-
-
Notifications
You must be signed in to change notification settings - Fork 1.7k
feat(browser): Add ElementTiming instrumentation and spans #16589
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
7680005 to
5379c13
Compare
size-limit report 📦
|
| // - `renderTime` if available (available for all entries, except 3rd party images, but these should be covered by `loadTime`, 0 otherwise) | ||
| // - `timestampInSeconds()` as a safeguard | ||
| // see https://developer.mozilla.org/en-US/docs/Web/API/PerformanceElementTiming/renderTime#cross-origin_image_render_time | ||
| const { spanStartTime, spanStartTimeSource } = loadTime |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
this can be a helper or return tuples, will help reduce bundle size of spanStartTime and spanStartTimeSource being un-minifiable.
| const activeSpan = getActiveSpan(); | ||
| const rootSpan = activeSpan ? getRootSpan(activeSpan) : undefined; | ||
| const route = rootSpan ? spanToJSON(rootSpan).description : getCurrentScope().getScopeData().transactionName; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
we can move these outside of the loop to avoid having to keep calling spanToJSON
This PR implements sending standalone LCP spans as an opt-in feature. Behaviour-wise, it's mostly aligned with our prior implementation of sending CLS standalone spans (#13056): - add an `_experiments.enableStandaloneLcpSpans` option and treat it as opt-in - keep collecting LCP values until users soft-navigate or the page is hidden - then, send the LCP span once - adds all `lcp.*` span attributes as well as the `lcp` measurement to the span (depending on if we merge #16589 or this first, we might need to readjust size limit) closes #13063 --------- Co-authored-by: s1gr1d <[email protected]> Co-authored-by: Sigrid Huemer <[email protected]>
# Conflicts: # .size-limit.js
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Bug: ElementTiming Timing Inconsistencies
ElementTiming spans for image-paint entries have inconsistent timing due to how loadTime and renderTime are used for span start time and duration calculation, especially when one of these values is 0 (e.g., for cross-origin images without CORS). This leads to:
- Zero-duration spans starting at
loadTimewhenrenderTimeis 0 butloadTimeis present. - Spans ending at
2 * renderTimewhenloadTimeis 0 butrenderTimeis present, as duration is incorrectly calculated asrenderTime.
packages/browser-utils/src/metrics/elementTiming.ts#L71-L85
sentry-javascript/packages/browser-utils/src/metrics/elementTiming.ts
Lines 71 to 85 in f6b225c
| // see https://developer.mozilla.org/en-US/docs/Web/API/PerformanceElementTiming/renderTime#cross-origin_image_render_time | |
| const [spanStartTime, spanStartTimeSource] = loadTime | |
| ? [msToSec(loadTime), 'load-time'] | |
| : renderTime | |
| ? [msToSec(renderTime), 'render-time'] | |
| : [timestampInSeconds(), 'entry-emission']; | |
| const duration = | |
| paintType === 'image-paint' | |
| ? // for image paints, we can acually get a duration because image-paint entries also have a `loadTime` | |
| // and `renderTime`. `loadTime` is the time when the image finished loading and `renderTime` is the | |
| // time when the image finished rendering. | |
| msToSec(Math.max(0, (renderTime ?? 0) - (loadTime ?? 0))) | |
| : // for `'text-paint'` entries, we can't get a duration because the `loadTime` is always zero. | |
| 0; |
Was this report helpful? Give feedback by reacting with 👍 or 👎
This PR adds support for instrumenting and sending spans from
ElementTimingAPI entries. Just like with web vitals and long tasks/animation frames, we register aPerformanceObserverand extract spans from newly emitted ET entries.Important:
enableElementTiming: falseinbrowserTracingIntegrationnavigationspans as well). We could also go the route of only sending until the first navigation as with standalone CLS/LCP spans. Happy to accept any direction we wanna take this.Some noteworthy findings while working on this:
loadTimewhich is the relative timestamp to the browser'stimeOrigin, when the image finished loading. For text nodes,loadTimeis always0, since nothing needs to be loaded.renderTimewhich is the relative timestamp to the browser'stimeOrigin, when the node finished rendering (i.e. was painted by the browser).renderTime - loadTimefor image nodes0for text nodestimeOrigin + loadTimefor image nodestimeOrigin + renderTimefor text nodesIn addition to the raw span and conventional attributes, we also collect a bunch of ET-specific attributes:
element.type- tag name of the element (e.g.imgorp)element.size- width x height of the elementelement.render-time-entry.renderTimeelement.load-time-entry.loadTimeelement.url- url of the loaded image (undefinedfor text nodes)element.identifier- the identifier passed to theelementtiming=identifierHTML attributeelement.paint-type- the node paint type (image-paintortext-paint)also some additional sentry-sepcific attributes:
route- the route name, either from the active root span (if available) or from the scope'stransactionNamesentry.span-start-time-source- the data point we used as the span start timeMore than happy to adjust any of this logic or attribute names, based on review feedback :)
closes #13675
also ref #7292