Profiling at the memory high-water mark - not at idle, which is what every earlier
round measured - found the largest single retainer in every snapshot was one V8
Context holding a 16.00 MiB ArrayBuffer. Expanding it through the dominator tree
named it: ssh2's crypto module, whose POLY1305_WASM_MODULE instantiates a
WebAssembly memory for ChaCha20-Poly1305 as soon as it is required.
Homarr has no SSH feature and does not depend on ssh2. The chain is
dockerode -> docker-modem -> ssh2, and docker-modem/lib/modem.js requires './ssh'
at module scope while using it in exactly one place, already guarded by
`protocol === 'ssh'`. A pnpm patch moves the require inside that branch. SSH-based
Docker hosts still work - require() is cached, so they pay it once - and every
socket-based install, which is everyone else, never loads it.
Measured over four alternated runs on a restored real backup, external memory:
before after
board loaded 25.4, 25.1 8.3, 11.2
nav-01 28.5, 25.4 10.1, 8.6
multi-tab 31.1, 35.8 13.0, 10.0
post-soak 21.4, 21.5 5.5, 7.2
Consistently -15 to -18 MiB, ranges non-overlapping, and 30 fewer files loaded
(1470 -> 1440).
Peak RSS does NOT improve, and that is worth stating plainly: V8 sizes its heap
against available headroom, so freeing native memory lets it grow further before
collecting. What changes is the composition - a fixed 16 MiB of unusable WASM
buffer becomes headroom the app can actually use, which matters under a container
memory limit but does not move the peak number.
Also adds the instruments this needed. A heap snapshot can never explain a peak,
because V8 runs a full GC before serialising one and reports only survivors:
- memory-probe.cjs gains in-process allocation sampling, on-demand snapshots via
v8.writeHeapSnapshot, an smaps endpoint, and require-graph recording.
- analyze-allocations.mjs diffs two stages to isolate what one stage allocated.
- analyze-smaps-stages.mjs splits each stage's RSS into JS heap, JIT, native and
file-backed pages.
- analyze-heap-dominators.mjs can now expand the largest retainer, which is what
identified ssh2. A plain graph walk escapes through __proto__ and is useless.
One trap recorded in the findings: taking the peak snapshot over CDP delivers it as
JS string chunks and added ~134 MiB of ArrayBuffers to the process being measured,
producing a fake 693 MiB reading. v8.writeHeapSnapshot streams from C++ instead, and
the probe now reports its own observerCostMiB so the cost of looking is visible.
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.
probe-fiber-tree.mts walks the live React tree and reports which components are
mounted, what DOM they render and their props. Heap snapshots only give minified
names like `y`, and source maps do not help because the snapshot records runtime
names — props shape plus rendered DOM identifies a component reliably. It injects
a DevTools hook stub before load so React registers every root; climbing from one
DOM node's fiber missed most of the tree (2,093 fibers vs 3,918).
What it found, on a 28-widget board:
- The most-mounted component is @mantine/core/Box (513). It is the primitive every
Mantine component renders through, so it is a symptom of how much is mounted
rather than a bug. Proportions match the 650 MB session snapshot (13% vs 17%),
so the minified `y` there is almost certainly Box.
- 71 Menu -> Popover -> PopoverDropdown triplets are mounted for context menus that
are almost never opened. Isolated with the existing enableRightClickOnWidgets
flag: 3,918 fibers with them, 3,554 without — 364 fibers, 9%, scaling with item
count. The dropdown contents are already lazy; the wrappers are not.
A fix that deferred the Menu until first right-click was reverted: it broke
right-click and inflated the tree to 20,107 fibers, because the replayed
contextmenu event re-entered the arming handler. Doing it properly needs one
board-level menu instead of one per item. The Playwright check that caught this is
in the findings — a fiber count alone would have read as a win.
Also records that the demo account had completed_board_tour = 0, so earlier rounds
measured the first-time-user path; completing it moves the tree by 38 fibers, so
the published numbers stand.
The container-only harness was measuring the wrong thing. Real usage — several
tabs, several boards, clicking into widgets — reaches ~880 MB server and a
650 MB *browser* heap, and the browser was never measured at all.
Profiling build: `docker build --build-arg HOMARR_PROFILING=true . -t
homarr:performance`. React's production bundle strips the hooks DevTools needs,
hence "Profiling not supported". Note reactProductionProfiling in next.config is
read only by the webpack path; under Turbopack it comes from `next build
--profile`, and turbo strips any env var not declared in the task's `env`, so all
three had to line up before the flag actually took effect — verified by markers
present only in React's profiling bundle, and static chunks going 20 -> 84 MB
with source maps.
inspect-container.mts attaches to an already-running container (SIGUSR1 opens
Node's inspector — no rebuild, no restart), reports cgroup/per-process/V8 memory
and can pull a heap snapshot out over CDP. The inspected context has no require
and no dynamic import, so it reads globals and uses CDP domains instead.
analyze-heapsnapshot / analyze-react-fibers / analyze-heap-retainers summarise a
snapshot too large for DevTools to open.
Harness now opens STRESS_BOARDS across STRESS_TABS_PER_BOARD tabs, clicks into
widgets, and records browser heap, DOM nodes and listener count via CDP.
Performance.getMetrics silently returns nothing unless the domain is enabled,
which is why the first run reported a 0 MiB client heap.
Two further wins, both measured on a restored 28-widget board behind a real
demo session (scripts/benchmarks/stress-restore.mts).
--max-semi-space-size=4 in run.sh. --max-old-space-size only bounds the old
generation; V8 also sizes the young generation generously (16 MB per
semi-space) and that slack is pure resident memory. 4 MB cut idle anonymous
memory 120.0 -> 97.2 MiB and peak 710.6 -> 560.2 MiB. It costs CPU, as any
GC-space reduction does — 15.8 -> 17.1 CPU-seconds over the run, ~8% — with
no board-load latency regression (median 594 -> 570 ms). Overridable via
HOMARR_MAX_SEMI_SPACE_SIZE.
serverExternalPackages for the remaining server-only dependencies. Bundling
a package inlines its transitive graph into a chunk that is compiled at boot
even when the code path never runs; leaving it external makes it a lazy
require. Adding winston, drizzle-orm, ical.js, jszip, ldapts, node-unifi,
@kubernetes/client-node and linkedom took idle to 90.3 MiB and board-loaded
209.6 -> 164.6 MiB, with slightly lower CPU (16.3s) and latency (549 ms).
The trade-off is image size: 388 -> 401 MB, because those packages now ship
in node_modules instead of being tree-shaken into chunks.
Cumulative vs the merge-base: idle 141.7 -> 90.3 MiB (-36%), board loaded
257.8 -> 164.6 MiB (-36%), under load 392.7 -> 302.5 MiB (-23%).
The harness now proves the session rather than inferring it: it asserts an
anonymous request to the private board is denied, requires the Auth.js
session cookie after sign-in, and reports whether it had to modify the
restored data at all (now "none"). It also records cgroup CPU seconds and
board-load latency, so a memory win bought with CPU cannot hide.
pnpm test 144/144 files, 1795 tests; typecheck and lint clean.
Two fixes found by running the harness against a real pair of images:
Login submitted before the form hydrated, so the click was a no-op that
looked identical to bad credentials. Load the page at networkidle and
retry the submit, surfacing any page alert if all attempts fail — the
same race already handled for the onboarding restore toggle.
memory.current counts reclaimable page cache, which moved the "same"
baseline between 192 and 261 MiB across runs. Anonymous memory is now
the headline metric (idle was 144 MiB anon vs 77 MiB of page cache), so
an A/B is not decided by unrelated disk I/O.
Sign-in redirects client-side, so page.waitForURL never reached its
implicit "load" state and every run failed after the restore step even
though authentication had succeeded. Poll window.location instead.
STRESS_NODE_OPTIONS injects NODE_OPTIONS directly, which is what makes
the V8 heap cap attributable: the merge-base run.sh does not read
HOMARR_MAX_OLD_SPACE_SIZE, so the cap can only be A/B'd against that
image by passing the flag in.
Container memory was only ever measured on an empty install, which hides
the cost that actually matters: a real board with real widgets and
integrations loaded.
stress-restore.mts boots a fresh container, restores a supplied backup
through the onboarding restore UI (including the "I understand"
confirmation), signs in, opens the busiest board, waits for every widget
to mount, then hammers navigation and soaks idle while sampling cgroup
v2 memory at each stage.
The backup is passed by path via STRESS_BACKUP_ZIP and never enters the
repo — it carries real integration secrets and an encryption key.
.gitignore now also refuses homarr-backup*.zip and loose *.sqlite.
Baseline reading (merge-base image, 28-item board, all widgets ready):
223 MiB idle -> 448 MiB board loaded -> 688 MiB peak.
Thomas Camlong