Make a large library affordable: flat idle CPU, and a third of the memory - #39

Merged

vyrti merged 23 commits into

mainfrom

perf/large-library

Aug 14, 2026

Commits on Aug 14, 2026

refactor(bench): move the library generator into its own crate
vyrti
committed
77af7e1
perf(scan): read only the files that changed, and read them in parallel
vyrti
committed
85f1dc3
perf(scan): make the five-minute tick cost what changed, not what exists
vyrti
committed
8401d7c
perf(database): stop pruning on every batch and re-counting on every poll
vyrti
committed
20b079e
fix(lifecycle): stop compacting the database on every shutdown
vyrti
committed
69462bb
fix(lifecycle): announce the server before scanning the library
vyrti
committed
1c91eca
perf(database): drop two indexes nothing queries, and one pointless delete
vyrti
committed
1c64937
docs(config): say that the index cache is per connection
vyrti
committed
969fba1
perf(web): stop counting the whole result to render one page
vyrti
committed
4b022ea
perf(watcher): bound the file-id cache instead of indexing the library
vyrti
committed
97ba1ae
fix(config): stop the config watcher walking the tree it sits in
vyrti
committed
7bae0de
docs(config): correct what cache_mb costs and what it buys
vyrti
committed
27279bf
perf(config): default the index cache to 8 MB
vyrti
committed
3eaaf0f
perf(scan): count what a scan changed instead of keeping every record
vyrti
committed
2f527d5
perf(metrics): stop refreshing every process on the host
vyrti
committed
21bddca
u
vyrti
committed
c6460e1
u
vyrti
committed
87cf201
perf(scan): read the index a page at a time when validating deletions
vyrti
committed
0e461a1
style(test): drop a trailing blank line rustfmt rejects
vyrti
committed
dad71a2
refactor(scan): size the read window to the write batch
vyrti
committed
3c66c1f
style: satisfy clippy on the new scan and watcher code
vyrti
committed
380a2a3
fix(config): watch the config file itself, not only its directory
vyrti
committed
9f80080
u
vyrti
committed
7a1f88a

Navigation Menu

Uh oh!

Make a large library affordable: flat idle CPU, and a third of the memory - #39

Make a large library affordable: flat idle CPU, and a third of the memory#39
vyrti merged 23 commits into
mainfrom
perf/large-library

Commits on Aug 14, 2026

refactor(bench): move the library generator into its own crate

perf(scan): read only the files that changed, and read them in parallel

perf(scan): make the five-minute tick cost what changed, not what exists

perf(database): stop pruning on every batch and re-counting on every poll

fix(lifecycle): stop compacting the database on every shutdown

fix(lifecycle): announce the server before scanning the library

perf(database): drop two indexes nothing queries, and one pointless delete

docs(config): say that the index cache is per connection

perf(web): stop counting the whole result to render one page

perf(watcher): bound the file-id cache instead of indexing the library

fix(config): stop the config watcher walking the tree it sits in

docs(config): correct what cache_mb costs and what it buys

perf(config): default the index cache to 8 MB

perf(scan): count what a scan changed instead of keeping every record

perf(metrics): stop refreshing every process on the host

u

u

perf(scan): read the index a page at a time when validating deletions

style(test): drop a trailing blank line rustfmt rejects

refactor(scan): size the read window to the write batch

style: satisfy clippy on the new scan and watcher code

fix(config): watch the config file itself, not only its directory

u