element-web icon indicating copy to clipboard operation
element-web copied to clipboard

Gracefully handle when rooms in space pagination token expires - Unable to paginate rooms in the space `/hierarchy` -> 400 `Unknown pagination token`

Open MadLittleMods opened this issue 3 years ago • 8 comments

Steps to reproduce

  1. Go to the space home
  2. Scroll and paginate the rooms in the space until you get stuck
    • Maybe this matters, I was also searching/filtering for a specific room before that
  3. Notice Failed to load list of rooms. message in Element and the underlying network request errors:

GET https://matrix-client.matrix.org/_matrix/client/v1/rooms/!OJBlkJuUrsKnqtNnTi%3Amatrix.org/hierarchy?suggested_only=false&from=iDciuFbCVinPIwUGYtUzQLHO&limit=20 -> ❌ 400 bad request

{"errcode":"M_INVALID_PARAM","error":"Unknown pagination token"}

I first created an issue in the Synapse tracker thinking the bad pagination token was being returned from the homeserver but it turns out there was just a 5 minute gap between pagination requests and Synapse only keeps track of pagination tokens within a 5 minute window.

What I think happened is this:

  1. Go to the space home
  2. Search/filter for a specific room
  3. Wait for the spinner appearing over and over for each /hierarchy request
    • It feels like at a certain point, the app gave up on paginating the rooms for some reason even though the loading spinner was still present. So I was waiting around even though it was doing nothing and pagination token expired behind the scenes. Is there something in the logs indicating why it would stop paginating?
  4. Get tired of waiting and clear the search/filter which reveals all rooms fetched so far
  5. Scroll down to the bottom of the list and try to paginate more
  6. Run into the Failed to load list of rooms. error and underlying network error /hierarchy ❌ 400 bad request

I think this is the last request Element made while I was searching/filtering even though it had a next_batch, and the spinner was still showing.

GET https://matrix-client.matrix.org/_matrix/client/v1/rooms/!OJBlkJuUrsKnqtNnTi%3Amatrix.org/hierarchy?suggested_only=false&from=rCkHRkRndYnCOIInVookuxrI&limit=20 -> ✅ 200 OK

date: Tue, 10 May 2022 19:02:38 GMT (response header)

{ "rooms": [...], "next_batch":"iDciuFbCVinPIwUGYtUzQLHO" }

Then 7 minutes later, I cleared the search/filter and tried paginating manually by scrolling to the bottom of the list:

GET https://matrix-client.matrix.org/_matrix/client/v1/rooms/!OJBlkJuUrsKnqtNnTi%3Amatrix.org/hierarchy?suggested_only=false&from=iDciuFbCVinPIwUGYtUzQLHO&limit=20 -> ❌ 400 Bad request

date: Tue, 10 May 2022 19:08:58 GMT (response header)

{"errcode":"M_INVALID_PARAM","error":"Unknown pagination token"}

Outcome

What did you expect?

  1. Element keeps paginating while the loading spinner is visible when searching/filtering
  2. Element gracefully fetches a new pagination token after the 5 minute expiration (we have to paginate from the beginning again)
  3. All of these problems are exacerbated because searching a space is painfully slow and there is no server endpoint to do it directly.
    • https://github.com/matrix-org/synapse/issues/11694
    • https://github.com/vector-im/element-web/issues/17550

What happened instead?

  • Element stopped paginating and showed an error: Failed to load list of rooms..
  • Element also continuously retries the request over and over with no back-off.

Logs: https://github.com/matrix-org/element-web-rageshakes/issues/12829

Operating system

Windows 10

Browser information

Chrome Version 100.0.4896.127

URL for webapp

https://develop.element.io/

Application version

Element version: 479d4bf64d97-react-14127c777b87-js-34cfa511049e Olm version: 3.2.8

Homeserver

matrix.org

Will you send logs?

Yes -> https://github.com/matrix-org/element-web-rageshakes/issues/12829

MadLittleMods avatar May 10 '22 21:05 MadLittleMods

Sorry, where does the https://spec.matrix.org/v1.2/client-server-api/#get_matrixclientv1roomsroomidhierarchy API say that the token can expire?

t3chguy avatar May 10 '22 21:05 t3chguy

Element is performing as per the spec, I suggest trying to reproduce on Android & iOS and I'm sure you'll find the same failure mode. Synapse or the Spec are wrong, the onus is on Synapse to prove which is which.

t3chguy avatar May 10 '22 22:05 t3chguy

Regardless of if the pagination token should expire or not, when I filter the space in Element, it should keep paginating the rooms in a space until it reaches the end. There were no results to fill up the scroll area to stop it and the loading spinner was still showing when I stopped filtering. I can actually confirm this because I took a screenshot of the unfiltered list with the loading spinner at the bottom which clearly indicates it wasn't done paginating and should have kept going during my filter (to extra confirm it's not just a stuck spinner, the list isn't complete).

If preferred, I can create a separate issue for this though. It's very related to how I ran into this situation in the first place though.


I've just searched again and reached a proper end state which wasn't there:

No results found
You may want to try a different search or check for typos.

MadLittleMods avatar May 10 '22 22:05 MadLittleMods

per https://github.com/matrix-org/synapse/issues/12697#issuecomment-1122947619 I am reopening this as a client-side problem.

turt2live avatar May 10 '22 22:05 turt2live

spec issue: https://github.com/matrix-org/matrix-spec/issues/1058

turt2live avatar May 10 '22 22:05 turt2live

In terms of design decisions:

largely what the UI should do in this case. It's not immediately clear whether we should throw everything away and refresh or if we should try and do something more intelligent

or maybe even a button which says "try again" or whatever

-- @turt2live, https://matrix.to/#/!bEWtlqtDwCLFIAKAcv:matrix.org/$I6cbmSawd27Nq6w9xw-68vj8MoJO3q6yE9x5MrVmLps?via=matrix.org&via=element.io&via=mozilla.org

MadLittleMods avatar May 11 '22 02:05 MadLittleMods

Downgrading this to occasional and minor, since I don't believe this happens super regularly, and the workaround is to reopen the space landing page and try again

robintown avatar May 12 '22 00:05 robintown

In terms of design decisions:

largely what the UI should do in this case. It's not immediately clear whether we should throw everything away and refresh or if we should try and do something more intelligent or maybe even a button which says "try again" or whatever -- @turt2live, https://matrix.to/#/!bEWtlqtDwCLFIAKAcv:matrix.org/$I6cbmSawd27Nq6w9xw-68vj8MoJO3q6yE9x5MrVmLps?via=matrix.org&via=element.io&via=mozilla.org

Just noting on this - it's really tough for us to make the right design decisions without pairing closely on the considerations or tradeoffs of any path with engineering. So, whoever ends up looking at this, please feel free to reach out for us to figure out the right path.

nadonomy avatar Aug 10 '22 15:08 nadonomy

Spec still hasn't been clarified.

t3chguy avatar Nov 24 '22 16:11 t3chguy