The ETL re-reads the Omni model, re-embeds it, and ends by dropping every cached answer. It runs nightly at 07:00 UTC and no longer runs on deploy — so after a change to the Omni model, or to what the ETL indexes, the index holds the old shape until this button is pressed or 07:00 comes round. A run takes several minutes and is rate-limited against Omni on purpose; only one can be in flight at a time.
No run in progress.
| Started | Trigger | Status | Duration | Documents | Approved | Embedded | Pruned | Cache dropped | Detail |
|---|---|---|---|---|---|---|---|---|---|
| 7 hours ago | cron | success | 367.6 s | 6,496 | 465 | 6,961 | 0 | 1 | — |
| 1 day ago | cron | success | 366.8 s | 6,496 | 465 | 6,961 | 0 | 2 | — |
| 2 days ago | cron | success | 365.5 s | 6,496 | 465 | 6,961 | 0 | 7 | — |
| 3 days ago | cron | success | 372.3 s | 6,458 | 465 | 6,923 | 0 | 9 | — |
| 4 days ago | cron | success | 368.0 s | 6,458 | 465 | 6,923 | 0 | 0 | — |
| 5 days ago | cron | success | 367.7 s | 6,458 | 465 | 6,923 | 0 | 0 | — |
| 6 days ago | cron | success | 415.4 s | 6,458 | 426 | 6,884 | 0 | 10 | — |
| 7 days ago | cron | success | 358.9 s | 6,391 | 464 | 6,855 | 0 | 2 | — |
| 8 days ago | cron | success | 371.0 s | 6,379 | 464 | 6,843 | 0 | 5 | — |
| 9 days ago | cron | success | 368.3 s | 6,379 | 464 | 6,843 | 0 | 0 | — |
| 10 days ago | cron | success | 354.1 s | 6,337 | 453 | 6,790 | 0 | 4 | — |
| 11 days ago | cron | success | 351.5 s | 6,337 | 453 | 6,790 | 0 | 0 | — |
| 12 days ago | cron | success | 410.8 s | 6,337 | 403 | 6,740 | 0 | 1 | — |
| 13 days ago | cron | success | 359.6 s | 6,320 | 453 | 6,773 | 0 | 5 | — |
| 14 days ago | cron | success | 369.1 s | 6,320 | 453 | 6,773 | 0 | 1 | — |
| 15 days ago | cron | success | 364.4 s | 6,320 | 453 | 6,773 | 0 | 2 | — |
| 16 days ago | manual | success | 393.2 s | 6,320 | 453 | 6,773 | 0 | 5 | — |
empty is not a failure: Omni returned nothing to index, and the run still dropped the answer cache, because the warehouse refreshed on its own schedule either way. interrupted means the run was still marked running when a later one started, which proves its process died — a deploy or a crash — so its outcome is genuinely unknown. A failed run records no counts: it never committed the indexing transaction, so there is no figure that describes anything that reached the database.