mirror of https://github.com/Wei-Shaw/sub2api.git synced 2026-10-06 14:33:10 +08:00

Files

T

IanShaw027 825f9c78d5 fix(channel-monitor-v2): default mode v1 and gentle adaptive backfill

Default channel_monitor_mode to v1 (opt-in V2) so upgrades keep active
probes; existing explicit v2 rows are left alone via ON CONFLICT DO NOTHING
plus migration checksum compatibility for already-applied 195.

V2 first-enable backfill no longer compresses ticks to 5s or uses 24h
chunks. Each tick does recent overlap plus at most one historical chunk
with depth-based ceilings (2h/4h/6h), adaptive grow/shrink, and failure
backoff. Error request_id dedup is bounded by a 90-minute lookback so
ops_error_logs is not scanned for full history.

Also align hide_throughput parse default with privacy-preserving public
runtime (missing key → true).

2026-08-08 13:59:11 +08:00

2.7 KiB

Raw Permalink Blame History

Channel Monitor V2 Safe Defaults & Gentle Backfill

Date: 2026-08-08
Status: Approved for implementation
Branch: fix/channel-monitor-v2-ops-ui-blockers (onto channel-monitor-v2)

Problem

H2 — Migration 195 and code defaults set channel_monitor_mode=v2, so every upgrade without an existing key auto-opts into passive V2, stops V1 probes, and may leave users on an empty V2 view.
M1 — First-enable bootstrap forces 5s ticks and ~24h RecomputeRange chunks until now-30d, hammering the primary DB.
Error dedup SQL matches request_id without a created_at bound, scanning ops_error_logs history via candidate_ids.

Decisions

Topic	Decision
Default mode	v1 (keep active probes). V2 is explicit opt-in.
Existing `channel_monitor_mode=v2` rows	Unchanged (`ON CONFLICT DO NOTHING` / no force rewrite).
Switch to v2	V1 runner `fire()` and `RunCheck` require `ActiveProbesAllowed()` → probes stop.
Backfill	Unified hard gates + adaptive soft gates (same rules, different observed load).
Optional profile knob	Not in this change (YAGNI).

Mode defaults

Migration 195_channel_monitor_mode.sql: insert 'v1' when key missing.
If 195 already applied with old checksum, add migration checksum compatibility (do not force rewrite applied DBs to v1).
Code: defaultChannelMonitorMode = v1; empty/invalid normalize → v1.
Frontend Settings form default and public feature flag fallback → v1 when missing/invalid.
Nil settings on V1 RunCheck path remains fail-closed (no probes) for test safety — independent of product default.

Gentle backfill (low-resource phases)

Hard gates (all servers)

Tick interval = refresh_interval (60 or 300s). No 5s bootstrap override.
Each tick: (1) recent overlap (~10m), then (2) at most one historical chunk while product bootstrap incomplete / retention walk ongoing.
Single leader lock; single short transaction budget (~55s).
Chunk ceiling by depth (product phases 90m → 1d → 7d → 30d → silent 90d).

Soft gates (adaptive)

Initial historical chunk: 1h (first seed still 2h for 90m UI).
Min chunk: 15m; max by depth: ≤1d → 2h, ≤7d → 4h, older → 6h.
Success + fast: grow slowly (×1.5) up to depth max.
Failure/timeout: halve chunk; exponential backoff on wait (cap 10m).
Progressive UI: show whatever is covered; bootstrap banner vs 30d product window.

Error dedup

Bound the request_id IN candidate_ids branch with
created_at >= $1 - INTERVAL '90 minutes' AND created_at < $2.

Out of scope

Force existing v2 → v1.
Separate backfill workers / read replicas.
Admin backfill_profile setting (may add later).

2.7 KiB Raw Permalink Blame History Unescape Escape