fix: verify capacity from dual paginated active-task scans

This commit is contained in:
CoolBoy
2026-10-28 00:38:09 +08:00
parent 0ba4977e88
commit 5dead8172c
5 changed files with 148 additions and 53 deletions

View File

@@ -2,7 +2,23 @@
This repository is packaged for the ModelHub XC agent platform.
## Capacity refresh fix 2026.10.27.1
## Capacity pagination verification 2026.10.28.1
Supersedes the 2026.10.27.1 policy of using the larger reported total.
The platform can return inflated totals, incorrect page counts, and short pages.
Capacity now reads waiting/running pages through an explicit empty page, ignores
total/pages for termination, and deduplicates task IDs. It repeats the scan with
page sizes 100 and 50; only matching active ID sets confirm occupancy. Reported
totals are diagnostic only. Repeated pages, malformed records, request failures,
changing snapshots or the 100-page-per-status budget fail closed for that account.
This costs additional read requests but scans active tasks, not full history.
Two agreeing reads are not an atomic snapshot or a guarantee against a platform
omission shared by both queries; server submission limits remain authoritative.
Logs report `method=dual_page_scan`, `confirmed_active`, and safe failure reasons.
No age cleanup, routing, success-rate gates, recent-five-slot reserve, or weekly
capacity exploration policy changes. Existing tasks are not cancelled.
## Previous capacity refresh fix 2026.10.27.1
All task-page requests are clamped to the platform's 100-record limit, including
community sampling. Normal and weekly capacity refreshes now use the same