This specification defines an interface to help web developers measure the performance of their applications by giving them access to high precision timestamps.
This User Timing specification is intended to supersede [[USER-TIMING-2]] and includes:
Web developers need the ability to assess and understand the performance characteristics of their applications. While JavaScript [[ECMA-262]] provides a mechanism to measure application latency (retrieving the current timestamp from theDate.now()
method), the precision of this timestamp varies between user agents.
This document defines thePerformanceMark andPerformanceMeasure interfaces, and extensions to thePerformance
interface, which expose a high precision, monotonically increasing timestamp so that developers can better measure the performance characteristics of their applications.
The following script shows how a developer can use the interfaces defined in this document to obtain timing data related to developer scripts.
async function run() { performance.mark("startTask1"); await doTask1(); // Some developer code performance.mark("endTask1"); performance.mark("startTask2"); await doTask2(); // Some developer code performance.mark("endTask2"); // Log them out const entries = performance.getEntriesByType("mark"); for (const entry of entries) { console.table(entry.toJSON()); } } run();
[[PERFORMANCE-TIMELINE-2]] defines two mechanisms that can be used to retrieve recorded metrics:getEntries()
andgetEntriesByType()
methods, and thePerformanceObserver
interface. The former is best suited for cases where you want to retrieve a particular metric by name at a single point in time, and the latter is optimized for cases where you may want to receive notifications of new metrics as they become available.
As another example, suppose that there is an element which, when clicked, fetches some new content and indicates that it has been fetched. We'd like to report the time from when the user clicked to when the fetch was complete. We can't mark the time the click handler executes since that will miss latency to process the event, so instead we use the event hardware timestamp. We also want to know the name of the component to have more detailed analytics.
element.addEventListener("click", e => { const component = getComponent(element); fetch(component.url).then(() => { element.textContent = "Updated"; const updateMark = performance.mark("update_component", { detail: {component: component.name}, }); performance.measure("click_to_update_component", { detail: {component: component.name}, start: e.timeStamp, end: updateMark.startTime, }); }); });
Some conformance requirements are phrased as requirements on attributes, methods or objects. Such requirements are to be interpreted as requirements on user agents.
The IDL fragments in this specification MUST be interpreted as required for conforming IDL fragments, as described in the Web IDL specification. [[WEBIDL]]
Performance
interfaceThePerformance interface andDOMHighResTimeStamp are defined in [[HR-TIME-2]]. ThePerformanceEntry interface is defined in [[PERFORMANCE-TIMELINE-2]].
dictionary PerformanceMarkOptions { any detail; DOMHighResTimeStamp startTime; }; dictionary PerformanceMeasureOptions { any detail; (DOMString or DOMHighResTimeStamp) start; DOMHighResTimeStamp duration; (DOMString or DOMHighResTimeStamp) end; }; partial interface Performance { PerformanceMark mark(DOMString markName, optional PerformanceMarkOptions markOptions = {}); undefined clearMarks(optional DOMString markName); PerformanceMeasure measure(DOMString measureName, optional (DOMString or PerformanceMeasureOptions) startOrMeasureOptions = {}, optional DOMString endMark); undefined clearMeasures(optional DOMString measureName); };
Stores a timestamp with the associated name (a "mark"). It MUST run these steps:
Removes the stored timestamp with the associated name. It MUST run these steps:
Stores the {{DOMHighResTimeStamp}} duration between two marks along with the associated name (a "measure"). It MUST run these steps:
Performance
object'snow()
method.DOMString
, letstart time be the value returned by running theconvert a mark to a timestamp algorithm passing instartOrMeasureOptions.0
.name
attribute tomeasureName.entryType
attribute toDOMString "measure"
.startTime
attribute tostart time.duration
attribute to the duration fromstart time toend time. The resulting duration value MAY be negative.detail
attribute as follows:null
.Removes stored timestamp with the associated name. It MUST run these steps:
name
[=string/is=]measureName.ThePerformanceMark interface also exposes marks created via the {{Performance}} interface's {{Performance/mark()}} method to thePerformance Timeline.
[Exposed=(Window,Worker)] interface PerformanceMark : PerformanceEntry { constructor(DOMString markName, optional PerformanceMarkOptions markOptions = {}); readonly attribute any detail; };
ThePerformanceMark interface extends the following attributes of the {{PerformanceEntry}} interface:
Thename
attribute must return the mark's name.
TheentryType
attribute must return theDOMString
"mark"
.
ThestartTime
attribute must return a {{DOMHighResTimeStamp}} with the mark's time value.
Theduration
attribute must return a {{DOMHighResTimeStamp}} of value0
.
ThePerformanceMark interface contains the following additional attribute:
Thedetail attribute must return the value it is set to (it's copied from thePerformanceMarkOptions dictionary).
ThePerformanceMark constructor must run the following steps:
Window
object andmarkName uses the same name as aread only attribute in thePerformanceTiming
interface,throw aSyntaxError
.name
attribute tomarkName.entryType
attribute toDOMString "mark"
.startTime
attribute as follows:duration
attribute to0
.ThePerformanceMeasure interface also exposes measures created via the {{Performance}} interface's {{Performance/measure()}} method to thePerformance Timeline.
[Exposed=(Window,Worker)] interface PerformanceMeasure : PerformanceEntry { readonly attribute any detail; };
ThePerformanceMeasure interface extends the following attributes of the {{PerformanceEntry}} interface:
Thename
attribute must return the measure's name.
TheentryType
attribute must return theDOMString
"measure"
.
ThestartTime
attribute must return a {{DOMHighResTimeStamp}} with the measure's start mark.
Theduration
attribute must return a {{DOMHighResTimeStamp}} with the duration of the measure.
ThePerformanceMeasure interface contains the following additional attribute:
Thedetail attribute must return the value it is set to (it's copied from thePerformanceMeasureOptions dictionary).
A user agent implementing the User Timing API would need to include"mark"
and"measure"
in supportedEntryTypes. This allows developers to detect support for User Timing.
Toconvert a mark to a timestamp, given amark that is aDOMString
or {{DOMHighResTimeStamp}} run these steps:
DOMString
and it has the same name as aread only attribute in thePerformanceTiming
interface, letend time be the value returned by running theconvert a name to a timestamp algorithm withname set to the value ofmark.DOMString
, letend time be the value of thestartTime
attribute from the most recent occurrence of aPerformanceMark object in theperformance entry buffer whosename
[=string/is=]mark. If no matching entry is found,throw aSyntaxError
.TypeError
.Toconvert a name to atimestamp given aname that is aread only attribute in thePerformanceTiming
interface, run these steps:
Window
object,throw aTypeError
.navigationStart
, return0
.navigationStart
in thePerformanceTiming
interface.PerformanceTiming
interface.0
,throw anInvalidAccessError
.ThePerformanceTiming interface was defined in [[NAVIGATION-TIMING]] and is now considered obsolete. The use of names from thePerformanceTiming interface is supported to remain backwards compatible, but there are no plans to extend this functionality to names in thePerformanceNavigationTiming interface defined in [[NAVIGATION-TIMING-2]] (or other interfaces) in the future.
Developers are encouraged to use the following recommended mark names to mark common timings. Theuser agent does not validate that the usage of these names is appropriate or consistent with its description.
Adding such recommended mark names can help performance tools tailor guidance to a site. These mark names can also help real user monitoring providers and user agents collect web developer signals regarding their application's performance at scale, and surface this information to developers without requiring any site-specific work.
In this example, the page asynchonously initializes a chat widget, a searchbox, and a newsfeed upon loading. When finished, the "mark_fully_loaded" mark name enables lab tools and analytics providers to automatically show the timing.
window.addEventListener("load", (event) => { Promise.all([ loadChatWidget(), initializeSearchAutocomplete(), initializeNewsfeed()]).then(() => { performance.mark('mark_fully_loaded'); }); });
In this example, the ImageOptimizationComponent for FancyJavaScriptFramework is used to size images for optimal performance. The code notes this feature's usage so that lab tools and analytics can measure whether it helped improve performance.
performance.mark('mark_feature_usage', { 'detail': { 'feature': 'ImageOptimizationComponent', 'framework': 'FancyJavaScriptFramework' } })
The interfaces defined in this specification expose potentially sensitive timing information on specific JavaScript activity of a page. Please refer to [[HR-TIME-2]] for privacy and security considerations of exposing high-resolution timing information.
Because the web platform has been designed with the invariant that any script included on a page has the same access as any other script included on the same page, regardless of the origin of either scripts, the interfaces defined by this specification do not place any restrictions on recording or retrieval of recorded timing information - i.e. a user timing mark or measure recorded by any script included on the page can be read by any other script running on the same page, regardless of origin.
Thanks to James Simonsen, Jason Weber, Nic Jansma, Philippe Le Hegaret, Karen Anderson, Steve Souders, Sigbjorn Vik, Todd Reifsteck, and Tony Gentilcore for their contributions to this work.