Messages disappearing or missing on Riot Desktop

[turt2live]

...on Riot Nightly (Desktop).

Probably a different failure mode of https://github.com/vector-im/riot-web/issues/1541 or #10032

[ara4n]

Just had it again, again in Nightly (albeit today's build rather than Sunday's). Worried it might be a regression in Nightly.

[turt2live]

@ara4n are you getting this on web at all?

[bwindels]

Only red flag I've seen so far in the rageshake is:

2020-03-11T00:20:33.614Z E Caught /sync error TypeError: Cannot read property 'highlight' of undefined
    at MatrixClient.eval (webpack-internal:///40:412:68)
    at MatrixClient.emit (webpack-internal:///38:157:7)
    at ReEmitter._handleEvent (webpack-internal:///66:33:17)
    at ReEmitter.forSource (webpack-internal:///66:41:24)
    at Room.emit (webpack-internal:///38:152:5)
    at Room.addReceipt (webpack-internal:///57:1708:8)
    at Room.addEphemeralEvents (webpack-internal:///57:1508:12)
    at eval (webpack-internal:///56:1253:10)
    at Module.promiseMapSeries (webpack-internal:///33:894:11)
    at async SyncApi._processSyncResponse (webpack-internal:///56:1165:3)

I assume that the line that caused the error is client.js:412, throwing there would prevent event being emitted on client for a sync, but perhaps not much more...

[bwindels]

Apart from the above PR, not sure what else to do here apart from waiting for more occurences/specifics to narrow it down. Please report here if you see it again @ara4n

[bwindels]

Interesting, Dave's rageshake contained the same error:

2020-03-12T18:26:57.178Z E Caught /sync error TypeError: Cannot read property 'highlight' of undefined
    at D.<anonymous> (vector://vector/webapp/bundles/5913369ec59578312adc/bundle.js:2:758309)
    at D.s.emit (vector://vector/webapp/bundles/5913369ec59578312adc/bundle.js:2:82475)
    at o._handleEvent (vector://vector/webapp/bundles/5913369ec59578312adc/bundle.js:2:589940)
    at o.n (vector://vector/webapp/bundles/5913369ec59578312adc/bundle.js:2:589991)
    at g.s.emit (vector://vector/webapp/bundles/5913369ec59578312adc/bundle.js:2:82418)
    at g.addReceipt (vector://vector/webapp/bundles/5913369ec59578312adc/bundle.js:2:822703)
    at g.addEphemeralEvents (vector://vector/webapp/bundles/5913369ec59578312adc/bundle.js:2:820798)
    at vector://vector/webapp/bundles/5913369ec59578312adc/bundle.js:2:730309
    at Object.I (vector://vector/webapp/bundles/5913369ec59578312adc/bundle.js:2:25446)
    at async m._processSyncResponse (vector://vector/webapp/bundles/5913369ec59578312adc/bundle.js:2:728881)

Which should be fixed with the above PR, so some evidence this is indeed the issue, although still not entirely sure how this is happening. It was also from the desktop fwiw.

[ara4n]

Just had this again on latest Nightly: https://github.com/matrix-org/riot-web-rageshakes/issues/2472

(Also https://github.com/matrix-org/riot-web-rageshakes/issues/1747 is a much older instance of the same symptoms)

[bwindels]

Changed to title as this is not just about messages the syncing user has sent, but also about the messages of other users not appearing.

[jryans]

So far, I am hearing that this only happens on nightly builds, so it seems like a bad interaction with Seshat's room crawling somehow. @poljar, would you happen to have any theories or ideas?

[poljar]

I looked at this already but concluded that it's fairly unlikely that Seshat is eating messages in unencrypted rooms as @bwindels rageshake points out so.

Now if it is indeed Seshat's fault it might have something to do that we have additional event listeners set up here.

I saw something similar happen if I would await the decryption promise in the onRoomTimeline handler instead of adding it to a set like here.

Though that happened only for encrypted rooms.

[jryans]

Hmm, I wonder if we may want to try / catch on the event emitter side (and possibly in each handler as well) of timeline events like that to ensure things can't go wrong...

[ara4n]

I haven't seen this in over a week. Either it was being triggered by Matrix.org being completely overloaded, or possibly a bad interaction with the /sync corruption we were seeing due to the replication rewrite, or some weirdness that has since gone away. I would deprioritise it for now.

[turt2live]

Sync corruption is matrix-org/synapse#7206 ftr

[turt2live]

see #10032 (comment)