# Pages (4 KiB) read per statement, cold cache (cache_size 2 MB), on a copy of hub.db (snapshot 2026-10-30 07:00Z,
# 572 MB) in tmpfs on DooPlex. Before = live schema; after = with the four R-1346 indexes / the INDEXED BY hint.
statement                                   before   after
NewestReportedControllerVersion (vouch)      35623      90
HostAgentVersionsSeen demo-hp (OS alarm)     38570     118
Prune: DELETE reports (in the write lock)    16538     309
Prune: DELETE host_reports (write lock)      13559     704
GetFleetAppSummary 7 d (apps page)           15404    1437
# Table sizes: host_reports 327 MB (19 928 rows), reports 139 MB (21 742 rows), app_telemetry 33 MB (123 115 rows).
# controller_version and agent_version sit AFTER report_json in their rows: reading them walks the JSON's overflow pages.
# VACUUM INTO (the 6-hourly snapshot) while 9 600 writes ran on the same copy: the slowest write 1.5 ms — the snapshot
# takes no write lock; its cost on the live hub is 3 min of disk load (570 MB read + 570 MB written on Longhorn).
# Hub pod cgroup 2026-10-30 09:3x: memory.current 268 013 568 of max 268 435 456 (anon 85 MB, file cache 177 MB);
# cpu.stat nr_throttled 45 419 of 85 460 periods.
