Hacktoberfest 2026：メンテナが10月に向けて印を付けた、オープンで初心者向けの issue。 Hacktoberfest の issue を見る

notification-acks: Track clients in memory and give up when they fall too far behind

オープン
  • 何か問題がありますか？
#892 コメント 1 件 リアクション 0 件 担当者 1 名 GitHub で見る

メンテナーはふだん 2 日以内に返信

@t-obr-iv がすでに取り組んでいます。

2026年10月18日 から。

評価

この issue はまだ評価されていません。

説明

backend by: ai-assisted frontend good-for: new-member spike
Exploring

scylla-server keeps an in-memory queue of numbered notifications for each client ID. The client reports the last number it received, and the server resends the rest. The server, not the socket, decides when a client is gone, and then drops its queue and subscriptions. A client is gone once its queue passes a size or age limit, regardless of time away or socket state.

Hypothesis

Tying "gone" to how much the server is holding keeps memory bounded under any outage pattern without picking a timeout, and a client that comes back within the limit loses nothing.

Context transfer
SUMMARY
#845: make sure a Rule's notification is actually received. The notification-acks spikes try different ways for the client to confirm what it got, so scylla-server can resend what it missed, for example after a Wi-Fi drop at the track or a laptop sleeping.

CONTEXT
- A Rule is a user-defined condition on a topic's values (e.g. a > 100) that notifies its subscribers when it matches. Logic in scylla-server/src/rule_structs.rs; a Rule fires after its debounce, then waits a fixed 5s cooldown.
- Delivery: handle_rule_processing in scylla-server/src/socket_handler.rs emits the rule_notify Socket.IO event (RuleNotification: id, topic, values, time) to the subscribing client, inline in the socket delivery loop. Nothing confirms the browser received it; a successful emit only means it was queued on the socket (buffer of 4096, shared with telemetry).
- RuleNotification.id is the Rule's ID, not unique per firing.
- If the subscribing client is not connected when a Rule fires, the notification is dropped and the client's subscriptions are deleted (socket_handler.rs, around lines 321-339). A failed emit is only logged.
- Clients identify with a client ID kept in localStorage (angular-client/src/services/socket.service.ts) and sent as the clientId query parameter when the socket connects (and as the x-client-id header on REST calls).
- Socket.IO is configured in scylla-server/src/main.rs: ping_timeout 5s, default ping_interval 25s, ack_timeout 1500ms. A silently dead connection can go unnoticed for up to about 30s; notifications sent in that window count as sent.
- Fired notifications are not stored on the server and there is no endpoint to fetch them. The client's notification log (angular-client/src/services/notification-log.service.ts) is in browser memory and cleared on reload.
- Server: socketioxide 0.18.7. Client: socket.io-client 4.7. socketioxide has no connection state recovery (Node Socket.IO's built-in replay of missed messages).
- Related: #801 moves rule processing off the delivery loop. Tests: scylla-server/tests/rule_structs_test.rs; nothing covers socket_handler.rs.
- Notifications carry no sequence number today, and the client does not send anything back when it receives one.
- Today the first notification that cannot be delivered deletes all of that client's subscriptions, plus any Rules left with no subscribers.
- The client reconnects automatically (socket.io-client default) and sends the same client ID each time it connects.
主要言語
TypeScript
スター
5
フォーク
1
平均マージ
2日 14時間
マージ済み PR（30日）
51

環境構築

  • Dockerfile・Docker Compose ファイルなし
  • プルリクエストのテンプレートあり
  • コントリビューションガイドなし

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。

Northeastern-Electric-Racing/Argos のほかの issue

Northeastern-Electric-Racing/Argos の issue をすべて見る

似ている issue

TypeScript 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。