Converse.js: Large rooms will eat RAM until killed by the OS

[licaon-kter]

Eg.

Steps? Join xmpp:#matrix#[email protected]?join ~5k users

FYI, Conversations (at least 2.4.0 or later) and apparently Yaxim handle these ok.

[darnuria]

I made a performance snapshot in firefox 67.0a1 (2019-03-06) (64-bit) before it become unusable.

Here the link to the performance snapshot (1.3 Mb compressed 8.1Mb uncompressed) it can be loaded in performance tab of firefox devtools.
https://framadrop.org/r/yXGF9VPBpl#zQRp+XTfDtkvPrCeTtLRaRkVDeoQMCGzQCumtd3Hya8=

[jcbrand]

@licaon-kter, @darnuria: looks like bridge.xmpp.matrix.org is no longer running, do you have a different bridge I could try?

[licaon-kter]

None that I know unfortunately.

[poVoq]

"muc_fetch_members: false" setting might be a temporary solution, but it seems like this isn't respected in ConverseJS version 5.0.5.

[licaon-kter]

The bridge is back, new syntax: xmpp:#matrix#[email protected]?join

Beware, use a disposable account as https://github.com/matrix-org/matrix-bifrost/issues/53 is not yet fixed

[licaon-kter]

Looks like on HEAD it's not leaking any more (at least with the current ~2.2k users), it did jump to 1.5Gb RAM for Firefox, but after the roster was loaded is backed down to 500Mb.

Good job @jcbrand

[jcbrand]

Thanks, I'm closing this then.