-age" stuff, which is a bonus I never expected anyone to do. That's awesome. The useless referrers are also gone, so hooray! Lots of good stuff happening here.
FreshRSS/1.24.3. Didn't get tripped up on my last round of pathological header values. Will probably get even better when it picks up the latest dev work, like dumping the referrers (see earlier dev entry).
Audrey2 RSS reader, various versions. Seems pretty chill. Also seems to be honoring Cache-Control max-age! Great stuff.
Emacs Elfeed 3.4.1. No idea what controls the polling interval but it's doing fine. Some early weirdness that might be multiple instances running in parallel but nothing recently.
Slackbot. Broken beyond belief. Must have been some intern's project that's been left online to irritate people. Sends HEADs, gets 405s, sends more. Probably will force me to write "let's see what happens when you start getting 403s and 404s and 410s" and whatnot, and then I'll complain about it not honoring those, too. Watch and see.
Newsboat, various versions. Probably okay now in terms of caching behavior but I need to shove some more pathological cases at it to be sure.
feedmail.org/0. This one looks like it's trying desperately to honor the max-age thing, but sometimes it arrives a hair too early - on the order of about 100 msec here. So weird. How damn often is it considering these polls for it to be able to hit it that close? Does it ever rest?
SpaceCowboys RSS Reader, 2.6.{29,30,31,32,33}. This one doesn't seem to be doing the unconditionals or short timing any more. It did some early on but then it stopped the silliness.
Unread RSS Reader. Plugging away at a 21 minute interval, up from 15. At least they're conditional. (At least two distinct test keys.)
Some browser UA. Godawful timing, always unconditional. This would trip automated blocks on the real site, no doubt.
cry-reader v0.0. No idea how it picks its intervals, but they're always far enough apart and are always conditional, so that's great. Haven't managed to trip it up with any pathological header values.
Netvibes. Same super broken caching behavior, still.
SpaceCowboys ... etc ... 2.6.31. Never does IMS, only occasionally does INM, wildly unpredictable polling intervals. Not good behavior. No idea how there can be so much variance between these things (see above).
NextCloud-News dev. The caching behavior is getting better, and -huzzah- it has a real version number at last! This means it will be possible to treat all of those old versions as damaged and route around them. Would be nice if it didn't keep pulling / 3x and favicon 1x for every. single. feed. request. Those are wholly unnecessary.
NextCloud-News/1.0. This version is heavily broken in the caching department, but help is on the way (see dev version above). If you're running this, you should plan to upgrade to pick up those essential fixes.
NewsBlur. Usually okay but sometimes it does crazy stuff like 3 *unconditional* polls in two seconds. What? Why?! (UA is also crazy long since it's waving a dead chicken in the form of including a complete forgery of a Safari UA, (nested parens) and all).
Friendica, various versions. 100% unconditional requests. Occasionally shows up < 10s after the previous poll. Not good.
TTRSS. Not recommended.
ureq, various versions. Seems okay. Occasional short-spacing between polls, but at least they're conditional.
NetNewsWire. Multiple distinct test keys, but none of them have a version number so there's no way to know if anything's changed upstream. Since I started writing this list, one instance came back fast enough to demonstrate that it still has the buggy caching behavior.
haven, 100% unconditional, bumpy timing, unchanged from before.
inforss, multiple versions. Seems fine.
curl, various versions. It does conditionals properly and just has the usual 59m/60m fencepost scheduling thing.
Inoreader/1.0. This thing periodically flips to unconditional mode, and also has the 59m/60m thing going on. On the prod side of the house, it gets itself auto-blocked fairly often for bad behavior.
Feedbin. Seems fine now. (Three distinct test keys, same behavior.)
NetNewsWire. Probably still buggy - was about three weeks ago, at least. Need to see what happens when I do a fresh round of tests.
Liferea/1.15.3. Something is very wrong with the caching on this thing. It drops to unconditionals a lot, and recently started sending some bullshit 1970 IMS header with things like 5s and 14s intervals. No idea what the hell is going on there.
rawdog/2.24rc1. 138 days with no complaints from me.
NetNewsWire. Just like the last one: was showing signs of buggy cache behavior before, and probably will again when I next launch into the tests.
NewsBlur. Not quite as animated as the earlier instance but still uncorks an unconditional from time to time for some reason.
rss2email, various versions. Occasional unconditionals for some reason.
Newsboat/2.36.0. Jumpy timing, sometimes too fast. At least they're conditional.
CommaFeed, various versions. Has had caching bugginess in the past, and did during the last test. Probably still does.
Some browser UA. Keeps banging away with <1s unconditionals. Would definitely get auto-blocked in prod. Maybe multiple instances are running, but why would one always send unconditionals?
theoldreader.com. This seems to have switched behaviors to conditional polls in early September. I see a pair of too-quick polls since then, but even those were conditional so it's not that bad. This is also way better than before. Thank you for the fix, devs!
Generic "Go-http-client/1.1" UA. Seems fairly chill.
Rapids. This one turned a corner with some upgrades back in July and has been delightful since.
er0k feeds. Was upgraded at some point and then didn't trip over any of my "inject some crazy" stuff over the past month. Smooth sailing.
Emacs Elfeed 3.4.1. Mildly wacky behavior that suggests multiple instances banging away at the same key - either that, or it has really scary caching bugs. I hope it's the former and not the latter.
walrss/0.3.7. 145 days of complaint-free behavior. Nice.
Yarr/1.0. Just needs to tweak the 59m/60m fenceposting thing.
Some browser UA. This one settled down into a good polling interval and does conditional requests properly. Good enough.
Feedly. Still doing the unconditional requests daily for some reason. Fail. (Multiple distinct test keys, same behavior.)
feedparser/6.0.2. Just the 59m/60m thing. Otherwise is managing to do conditional (INM) requests so it's not the end of the world.
Reeder, various versions. This one acts strangely sometimes, and sends the wrong value in an INM header, like it got hit over the head and reverted to something from a few days *or weeks* before. Also sometimes sends unconditional requests. Could be multiple instances, so don't do that. One key = one instance of one reader. Anything else is chaos. (If this is you, ask me for more keys!)
Lots of mixed up stuff, but of late it's something claiming to be a browser that bangs away with unconditionals at 10 minute intervals. Would be auto-blocked in prod in a jiffy.
Yarr/1.0. Sends conditionals but has terrible timing - *way* too fast on those polls.
Bloggulus/0.4.0. Seems to be doing fine.
newsraft/0.27. This got some tweaks recently to deal with pathological ETag/LM behavior. Thanks for that!
feedparser/6.0.10. Still running too quickly: 29m/30m. They're conditional, but ... eh.
Mojolicious (Perl). 147 days of polls with no issues to report.
Some browser UA. Seems to be fine.
com.vanniktest.rssreader, various versions. Weird timing, as before. No idea what kind of scheduling it's using - repeats within 1-2s make no sense.
Broadsheet/0.1. Late update to this post: it's doing fine, but there's an unusual bit of wackiness showing up here. The URL it's using is checked into a github repo, and something or someone (not the author, and not Broadsheet) is pulling them. Maybe this is github trying to be "helpful". Joy.
[This entry originally said it was changing UAs and doing wild unconditional requests from time to time. It is not. Something else with the key is taking it for a ride. Raar.]
Occasional odd behavior - UA changes to something else, and it fires off multiple unconditional requests within a second. No idea if this is someone running multiple programs against the same key. (If this is you, ask for another key, or maybe get two, and stop using this one.)
feedbase-fetcher.pl/0.5. This settled into a good groove.
Artykul/1.0. The clown leak. I really need to just block them, or at least, start sending them really broken stuff to see what happens. Installing their app shouldn't have subscribed their backend to my early testing URL, but it did, and uninstalling it didn't stop it!
Instagram Engineering
Shopify Engineering