Fewer server round trips when showing posts and threads #234

New issue

Closed

opened 2026-10-21 19:08:24 +00:00 by littlefox · 1 comment

littlefox commented

2026-10-21 19:08:24 +00:00

Owner

Found in the codebase review (#230, D-058). Each round trip to lyskomd costs latency, and more over the relay. These are the places that multiply them:

Thread cards are loaded one at a time. The thread loader in app/src/main_view/mod.rs awaits kom.preview(t) for each card in turn. 100 cards at 30 ms take 3 s before the last one shows.
- Load them in small parallel batches, as Kom::thread already does for stats with BATCH. lyskomd penalises floods, so keep the batches small.
- Load the cards near the current post first.
A preview fetches a stat for every reply, just for the author's name. TextView::build calls kom.text_stat for each comment-to, footnote-to, commented-in and footnoted-in link. A post with 500 comments costs 500 get-text-stat calls to show its card, though the card doesn't show who wrote the replies.
- Let previews skip the link authors (a lighter preview), or fetch them lazily when the full post is opened.
Marking read is one call per text. Kom::mark_read sends one mark-as-read per text per conference, while the protocol takes a list. Reading by scrolling (#180, #228) marks many in a row: collect them for a short moment and send one call per conference.
session::who_is_on looks up every session's name with get-uconf-stat, bypassing the uconfs cache in Kom.
Hot paths in kom-client copy state:
- read_set clones the conference's RangeSet on every is_unread. Elisp users often have fragmented read ranges.
- advance clones the whole seen set on every batch, and seen grows with every post read in the session.
- Borrow these instead.

Measure with #225: calls per "Läs nästa" and per thread shown (count frames in a test client), and the time to show the last card.

Found in the codebase review (#230, D-058). Each round trip to lyskomd costs latency, and more over the relay. These are the places that multiply them: - **Thread cards are loaded one at a time.** The thread loader in `app/src/main_view/mod.rs` awaits `kom.preview(t)` for each card in turn. 100 cards at 30 ms take 3 s before the last one shows. - Load them in small parallel batches, as `Kom::thread` already does for stats with `BATCH`. lyskomd penalises floods, so keep the batches small. - Load the cards near the current post first. - **A preview fetches a stat for every reply, just for the author's name.** `TextView::build` calls `kom.text_stat` for each comment-to, footnote-to, commented-in and footnoted-in link. A post with 500 comments costs 500 `get-text-stat` calls to show its card, though the card doesn't show who wrote the replies. - Let previews skip the link authors (a lighter `preview`), or fetch them lazily when the full post is opened. - **Marking read is one call per text.** `Kom::mark_read` sends one `mark-as-read` per text per conference, while the protocol takes a list. Reading by scrolling (#180, #228) marks many in a row: collect them for a short moment and send one call per conference. - **`session::who_is_on`** looks up every session's name with `get-uconf-stat`, bypassing the `uconfs` cache in `Kom`. - **Hot paths in `kom-client` copy state:** - `read_set` clones the conference's `RangeSet` on every `is_unread`. Elisp users often have fragmented read ranges. - `advance` clones the whole `seen` set on every batch, and `seen` grows with every post read in the session. - Borrow these instead. Measure with #225: calls per "Läs nästa" and per thread shown (count frames in a test client), and the time to show the last card.

littlefox added this to the 0.6: Performance milestone

2026-10-21 19:08:24 +00:00

littlefox added the

area/protocol

area/reading

kind/bug

labels

2026-10-21 19:08:24 +00:00

littlefox referenced this issue

2026-10-21 19:08:41 +00:00

Jittery scrolling through unread posts #228

littlefox referenced this issue

2026-10-21 19:08:42 +00:00

Performance review of the whole codebase #230

littlefox referenced this issue

2026-10-21 19:53:18 +00:00

Channels as lists of threads, threads as chats (Teams-style) #242

littlefox referenced this issue

2026-10-21 22:17:08 +00:00

Inbox reads the stat of every text in the mailbox, again on every login anywhere #237

littlefox referenced this issue from a pull request that will close it,

2026-10-22 02:01:20 +00:00

Fewer round trips: light previews, cached names, cards a few at a time #261

littlefox closed this issue

2026-10-22 02:34:26 +00:00

littlefox referenced this issue from a commit

2026-10-22 02:34:26 +00:00

Fewer calls: previews without link authors, names from the cache

littlefox referenced this issue from a commit

2026-10-22 02:34:26 +00:00

Merge origin/main; thread cards and inbox previews a few at a time

littlefox commented

2026-10-22 04:04:13 +00:00

Author

Owner

Merged in #261; main run #437 green (#432 was the npm registry outage, #263). Verified: devserver tests, full e2e on three profiles (332 passed), and tools/perf.sh: inbox open 2.45 → 1.94 s; reading 120 cards by scrolling: layouts 35 → 18, main-thread 0.37 → 0.27 s. Marks were batched in #254.

Merged in #261; main run #437 green (#432 was the npm registry outage, #263). Verified: devserver tests, full e2e on three profiles (332 passed), and `tools/perf.sh`: inbox open 2.45 → 1.94 s; reading 120 cards by scrolling: layouts 35 → 18, main-thread 0.37 → 0.27 s. Marks were batched in #254.