Jigasi vs Jicofo participant count mismatch (165 vs 29 for same instance)

[Jake-Chae]

I collected some additional data that shows a clear mismatch between how Jigasi and Jicofo count participants for the same instance.
From inside the Jigasi container:

docker exec -it curl -s http://127.0.0.1:8788/about/stats | jq '.participants, .conferences'

Example output:

165

6

So Jigasi’s own /about/stats endpoint reports 165 participants across 6 conferences.
At (almost) the same time, from the Jicofo host:

curl -s http://localhost:8888/debug | jq '.jigasi_detector'

Example output:

{
  "brewery_jid": "jigasibrewery@internal.auth.example.net",
  "jitsi-xxxxxx": {
    "supports_sip": false,
    "supports_transcription": true,
    "is_in_graceful_shutdown": false,
    "participants": 29,
    "region": "null"
  }
}

So Jicofo’s jigasi_detector sees the same Jigasi instance (the single entry under jigasi_detector) as having only 29 participants, while Jigasi itself reports 165 participants at the same time.
Given that:

• Jigasi is successfully joining the Jigasi brewery MUC (e.g. JigasiBrewery@internal.auth.example.net) and handling STT for multiple rooms (we see accepted jigasi dial IQs for those rooms in the Jicofo logs), and

• There are no obvious errors in Jicofo logs related to ColibriStatsExtension, SUPPORTS_TRANSCRIPTION, or PARTICIPANTS,

it seems that either:

• Jigasi is exposing one participant count on /about/stats but advertising a different count in the XMPP presence stats to the brewery, or

• Jicofo is interpreting the Jigasi stats in a way that results in jigasi_detector.participants being much lower than Jigasi’s own internal count.

Could you clarify what the intended semantics of jigasi_detector.*.participants are (for example: total participants handled by this Jigasi instance vs. only STT-enabled participants vs. something else), and whether this 165 vs 29 mismatch is expected or indicative of a bug?

If it helps, I can enable more verbose logging (Prosody/Jigasi/Jicofo) or capture example XMPP stanzas with the ColibriStatsExtension from the Jigasi brewery MUC to help pinpoint where the discrepancy comes from.

[damencho]

Is jigasi several times in the same conference?
You cannot make comparison between those two stats. Jicofo joins once per conference where jigasi is like a client it can join several times(more than one participant dial in) and every time it joins it will increase the number of conferences and participants it has connected to.