Observability blind spots found in the 11-12 live investigation (item ledger, history retention, alerts)

オープン
#3,307 コメント 1 件 リアクション 0 件 担当者 0 名 GitHub で見る

メンテナーはふだん 1 日以内に返信

まだ誰も着手していません。

評価

難易度
5/5
見積もり時間
1週間以上
初心者へのやさしさ
12/100
issue の種類
機能追加
明瞭さ
おおむね明確
活発さ
活発
技術スタック
go, sqlite

調査の方向性

This is a multi-part design issue, not a single change. Start with the hourly bucket code behind npc_sale and monster_gold, then the economy_ledger writer and the history.sqlite pruner, and read #3302 for the census index. Done when each checklist box is a merged change or a recorded decision not to do it.

索引モデルが issue の本文から書いたものです。

説明

area: server enhancement priority: high

The 2026-11-12 live investigation (#3302, #3303 and the starter-refill issue) had to infer facts the server should record. Each line is data we lacked; fix them so the next investigation reads facts instead of joining timestamps.

Items

  • No item ledger. Gold has economy_ledger; items have nothing. A grant (starter kit, refill, quest, GM, mall, drop pickup), a merchant sale, a stall or exchange transfer and a destroy leave no row saying which item, how many, and through which door. A duplication cannot be proved or cleared after the fact without diffing nightly backups.
  • npc_sale and monster_gold fold into hourly buckets holding only total and n. The item sold is lost at once. Keep a per-character, per-item hourly bucket (item ref, count, gold).
  • stall_sale and exchange rows name no counterparty and no item. Finding who paid 1,000,000,000 to whom, and for what, took a self-join on at_ms. Record the counterparty character and the items on both sides.

Sessions

  • history.sqlite kept only about 90 minutes: two accounts wrote most of 200k events (every transport log line of every session becomes an event) and the pruner dropped everyone's older history. Budget events per account, and do not store per-session transport info lines as history events.
  • GameWorld and Agent logs live only in the Nomad alloc and vanish at the next release. Ship them to durable storage with a retention rule (they hold player names, so staff-only).
  • The Agent logs only refused admission tickets. Count minted tickets per account (a counter and a journal column), so a mint rate is visible.
  • The connection journal has no per-session summary of what the session did: frames in, opcodes used, time to first action. "Entered, sold, left in one second" was inferred from three stores.

Detection

  • Nothing alerted. account_session_evictions reached 74k, one account opened 3 sessions a second for a day, and 97% of the connection journal came from four accounts. The monitor needs rate alerts: sessions per account per minute, evictions per minute, entries per character per hour, merchant sales per entry.
  • A websocket session with no User-Agent is not a browser (every browser sends one). It is recorded and never surfaced. Show it in the console and count it.
  • The console has no "top accounts by session churn / by minted gold per hour" view; the hourly buckets already hold the gold half.
  • Every free grant (kit, refill, beta silk, starter silk) needs a counter per account per day, so a mint shows as a number that grows.

Performance of the tools themselves

  • The census index is quadratic in sessions per account (#3302), so the abuse it should expose degrades the server.

Done when each box is a merged change or a recorded decision not to do it.

主要言語
Go
スター
97
フォーク
61
平均マージ
4時間 14分
マージ済み PR(30日)
913

環境構築

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

opensro-dev/opensro のほかの issue

opensro-dev/opensro の issue をすべて見る

似ている issue

Go の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。