## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - Hosted instances can send browser page-load traces to Honeycomb (#21156, #21392) > - On staging, one hour in one workspace sent about 64,000 Honeycomb events, and only about 15 were `page.load` > - Two sources make the volume: nine network timing span events on each API request span, and request spans for background polling in idle tabs > - The page-load metrics use neither of them > - This pull request drops the timing events and records API request spans only while a page load is measured > - The benefit is about 30x fewer events in a browser test, with the same page-load data and the same server trace joins ## Linked Issues or Issue Description Refs #21156, Refs #21392 **What existing behavior does this improve?** Browser tracing exports a span for every same-origin `/api/` fetch and nine network timing events on each of those spans. **Subsystem affected** UI browser tracing (`ui/src/lib/browser-tracing/`), on hosted instances with browser tracing turned on. **Current behavior** Honeycomb stores each span event as its own event. Each request span becomes about 10 events. Idle tabs poll (for example live runs every 3 s on a task page), so most spans are requests that no page load made. On staging: about 6,600 `GET` spans plus about 6,400 of each timing event per hour, against about 15 `page.load` spans. **Proposed behavior** Request spans carry no timing events. A request is recorded only when it starts while a `page.load` measurement is open. Other requests send no span and no `traceparent`. **Reason and benefit** Event volume drops by about 30x for an idle tab. `page.load`, Web Vitals, and the Resource Timing backfill do not change. No exported request span lacks its `page.load` parent, and no server span points at a browser span that was never sent. ## What Changed - `sdk.ts`: the fetch instrumentation runs with `ignoreNetworkEvents: true` (an option of the installed `@opentelemetry/instrumentation-fetch` 0.221.0). - `sdk.ts`: new `measuredRequestsOnly` wraps the tracer provider that only the fetch instrumentation uses. When no `page.load` is open, it returns a non-recording span with an invalid context. Nothing is exported. The W3C propagator sends no `traceparent` for an invalid context, so the server traces that request alone, as for an unsampled tab. The `page.load`, Web Vitals, and backfill spans use the normal tracer and do not change. - Failed idle requests are not kept. The server's own request span records server errors. To keep them, the browser must record every idle request and send its `traceparent` before it knows the status, and that brings back the dangling joins. - `span-sanitizer.ts`: comment update only. - `doc/browser-performance-tracing.md`: new "Live API spans" section with the rule, the note that timing events are not sent, and updated volume and trace-header text. ## Verification - Red then green (`ui/src/lib/browser-tracing/sdk.test.ts`): - `records no span and sends no traceparent for an API request made while no page load is measured`: failed on the old code (an idle request sent `traceparent`), passes now. - `sends request spans without per-request network timing events`: failed on the old code (9 timing events on the request span), passes now. - All 56 tests in `ui/src/lib/browser-tracing/` pass. `tsc --noEmit` for `ui` passes. - `pnpm verify` exits 2 because a doc is a global CI input. CI supplies the broad coverage. - Real browser: local dev server on an isolated empty instance with 8 tasks, a stand-in OTLP collector, headless Chromium, and the tracing config injected into the session response. Scenario: hard load of the all-tasks page, click a task, then 2 minutes idle. One Honeycomb event = one span or one span event. | Name | master | this branch | | --- | ---: | ---: | | `page.load` | 2 | 2 | | `web_vital.lcp` | 1 | 1 | | `HTTP` (backfill) | 33 | 32 | | `GET` | 165 | 18 | | `POST` | 1 | 1 | | each of `fetchStart`, `domainLookupStart`, `domainLookupEnd`, `connectStart`, `connectEnd`, `requestStart`, `responseStart`, `responseEnd` | 165 | 0 | | **Total events** | **1,522** | **54** | - `secureConnectionStart` does not appear locally because the dev server is plain HTTP. On staging it is the ninth event. - Branch: the hard `page.load` has 32 `HTTP` children and the client `page.load` of the task has 19 live children (`GET`, `POST`). 0 request spans without a `page.load` parent (master: 146). 19 requests sent `traceparent`, and all 19 point at an exported span (master: 172 sent, 6 point at spans never exported). - Idle traffic in this test (210 `/api/` requests in total): `GET /api/issues/:id/live-runs` 88 times in 2 minutes, and `GET /api/health` 44 times (a dev-server-only 2 s poll). Most of the rest are one-off refetches. - Bundle (`vite build`): entry 3,826,982 B raw both, gzip 1,156,148 → 1,156,173 B (+25 B, chunk hash). Lazy SDK chunk 85,423 → 85,825 B raw, 27,723 → 27,859 B gzip. ## Risks - Low risk. The change is limited to the operator-gated browser tracing path on hosted instances. - Behavior shift: Honeycomb no longer has browser spans for requests made outside a page load, such as polling and mutations in a loaded page. Their server spans are still recorded as their own traces. - A request that starts during a page load and finishes after it is still recorded, as a child of that load. > For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and discuss it in `#dev` before opening the PR. Feature PRs that overlap with planned core work may need to be redirected — check the roadmap first. See `CONTRIBUTING.md`. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, via Claude Code with tool use (shell, file edits, headless Chromium). ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [ ] All Paperclip CI gates are green - [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
20 KiB
Browser Performance Tracing
The board UI can send real-user page-load measurements, Core Web Vitals, and API request spans to Honeycomb (or any OTLP/HTTP collector). The data answers questions such as "what is the P95 time to load the all-tasks page?" and can back a Honeycomb SLO with an owner.
This is part of the operator-gated Observability path (see
observability.md and rules/data-paths.mdc). It is not the
first-party Telemetry system and adds no Telemetry events.
Who gets it
Browser tracing runs only on hosted Paperclip Cloud instances that an operator explicitly configures. It is off by default.
The server decides, in server/src/browser-tracing-config.ts. It sends a
browserTracing object in the authenticated GET /api/auth/get-session
response only when both hold:
- The instance is cloud managed (
isCloudManagedInstance: aPAPERCLIP_CLOUD_TENANT_SERVER_TOKENor aPAPERCLIP_MANAGED_CONFIGdocument is present). PAPERCLIP_BROWSER_TRACES_ENDPOINTis set and every setting is valid.
Otherwise the field is null. Self-hosted installs, local development, and
tests never satisfy (1). With null, or with no session (signed out), the
browser never downloads the tracing code: it lives in its own lazily loaded
chunk (ui/src/lib/browser-tracing/sdk.ts). An invalid setting fails closed:
the field is null and the server logs one warning naming the variable, never
its value.
Configuration
Set these on the hosted server process:
| Variable | Required | Default | Meaning |
|---|---|---|---|
PAPERCLIP_BROWSER_TRACES_ENDPOINT |
yes | — | OTLP/HTTP base URL, for example https://api.honeycomb.io (US) or https://api.eu1.honeycomb.io (EU). /v1/traces is appended unless the path already ends with it. Must be https (plain http only for a loopback host). |
PAPERCLIP_BROWSER_TRACES_INGEST_KEY |
for Honeycomb | — | A Honeycomb ingest key (prefix hcaik_). Sent as x-honeycomb-team. Required when the endpoint is a Honeycomb host; optional for a collector you run. Any other key format is refused. |
PAPERCLIP_BROWSER_TRACES_SAMPLE_RATE |
no | 1 |
Fraction of page sessions traced, 0–1, rounded to an exact 1 in N (0.3 becomes 1/3; 0.75 becomes 1). 0 turns tracing off. |
PAPERCLIP_BROWSER_TRACES_SERVICE_NAME |
no | paperclip-web |
service.name; in Honeycomb this is the dataset name. |
PAPERCLIP_BROWSER_TRACES_ENVIRONMENT |
no | SENTRY_ENVIRONMENT |
deployment.environment.name, for example staging. |
Key handling
Every signed-in browser can read the key, the same way it can read the
front-end Sentry DSN. That is why only a Honeycomb ingest key is accepted:
it can send events but cannot read data or change the Honeycomb account. Use a
key scoped to the environment that should receive browser data, keep it in the
hosting secret store, and never commit it. To keep the key out of browsers
entirely, point the endpoint at an OTLP collector you run (it adds the key) and
leave PAPERCLIP_BROWSER_TRACES_INGEST_KEY unset. That collector must allow
CORS POST from the instance origin with the content-type header.
Sampling
The sample decision is made once per page session (one tab load), before the
SDK downloads. An unsampled session downloads and sends nothing. A sampled
session sends everything, so each trace is complete. When the rate is below 1,
every span carries Honeycomb's SampleRate attribute (1 in N, for example 4
for 0.25), so Honeycomb's COUNT and percentiles are weighted correctly.
Volume per sampled session: one page.load span per navigation, up to three
web_vital.* spans per page lifetime, and one span per same-origin /api/
request that starts while a navigation is being measured (see
Live API spans). Requests made while no navigation is being
measured, such as background polling in an idle tab, send nothing, and request
spans carry no span events.
Load-speed cost
Nothing loads on the critical path. The main bundle carries only the
measurement code (route matching, readiness checks); on the ui build of this
change the initial JavaScript grew by about 6.8 KB raw / 2.6 KB gzip (0.2%). The
OpenTelemetry SDK, a small OTLP JSON exporter, fetch instrumentation, and
web-vitals are a separate chunk (about 84 KB raw / 27 KB gzip) fetched only when tracing is on
and the session is sampled, and only after the page settles (painted, no query
fetching, browser idle; ui/src/lib/when-page-settles.ts). If the page never
settles (for example a request hangs), the chunk loads 35 seconds after tracing
starts anyway, so a hard load that ended as timeout is still exported.
Because the SDK loads after the first page settles, it cannot trace the hard
load's API requests live. Instead the main-bundle controller registers one
PerformanceObserver for resource entries and keeps only this
origin's /api/ fetch/XHR entries, at most 500. buffered: true delivers
the requests that finished before tracing started (it starts once the session
request returns). After that, the observer receives every entry even when the
browser's resource buffer (250 entries by default) is full; only entries that
overflowed that buffer before tracing started are missed. When the SDK loads it turns
those entries into child spans of the page.load they belong to (see
Backfilled API spans). The observer stops and the
entries are dropped once no navigation that started before the SDK loaded can
still claim them: every such navigation has ended, and either a navigation
that started after the SDK loaded has begun or 60 seconds have passed (a
navigation whose history change came before the SDK can commit after it).
Measurements taken before the SDK loads are buffered and exported when it
arrives.
What "loaded" means
Each navigation produces one page.load span. It ends at the first animation
frame, held for two consecutive frames, in which:
- no full-page loading state is mounted (the
PaperclipLoadingspinner, which is also the routeSuspensefallback), and - one of:
- Content signal (
ready_signal = content): when the page declared a primary-content signal withusePageContentReady, every declared signal is ready. The end time is the paint time the page reported. - Idle signal (
ready_signal = idle): otherwise, no React Query query that has no data yet is fetching. Background refetches of data already on screen do not count. The end time is the first such frame.
- Content signal (
The rule lives in one place (ui/src/lib/browser-tracing/page-load-recorder.ts)
and the route comes from the router's own route tree (App.tsx passes the tree
it renders to PageLoadTracker), so a new page is measured without page code.
- Hard load (
load_type = hard): starts at navigation start (the document request, time 0) and includes HTML, JavaScript download and start up, authentication, and the page's data and render. Redirects during the load (for example/to/:companyPrefix/dashboard) keep the same span, reported under the final route. - Client navigation (
load_type = client): starts when the URL path changes in history (a link click,navigate(), or back/forward) and includes lazy route code download, data loading, and render. Query-string and hash changes are not navigations. A history replace while a navigation is still open is a redirect and continues it.
Outcomes:
ready: loaded, as defined above.interrupted: another navigation started first.hidden: the tab was hidden when the navigation started (a link opened in a background tab) or became hidden before the page was ready. The span ends at the moment the tab hid, so a load in a background tab has a duration of about 0. Browsers do not paint hidden tabs, so such a load could never become ready; this is the same ruleweb-vitalsapplies to LCP and FCP. For a hard load, Chromium'svisibility-stateperformance entries give the time the tab first hid, even before tracing started; elsewhere a tab that is hidden when tracing starts counts as hidden from navigation start. A redirect while the tab is still hidden renames the hidden span rather than starting another one.was_hiddenistrueon these spans.timeout: not ready 30 seconds after it started. The span always ends at exactly start + 30 s, however late the browser ran the timer. Any end observed after that deadline (for example a page that only became ready after the computer woke up) is also reported as this timeout.
paperclip.page.suspended = true marks a span during which the computer slept
or the tab was frozen: the observed end was more than 5 seconds past the
30 s deadline, or two consecutive animation frames were more than 5 seconds
apart. Its duration does not describe the page; exclude it like hidden. A
suspend that ends before the next frame is not detectable, but it can never
stretch a span beyond 30 s.
A navigation open when the tab closes is not sent. For load-time percentiles
and SLOs, filter on outcome = ready; that excludes hidden tabs, timeouts and
interruptions, which you count separately.
Pages with an explicit content signal:
| Page | Route | Loaded when |
|---|---|---|
| Task (issue) page | /:companyPrefix/issues/:issueId |
The conversation content has painted: the same point as the issue-detail:navigate→content-paint User Timing measure. |
| All-tasks list (plain task list) | /:companyPrefix/issues |
The first page of task rows has rendered (or the list shows its error state). The combined inbox and coordinator task views use the idle signal. |
An agent chat with no saved task has no content signal and uses the idle rule.
Page views
Some pages show a different list depending on ?view=. For those,
paperclip.page.view records the view (ui/src/lib/browser-tracing/page-view.ts),
read from the URL when the span ends, so it is the view the page resolved and
wrote into the URL. Only names from the code's view registries are sent; any
other value is other.
| Route | Values |
|---|---|
/:companyPrefix/issues |
The keys of TASK_VIEW_KEYS in ui/src/lib/task-views.ts (mine, unread, blocked, recent, everything, needs_attention, all, active, backlog, done), default when the URL has no view yet, else other |
/:companyPrefix/decisions |
ATTENTION_PERSONAL_VIEWS (mine, unassigned, company, archived, all), mine when the URL has no view, else other |
Switching views on the same page changes only the query string, so it is not a
new navigation and has no page.load span; the attribute describes the view a
navigation landed on.
To add one, call usePageContentReady(ready) in the page component. Use it
only when the idle rule is inaccurate for that page.
Emitted spans
All spans pass through an allowlist (ui/src/lib/browser-tracing/span-sanitizer.ts)
before export. Any other attribute an instrumentation adds, including full
URLs, server addresses, status messages, and event attributes, is dropped.
| Span name | When |
|---|---|
page.load |
One per hard load or client navigation. |
web_vital.lcp, web_vital.inp, web_vital.cls |
Core Web Vitals, from Google's web-vitals, usually reported as the tab hides. Zero duration; the value is an attribute. |
GET, POST, … |
A fetch to this origin's /api/ that starts while a page.load span is open, as its child. Its traceparent header joins the server's request span to the same trace, so one trace shows the navigation, its API calls, and (when server tracing is on) their server spans. See below. |
HTTP |
A request to this origin's /api/ made before the SDK loaded, during a measured navigation (in practice the hard load), rebuilt from Resource Timing. See below. |
Live API spans
The rule: a fetch to this origin's /api/ is recorded only when it starts
while a page.load measurement is open. It is a child of that page.load
and sends a traceparent, even if it finishes after the navigation ends.
A request that starts while no measurement is open (background polling, a
refetch after a live update, a mutation in a page that already loaded) gets no
span and no traceparent. The fetch instrumentation receives a non-recording
span with an invalid trace context for it, so nothing is exported and the
standard trace context propagator sends no header
(measuredRequestsOnly in ui/src/lib/browser-tracing/sdk.ts). The server
traces such a request on its own, exactly as it does for a tab that is not
sampled. There is therefore no browser span without a page.load parent and
no server span that points at a browser span that was never sent.
Failed idle requests are not kept either: the server's own request span
already records a server error, and keeping them would mean recording every
idle request and sending its traceparent before its status is known.
Request spans carry no span events. The fetch instrumentation runs with
ignoreNetworkEvents: true, so the per-request Resource Timing events
(fetchStart, domainLookupStart, domainLookupEnd, connectStart,
secureConnectionStart, connectEnd, requestStart, responseStart,
responseEnd) are not sent. Honeycomb stores each span event as its own
event, so they multiplied request volume by about 10, and the span's own
duration already covers the request.
Backfilled API spans
The hard load's API requests happen before the SDK exists. When the SDK loads,
it creates one HTTP child span under that load's page.load for each
same-origin /api/ fetch/XHR Resource Timing entry that started inside the
load's window (and before the SDK started tracing requests live, so nothing is
counted twice). Each runs from the entry's startTime to its responseEnd.
They differ from live fetch spans in three ways:
- The name is
HTTP, OpenTelemetry's name for a request whose method is unknown: Resource Timing does not expose the method, so there is nohttp.request.method. http.response.status_codeis present only where the browser exposesresponseStatus(Chromium); it is never guessed.- No
traceparentwas sent with these requests, so they do not join the server's spans. Usepaperclip.api.areaand the timings; for the server side, query server spans by time.
They carry paperclip.api.timing_source = resource_timing. The URL is reduced
to paperclip.api.area by the same sanitizer as live fetch spans.
Attributes
Attributes (the complete allowlist):
| Attribute | On | Values |
|---|---|---|
paperclip.page.route |
all | Route pattern, for example /:companyPrefix/issues/:issueId; (unmatched) when no route matches. Built only from route definitions, never from the URL. On fetch spans: the route current when the request started. On web_vital.lcp: the hard load's route. |
paperclip.page.load_type |
page.load |
hard, client |
paperclip.page.outcome |
page.load |
ready, interrupted, hidden, timeout |
paperclip.page.ready_signal |
page.load |
content, idle, none |
paperclip.page.was_hidden |
page.load |
boolean |
paperclip.page.suspended |
page.load |
true when the computer slept or the tab froze during the measurement; absent otherwise |
paperclip.page.view |
page.load on pages with views |
The view from ?view=, from a fixed list per page, else other; absent on other pages |
paperclip.page.navigation_type |
hard page.load |
navigate, reload, back_forward, prerender |
paperclip.page.ttfb_ms, paperclip.page.dom_content_loaded_ms, paperclip.page.load_event_ms |
hard page.load |
Navigation Timing milestones, ms from navigation start |
paperclip.web_vital.name |
web_vital.* |
LCP, INP, CLS |
paperclip.web_vital.value |
web_vital.* |
ms for LCP and INP; unitless for CLS |
paperclip.web_vital.rating |
web_vital.* |
good, needs-improvement, poor |
paperclip.api.area |
fetch | First path segment after /api/ when it is a fixed word (issues, companies, …), else other. Use the joined server span's route for detail. |
paperclip.api.timing_source |
backfilled HTTP |
resource_timing |
http.request.method, http.response.status_code, error.type |
fetch | OpenTelemetry HTTP semantics (backfilled spans: status only, where the browser exposes it) |
SampleRate |
all, when sampling | 1-in-N weight |
Resource: service.name, service.version (the UI build commit),
deployment.environment.name.
Not sent: URLs, paths with ids, query strings, hashes, referrers, user agent, user, company, or task identifiers, titles, or content. Server spans carry no company identifier, so browser spans do not either. Uncaught browser errors are not sent here because Sentry already reports them (observability.md).
Trace headers go only to this origin (propagateTraceHeaderCorsUrls is empty),
and only on requests that start during a measured navigation, so no third
party receives a traceparent. Same-origin requests need no CORS
change. The server's OpenTelemetry HTTP instrumentation continues the trace
when server tracing is enabled (observability.md).
Example Honeycomb queries
Use the paperclip-web dataset (or your service.name) and the environment
that holds browser data. Load-time percentiles filter on outcome = ready,
which leaves out hidden tabs, timeouts, and interruptions. Also leave out
suspended spans: paperclip.page.suspended is only ever set to true, so
filter on it not existing.
P95 load time of the all-tasks page, by view
VISUALIZE P95(duration_ms), COUNT
WHERE name = page.load
paperclip.page.route = /:companyPrefix/issues
paperclip.page.outcome = ready
paperclip.page.suspended does-not-exist
GROUP BY paperclip.page.load_type, paperclip.page.view
P95 load time of an individual task page
VISUALIZE P95(duration_ms), COUNT
WHERE name = page.load
paperclip.page.route = /:companyPrefix/issues/:issueId
paperclip.page.outcome = ready
paperclip.page.suspended does-not-exist
GROUP BY paperclip.page.load_type
API requests during hard loads of the all-tasks page
VISUALIZE P95(duration_ms), COUNT
WHERE paperclip.api.timing_source = resource_timing
paperclip.page.route = /:companyPrefix/issues
GROUP BY paperclip.api.area
Group by service.version to compare builds. For a timeout rate, count
paperclip.page.outcome = timeout against page.load spans for the route
whose outcome is not hidden.
For Web Vitals, P75(paperclip.web_vital.value) where
name = web_vital.lcp (or inp, cls), grouped by paperclip.page.route.
SLO. A Honeycomb SLO needs a boolean SLI as a derived column. For example, "task page loads within 2 s":
IF(AND(EQUALS($name, "page.load"),
EQUALS($paperclip.page.route, "/:companyPrefix/issues/:issueId"),
IN($paperclip.page.outcome, "ready", "timeout"),
NOT(EQUALS($paperclip.page.suspended, true))),
AND(EQUALS($paperclip.page.outcome, "ready"), LTE($duration_ms, 2000)))
The expression is null for other events (including hidden and
interrupted loads), so they do not count; a timeout counts as a failure.
Verifying a deployment
- Set the variables above and deploy. A cloud-managed instance logs nothing on
success; an invalid setting logs one
browser tracing is offwarning. - Sign in, open the browser network panel, and confirm the session response
carries
browserTracing, asdk-*.jschunk loads after the page settles, and aPOSTto…/v1/tracesreturns 200. - In Honeycomb, query
name = page.loadin the browser dataset for the last 10 minutes and checkpaperclip.page.routeandservice.version.