Changeset 43273c9 in Klonkt

Timestamp:

07/01/2026 06:38:05 PM (2 months ago)

Author:

roboburr <roboburr@…>

Branches:

main

Children:

d18c60e

Parents:

dc41bef

Message:

refactor(fediverse): thread crawl one level per pass, like Mastodon

Match Mastodon's FetchReplies behaviour: crawl ONE hop of replies per pass instead
of a deep BFS. Deeper levels still fill in — a cached reply becomes a seed on the
next crawl, so its replies are pulled on a later view (Mastodon's per-status cascade).
More conservative + polite; same push-first + cache + bounded design.

• src/services/ActivityPubService.js — THREAD_MAX_DEPTH 3 → 1; comment updated.

Co-Authored-By: Claude <noreply@…>

File:

• src/services/ActivityPubService.js (modified) (1 diff)

src/services/ActivityPubService.js

@@ -1840 +1840 @@
 // Most replies reach us by delivery, but replies-to-replies that live on other servers and
 // aren't addressed to us are missed. This pulls the AS2 `replies` collections of the replies
-// we DO have, caching any newly-found ones in ap_interactions. Bounded (depth/fetch caps),
-// polite (serial), PULL only, and stale-while-revalidate: it never runs in a page request —
-// the view renders from cache; a stale post kicks off a background refresh for the NEXT view.
+// we DO have, caching any newly-found ones in ap_interactions.
+//
+// Matches Mastodon's behaviour: ONE level per crawl (like its FetchRepliesService), not a deep
+// recursive walk. Deeper levels fill in incrementally across crawls — once a fetched reply is
+// cached it becomes a seed itself, so its own replies are pulled on a later view (Mastodon's
+// per-status cascade). Bounded + polite (serial), PULL only, and stale-while-revalidate: it
+// never runs in a page request — the view renders from cache; a stale post kicks off a
+// background refresh for the NEXT view.
 const THREAD_TTL_MS = 15 * 60 * 1000;   // don't re-crawl a post more than ~4×/hour
-const THREAD_MAX_DEPTH = 3;             // replies-to-replies-to-replies
+const THREAD_MAX_DEPTH = 1;             // one hop per crawl (like Mastodon); deeper fills in over crawls
 const THREAD_MAX_FETCHES = 30;          // hard cap on remote GETs per crawl (be a good peer)
 const _crawlingThreads = new Set();     // per-post in-flight lock (no stampede across views)