Count archived history in the analysis warm-up, and read v_ views everywhere (#1809)

[erikdarlingdata]

Closes #1809.

The reported bug — confirmed exactly

AnalysisService.GetTotalDataSpanHoursAsync measured MIN..MAX(collection_time) on RAW wait_stats — the hot store only. The 512 MB threshold calls ArchiveAllAndResetAsync, which copies everything to Parquet and empties the hot store, so the measured span restarts at every reset. On a multi-server install that resets more often than daily, the Recommendations warm-up can never reach 24 hours — while the archived history sits in Parquet, readable through the v_ union views every other tab already uses. The reporter's option 1 (count archived + hot) is what shipped; no new knob.

Why it's a 23-site sweep, not a one-word fix

The broken gate was accidentally SHIELDING a second instance of the same defect. The fact collector read raw tables in 22 more places: the analysis-window reads (query stats/snapshots, memory, perfmon, file IO, blocking, deadlocks, storage, sessions, jobs, memory clerks/pressure, plan cache, CPU scheduler) and the on-load config snapshots (server/database config, trace flags, server properties) — which after a reset are EMPTY until the next app start. Fix only the gate and post-reset runs would analyze a thin hot window: the exact fraction-of-period distortion the 24-hour floor exists to prevent. So the gate and the reads move together — all 23 sites now read their v_ archive views.

Safety of the swap, verified not assumed: the views are column-identical to their tables by construction (only config_alert_log's view adds a source column, and analysis never reads it); the QUALIFY-dedup views make re-collected post-reset rows appear exactly once; there is no SELECT * anywhere in Lite/Analysis.

Guards

• Behavioral fixture (real DuckDB + Parquet, the ArchiveViewDedupTests harness): 28h archived + 2h hot across a simulated reset → span counts ~30h; a genuinely young install (2h, no archive) still gates.
• Catalog-driven sweep guard: for every table in ArchiveService.ArchivableTables, a raw FROM {table} anywhere in Lite/Analysis fails with the file named — a new collector's table is guarded the day it exists.
• Watched red: reverting the span query to the raw table fails BOTH the fixture test and the sweep guard.

Test plan

• New tests 3/3 green; red under mutation
• Full Lite.Tests: 1601/1601
• Lite builds, 0 warnings
• CI

🤖 Generated with Claude Code