[GH-ISSUE #1358] DATABASE_USER_ACTIVE_STATUS_UPDATE_INTERVAL env var value recommendation is wrong #164

New issue

Closed

opened 2026-10-18 14:09:33 +00:00 by bsakaguchi · 1 comment

bsakaguchi commented

2026-10-18 14:09:33 +00:00

Owner

Originally created by @dgarciaf18 on GitHub (Aug 25, 2026).
Original GitHub issue: https://github.com/open-webui/docs/issues/1358

Clearly 300s (5min) is not inside the 3min (180s) active window. I've been monitoring the webui_users_active OTEL metric and setting DATABASE_USER_ACTIVE_STATUS_UPDATE_INTERVAL > 180 seems to be making the metric not accurate. I configured it to 300 and the result was a sawtooth pattern with 5min interval.

Originally created by @dgarciaf18 on GitHub (Aug 25, 2026). Original GitHub issue: https://github.com/open-webui/docs/issues/1358 Here: [**Recommended:** a few hundred seconds, e.g. `300` to `500`. Presence accuracy stays within the interval, which is well inside the 3-minute active window.](https://github.com/open-webui/docs/blob/main/docs/reference/env-configuration.mdx?plain=1#L7461) Clearly 300s (5min) is not inside the 3min (180s) active window. I've been monitoring the webui_users_active OTEL metric and setting DATABASE_USER_ACTIVE_STATUS_UPDATE_INTERVAL > 180 seems to be making the metric not accurate. I configured it to 300 and the result was a sawtooth pattern with 5min interval.

bsakaguchi closed this issue

2026-10-18 14:09:33 +00:00

bsakaguchi commented

2026-10-18 14:09:34 +00:00

Author

Owner

@Classic298 commented on GitHub (Aug 25, 2026):

You're right, thanks for measuring it. The active-user count uses a fixed 180 second window, so 300 puts the interval outside it: a user ages out before the next write is allowed, then reappears when it lands, which is the sawtooth you saw.

The recommendation was wrong and is now 60 to 120 seconds across the reference, the scaling guide, the performance page and the multi-replica guide, each stating that the value has to stay below 180 and why.

@Classic298 commented on GitHub (Aug 25, 2026): You're right, thanks for measuring it. The active-user count uses a fixed 180 second window, so 300 puts the interval outside it: a user ages out before the next write is allowed, then reappears when it lands, which is the sawtooth you saw. The recommendation was wrong and is now 60 to 120 seconds across the reference, the scaling guide, the performance page and the multi-replica guide, each stating that the value has to stay below 180 and why.

bsakaguchi referenced this issue

2026-10-18 14:10:55 +00:00

[PR #164] [MERGED] fix typo; add more deets #294