Profiling the container end to end found one allocation site responsible for
most of the peak: a single 46.3 MiB JSON body, assembled from 3,946 socket
chunks, inside undici's fullyReadBody. That function holds the chunk list and
the joined copy at the same instant, so a body of size N costs about 2N.
The source was two Umami calls asking for pageSize=100000, with
getMultiEventTimeSeriesAsync running them for every tracked event name
concurrently via Promise.all. getEventNamesAsync was downloading every raw
event of the last 30 days purely to collect a handful of distinct name strings.
- getEventNamesAsync now reads the aggregated /metrics?type=event endpoint.
- getWebsiteEventTimeSeriesAsync pages 5,000 records at a time, folding each
page into the bucket map before requesting the next.
- getMultiEventTimeSeriesAsync runs series sequentially.
Instances that return aggregated {x,y} buckets still answer in one request, an
instance that accepts pageSize but ignores page is detected rather than
double-counted, and hitting the page ceiling logs instead of silently
undercounting.
Alternated runs, 2 per side, on a restored real backup:
peak container 498.9 -> 398.5 MiB (-20%)
peak node RSS 452.4 -> 353.3 MiB (-22%)
peak during 4 tabs 498.5 -> 356.3 MiB (-29%)
peak arrayBuffers 95.8 -> 6.6 MiB (-93%)
largest allocation 46.4 -> 2.4 MiB (-95%)
Costs +4% CPU from the extra round trips; board load latency unchanged.
Post-soak steady state and boot idle are unchanged within run-to-run spread.
Adds the attribution tooling this needed: attribute-memory.mts reconciles
cgroup -> processes -> address-space mappings -> V8 spaces -> loaded code,
memory-probe.cjs is an injectable diagnostics preload (never shipped) that
tracks buffer and fetch allocation sites, analyze-heap-dominators.mjs computes
real retained sizes, and compare-stress-runs.mjs reports alternated runs with
their spread.
Also corrects an earlier claim in PHASE2-FINDINGS.md: the reported "7.86 MiB of
duplicate chunk sources" was an artefact of V8 truncating string contents to
~1 KiB in heap snapshots. There are zero byte-identical duplicates among all
3,055 server JS files. The real doubling is character width - one codepoint
above U+00FF forces a whole source string to two bytes per character, which
costs 9.45 MiB across 80 loaded files.
Adds 6 tests; this integration had none.
Merge origin/dev to pick up service isolation (concurrently dev script,
turbopack aliases, instrumentation simplification), PostHog analytics,
color scheme, and other dev improvements while preserving all Phase 1
performance work (Suspense boundaries, cache components, Redis handler,
integration barrel diet, cache invalidation).
- Fix integration data cache: check pending promise before TTL to
prevent returning undefined on concurrent cold-cache access
- Add concurrent-miss test (two parallel callers, one fetch)
- SVG sanitizer: extract <svg> from DOMPurify output to strip
residual HTML wrapper
- Narrow safeRevalidateTag catch: log unexpected errors, silence only
known "outside request scope" messages
- Run oxfmt across all packages (206 files formatted)
- Remove all concrete integration class re-exports from barrel
- Replace instanceof check with kind check in health-monitoring
- Document cache invalidation as preparatory (cacheTag follow-up)
* fix(immich): use thumbnail endpoint instead of original for album carousel
Switches from /api/assets/:id/original to /api/assets/:id/thumbnail to fix
HEIC image incompatibility (thumbnails are always JPEG), reduce bandwidth,
and improve loading speed.
Closes#6388
* working on immich
* Improve image proxy error handling
* fix(image-proxy): use trusted fetch response type
* docs(immich): require Asset View permission
The widget-only integrations refactor incorrectly removed the
ArchiveTeam Warrior integration entirely. It only needed to be
excluded from auto-seeding (by removing defaultUrl), not deleted.
This restores the integration class, reverts the widget/router/handler
to use integration middleware, and removes archiveTeamWarrior from the
widget-only-integrations migration.
* fix(beszel): harden cache layer and prevent blank widgets
- Add debug logging across Beszel integration, request handler, and tRPC
router (auth, fetch, SSE, per-procedure entry/exit) gated behind
LOG_LEVEL=debug
- cached-request-handler: retry with backoff, isValid predicate to skip
caching invalid data, in-flight dedup (single-flight per channel) to
avoid thundering herd after cache invalidation
- redis channel: set-then-publish so storage is always consistent even
when pubsub publish fails
- Beszel handlers: opt into fallbackToStaleOnError and retry so brief
upstream hiccups don't throw away good cached data
- beszel-system-stats: add refetchInterval matching each time period's
cache TTL so historical charts refresh in the background
- tRPC router: switch Promise.all -> Promise.allSettled for getSystems
and getAlerts; getSystemStats returns an error object instead of
throwing. Each result carries an optional error field so one failed
integration no longer nukes the whole widget
- Widgets: add explicit isPending loader, empty state for grid and
stats, and a shared per-integration error indicator (tooltip with
failed integration names). Alerts and table now only show their
empty state after loading completes
- Translation: add empty.noSystems key to beszelSystemGrid and
beszelSystemStats
Addresses #6030
* style: auto-fix lint and format
---------
Co-authored-by: ajnart <ajnart@users.noreply.github.com>
ajnart
Thomas Camlong
Bergasha
Matt Van Horn
Tim MooreandTim Moore
Bergasha
Josh PrestonandThomas
Anil K TiwariandThomas
Felix-AyushandThomas
Nerdy TechnicianandThomas
Thomas VervestandThomas
Afonso Jorge Ramos