Refresh issue with matterbridge

[mutmut]

SPOILER : everything ended (almost) well

Hello,
so the AI tells me there’s no thread about a refresh problem with matterbridge, so I’ll go ahead and post:slight_smile:
Yet I remember the tests we did on the Matter integration and forced refresh with decommissioning/recommissioning of matterbridge, weird.

Anyway, I have a refresh issue, here’s my scenario.
I recently installed a Somfy garage door motor. The motor not appearing (immediately?) in the Matter integration, I disabled the matterbridge-somfy plugin and re-enabled it.
Result: my Somfy devices changed their unique ID (which is normal) and then all I have to do is go to the Matter devices on Gladys to trigger their updates (with the refresh button).
And there, nothing

And much later (a few hours), the devices update appears:

I didn’t understand why this delay.

I retest today by disabling the matterbridge-dyson plugin and nothing happens on the Gladys side even though I’m convinced a red « unreachable » label should appear.

And if I look in the Matter nodes, my Dyson devices (24 and 25) are still present:

I disable Matter and re-enable it, matterbridge reacts well:

Nodes 24 and 25 have now disappeared:

And in the devices they are still there, without a label (the AI confirmed that no thread mentions this kind of « unreachable » label):

EDIT: I just found that famous badge: 🚀 Matter & Gladys Assistant : C’est parti! - #117 par pierre-gilles
AI is great but I must have bad karma because it never gives me a positive answer when I know it’s positive…

I continue my scenario by reactivating the matterbridge-dyson plugin … and then by forcing the refresh, it works (well I just did it live, the demo effect is still as surprising )

So from what I was able to retest, there seems to be a refresh problem when a change occurs on the matterbridge side, I have the impression that it’s especially when « removing ».
@pierre-gilles do you have a polling interval/check delay following your implementation?
We discussed it here.
And for the « disabled » badge, when should it appear?

[pierre-gilles]

@mutmut

@pierre-gilles do you have a check delay after your implementation?

No, actually there’s no longer any delay!

Now, as soon as you load the page, the list shows exactly the paired devices.

Several possible explanations for your observation:

• Either it’s a Gladys bug,

[mutmut]

@pierre-gilles

Several possible leads for your observation:

oh it’s going to be hard to find the culprit

@pierre-gilles

Either it’s a Gladys bug, but that surprises me because normally we just display the list that comes directly from the Matter.js library

I’m seeing odd behavior with page refreshes:

• I updated my 2 Dyson devices and then removed the filter: I ended up with all the Somfy devices in Add mode
• I refreshed the page: same
• I switched back to the Settings tab and the page tells me that Matter is not enabled then after a few seconds, it returned to normal
• I went back to the Devices tab and everything was normal again, no more unadded devices visible.

It might be my production Gladys install, I’ll test on my test Gladys.

[pierre-gilles]

@mutmut

I’m experiencing weird behavior with page refreshes:

If you have a way to reproduce it, I’d appreciate it — these kinds of small bugs, however, are indeed Gladys bugs

But in my opinion, it’s unrelated to a problem with devices not appearing in Gladys.

If you can monitor your Gladys logs when this kind of problem occurs, that could help.

[mutmut]

I just made a video, it’s real-time.

On the infrastructure side, I don’t see any lag on CPU/RAM/IOPS.

[pierre-gilles]

Your instance looks really slow What’s the machine behind it? What kind of disk?

Ok, I see the issue, it’s just that your instance is pretty slow and the loading states could be improved! I’ll make a note to look into it.

[mutmut]

On the SW side: Proxmox with 2 LXC Debian + Docker, one for Gladys and one for Matterbridge, all on a Ceph pool for HA storage.
On the HW side: i7-6700T with 16 GB of RAM, an NVMe where the system and the Ceph DB/WAL are, a 7200 rpm HDD for the Ceph pool (LXC).

[quote=« pierre-gilles, post:6, topic:9591 »]
Your instance seems really slow

[pierre-gilles]

That’s what I was thinking, it looks 100% like a disk problem! You need an SSD

[mutmut]

I just tested with my Gladys test instance, it’s a bit faster even though the infrastructure is the same.
I’ll send you the video and the Gladys log in a DM.

[pierre-gilles]

Thanks for the video and the logs @mutmut

I don’t see anything special in the logs, and the video doesn’t help me much — I’m having trouble understanding what is supposed to work or not

In any case, I think the first step in your situation is to switch back to fast, full-SSD storage. If you stay on HDD, you’re likely to have a lot of problems in the future. Gladys isn’t designed to run on slow storage and that can cause issues all over the place!

On my side, I’ve noted your feedback on the loading screens and I’ll improve all that

[pierre-gilles]

I am proposing a PR to improve loading states.

For example, when the « Devices » page is loaded but the list of devices paired via Matter is being refreshed, the page now displays:

I also improved the « Add a device » and « Settings » views to clearly show an indication that the page is loading.

The PR:

Matter: Improve loading states

[mutmut]

@pierre-gilles

In any case, I think the first step in your situation is to switch back to fast, all-SSD storage. If you stay on HDD, you’re likely to have a lot of problems in the future.

Done (enterprise SATA SSD) and it’s much snappier (actually like before)!

@pierre-gilles

I don’t see anything special in the logs, and the video doesn’t help me much, I’m having trouble understanding what is supposed to work or not

Argh
When Matterbridge is no longer running (shutdown for example), I don’t have any indication that the devices are disconnected on my side (but I see that you have the Disconnected badge in your screenshot), which would let me check whether the matterbridge docker is up or down or if there’s a matterbridge plugin issue. But in my case, I don’t have that visual feedback, even after clicking the Refresh button (top right of Devices).
I don’t know whether Matter has a ping or if it’s just info sent by the protocol at regular intervals, and since I don’t understand that I don’t understand how the rest works (or doesn’t work).

@pierre-gilles

I propose a PR to improve the loading states.

Great!

EDIT :
after about ten manual refreshes (over ~20s), I finally get the Disconnected badges (but not automatically):

EDIT 2 :
ok, it seems to work fine automatically because I do get the badges after I’d say 1 to 2 minutes when returning to the Devices page.
The HDDs probably didn’t help… I’ll keep an eye on all this over time.

[pierre-gilles]

@mutmut

after about a good ten manual refreshes (over 20s), I finally get the Disconnected badges (but not automatically) :

Ok, this is indeed a user understanding issue, there is no bug.

Refreshing data from Matter can take time, especially if you have offline devices, because the Matter.js library tries to recontact those devices, and the timeout is likely several tens of seconds, hence the wait.

The loading message at the top of this screen should help better understand what’s happening

[mutmut]

@pierre-gilles

timeout should be several dozen seconds, hence the wait.

Well, that’s the info I was missing Thanks!

[pierre-gilles]

I’ve improved the message:

[pierre-gilles]

Improvements to the loading state of the Matter integration are available in Gladys Assistant 4.61 :

Gladys Assistant 4.61 : Passage à Zigbee2mqtt 2.4.0

Salut à tous Je sors aujourd’hui Gladys Assistant 4.61, avec comme amélioration majeure le passage à Zigbee2MQTT 2.4.0 ! Zigbee2MQTT mis à jour vers la version 2.4.0 Pour une meilleure compatibilité avec les nouveaux périphériques Zigbee, Gladys Assistant passe à la nouvelle version majeure de Zigbee2MQTT. Si vous utilisez le conteneur Zigbee2MQTT lancé par Gladys, celui-ci sera automatiquement mis à jour vers la version 2.4.0, sans action de votre p…