What the documents look like
A brand livestream operation carries its own script, inventory and after-sales obligations, and its revenue does not depend mainly on creator commissions. The working material for an operations handover therefore sits in four recurring documents: the livestream script, the product explanation card, the shift schedule and the livestream review. Typical roles are the brand livestream team, the floor controller, the host and the paid-traffic specialist. Data sources include the Douyin Shop Compass, Juliang Qianchuan, livestream comments and transaction details, and the material is refreshed per session and hourly. That refresh cadence affects how a review document is time-stamped; it does not change any permission or handover judgement on this page.
Constraints at the upgrades step
Upgrading to v4.17.2 requires updating the fastgpt-app and fastgpt-pro image tags and running two manually triggered scripts: 20261008_migrate_chunk_training and 20261009_migrate_dataset_rebuild_status. The first requires that all old-version App and Pro instances in the cluster have been fully stopped and that old task leases have expired before it is clicked. The second depends on the first having completed. Both are manually triggered and do not block service startup. Since 4.15.0 the open-source edition no longer reads config.json; former fields must be moved to environment variables, and optional configurations that were never used require no additions. App and Admin must both set AIPROXY_API_ENDPOINT and a non-empty AIPROXY_API_TOKEN; a missing item in either place causes startup validation to fail.
Hypothetical drill sample: a shift schedule for a brand livestream team contains the sentence "Floor controller A covers the 10 March 20:00 session, with an authorization scope limited to explanation pacing and inventory announcements, and does not cover after-sales ticket handling." The same schedule contains a drafted handover request: "Host B requests to take over after-sales ticket handling permission after the 10 March session ends." The request carries no signed authorization record, and the schedule contains no field stating who owns after-sales permission.
Configuration and implementation checks
| Configuration item | Hypothetical starting value | Reason and comparison object |
|---|---|---|
AIPROXY_API_ENDPOINT | http://aiproxy:3000 | Documented as a root address with no /v1 suffix; compare against a single-model custom request address |
AIPROXY_API_TOKEN | A non-empty value entered for the drill | Documented as non-empty and consistent with the server; compare against an empty value |
AGENT_SANDBOX_OPENSANDBOX_VOLUME_NAME_PREFIX | Reuse the former VM_VOLUME_NAME_PREFIX value | Documented instruction to carry the old value across the upgrade; compare against a newly created prefix |
FILE_TOKEN_KEY | A random strong key entered for the drill | Documented minimum of 6 characters and a warning against example values |
INVOKE_TOKEN_SECRET | A random strong key entered for the drill | Documented minimum of 32 characters |
SYNC_INDEX | true | Documented behavior: when disabled, indexes must be maintained manually |
These rows are implementation checks tied to the handover work on this page, not a full deployment checklist. The starting values are drill inputs chosen for the hypothetical sample, not official defaults, and each is compared against the documented condition stated in the third column.
Three common errors
- Treating the drafted handover request in the shift schedule as a completed authorization. The request carries no signed record, and the schedule contains no after-sales permission field, so the request remains in an unsupported, pending-verification state.
- Treating a click on
20261008_migrate_chunk_trainingas a completed migration. The document describes the script as manually triggered and requires confirmation that old instances are stopped and leases expired before execution. - Inferring current environment-variable behavior from an old
config.jsonfield name. Since 4.15.0 the open-source edition no longer reads that file, and the field mapping must be checked against the official table rather than assumed.
How to verify the setup
- Input the two sentences from the shift schedule and the drafted request. Compare whether "authorization scope limited to explanation pacing and inventory announcements" and "take over after-sales ticket handling permission" refer to the same field. If the schedule contains no after-sales permission field, the request stays pending verification and is not judged as authorized.
- Input the two script names
20261008_migrate_chunk_trainingand20261009_migrate_dataset_rebuild_statustogether with the documented trigger conditions. Compare whether any old-version instance is still running in the cluster. If one is running, the precondition for executing either script is not met. - Input the two configuration items
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKEN. Compare whether both are non-empty and whether the endpoint avoids a/v1suffix. If either is empty or the endpoint carries/v1, startup validation fails, and the values must be corrected per the document before restart.
Each check names the input, the comparison and the observable result. A pending request is not an approval, a clicked script is not a completed migration, and an unsigned handover is not a transferred permission. Where the schedule is silent, the gap stays open rather than being filled by inference.