Debugger.Memory =============== The [`Debugger API`][debugger] can help tools observe the debuggee's memory use in various ways: - It can mark each new object with the JavaScript call stack at which it was allocated. - It can log all object allocations, yielding a stream of JavaScript call stacks at which allocations have occurred. - It can compute a *census* of items belonging to the debuggee, categorizing items in various ways, and yielding item counts. If dbg is a [`Debugger`][debugger-object] instance, then the methods and accessor properties of `dbg.memory` control how dbg observes its debuggees' memory use. The `dbg.memory` object is an instance of `Debugger.Memory`; its inherited accessors and methods are described below. ## Allocation Site Tracking The JavaScript engine marks each new object with the call stack at which it was allocated, if: - the object is allocated in the scope of a global object that is a debuggee of some [`Debugger`][debugger-object] instance dbg; and - dbg.memory.[trackingAllocationSites][tracking-allocs] is set to `true`. - A [Bernoulli trial][bernoulli-trial] succeeds, with probability equal to the maximum of [`d.memory.allocationSamplingProbability`][alloc-sampling-probability] of all `Debugger` instances `d` that are observing the global that this object is allocated within the scope of. Given a [`Debugger.Object`][object] instance dobj referring to some object, dobj.[allocationSite][allocation-site] returns a [saved call stack][saved-frame] indicating where dobj's referent was allocated. ## Allocation Logging If dbg is a [`Debugger`][debugger-object] instance, and dbg.memory.[trackingAllocationSites][tracking-allocs] is set to `true`, then the JavaScript engine logs each object allocated by dbg's debuggee code. You can retrieve the current log by calling dbg.memory.[drainAllocationsLog][drain-alloc-log]. You can control the limit on the log's size by setting dbg.memory.[maxAllocationsLogLength][max-alloc-log]. ## Censuses A *census* is a complete traversal of the graph of all reachable memory items belonging to a particular `Debugger`'s debuggees. It produces a count of those items, broken down by various criteria. If dbg is a [`Debugger`][debugger-object] instance, you can call dbg.memory.[takeCensus][take-census] to conduct a census of its debuggees' possessions. Accessor Properties of the `Debugger.Memory.prototype` Object ------------------------------------------------------------- If dbg is a [`Debugger`][debugger-object] instance, then `dbg.memory` is a `Debugger.Memory` instance, which inherits the following accessor properties from its prototype: ## `trackingAllocationSites` A boolean value indicating whether this `Debugger.Memory` instance is capturing the JavaScript execution stack when each Object is allocated. This accessor property has both a getter and setter: assigning to it enables or disables the allocation site tracking. Reading the accessor produces `true` if the Debugger is capturing stacks for Object allocations, and `false` otherwise. Allocation site tracking is initially disabled in a new Debugger. Assignment is fallible: if the Debugger cannot track allocation sites, it throws an `Error` instance. You can retrieve the allocation site for a given object with the [`Debugger.Object.prototype.allocationSite`][allocation-site] accessor property. ## `allocationSamplingProbability` A number between 0 and 1 that indicates the probability with which each new allocation should be entered into the allocations log. 0 is equivalent to "never", 1 is "always", and .05 would be "one out of twenty". The default is 1, or logging every allocation. Note that in the presence of multiple Debugger instances observing the same allocations within a global's scope, the maximum allocationSamplingProbability of all the Debuggers is used. ## `maxAllocationsLogLength` The maximum number of allocation sites to accumulate in the allocations log at a time. This accessor can be both fetched and stored to. Its default value is `5000`. ## `allocationsLogOverflowed` Returns `true` if there have been more than [`maxAllocationsLogLength`][#max-alloc-log] allocations since the last time [`drainAllocationsLog`][#drain-alloc-log] was called and some data has been lost. Returns `false` otherwise. Debugger.Memory Handler Functions --------------------------------- Similar to [`Debugger`'s handler functions][debugger], `Debugger.Memory` inherits accessor properties that store handler functions for SpiderMonkey to call when given events occur in debuggee code. Unlike `Debugger`'s hooks, `Debugger.Memory`'s handlers' return values are not significant, and are ignored. The handler functions receive the `Debugger.Memory`'s owning `Debugger` instance as their `this` value. The owning `Debugger`'s `uncaughtExceptionHandler` is still fired for errors thrown in `Debugger.Memory` hooks. On a new `Debugger.Memory` instance, each of these properties is initially `undefined`. Any value assigned to a debugging handler must be either a function or `undefined`; otherwise a `TypeError` is thrown. Handler functions run in the same thread in which the event occurred. They run in the compartment to which they belong, not in a debuggee compartment. ## `onGarbageCollection(statistics)` A garbage collection cycle spanning one or more debuggees has just been completed. The *statistics* parameter is an object containing information about the GC cycle. It has the following properties: ## `collections` The `collections` property's value is an array. Because SpiderMonkey's collector is incremental, a full collection cycle may consist of multiple discrete collection slices with the JS mutator running interleaved. For each collection slice that occurred, there is an entry in the `collections` array with the following form: ``` { "startTimestamp": timestamp, "endTimestamp": timestamp, } ``` Here the `timestamp` values are [timestamps][timestamps] of the GC slice's start and end events. ## `reason` A very short string describing the reason why the collection was triggered. Known values include the following: * `"API"` * `"EAGER_ALLOC_TRIGGER"` * `"DESTROY_RUNTIME"` * `"LAST_DITCH"` * `"TOO_MUCH_MALLOC"` * `"ALLOC_TRIGGER"` * `"DEBUG_GC"` * `"COMPARTMENT_REVIVED"` * `"RESET"` * `"OUT_OF_NURSERY"` * `"EVICT_NURSERY"` * `"FULL_STORE_BUFFER"` * `"SHARED_MEMORY_LIMIT"` * `"PERIODIC_FULL_GC"` * `"INCREMENTAL_TOO_SLOW"` * `"DOM_WINDOW_UTILS"` * `"COMPONENT_UTILS"` * `"MEM_PRESSURE"` * `"CC_FINISHED"` * `"CC_FORCED"` * `"LOAD_END"` * `"PAGE_HIDE"` * `"NSJSCONTEXT_DESTROY"` * `"SET_NEW_DOCUMENT"` * `"SET_DOC_SHELL"` * `"DOM_UTILS"` * `"DOM_IPC"` * `"DOM_WORKER"` * `"INTER_SLICE_GC"` * `"REFRESH_FRAME"` * `"FULL_GC_TIMER"` * `"SHUTDOWN_CC"` * `"USER_INACTIVE"` ## `nonincrementalReason` If SpiderMonkey's collector determined it could not incrementally collect garbage, and had to do a full GC all at once, this is a short string describing the reason it determined the full GC was necessary. Otherwise, `null` is returned. Known values include the following: * `"GC mode"` * `"malloc bytes trigger"` * `"allocation trigger"` * `"requested"` ## `gcCycleNumber` The GC cycle's "number". Does not correspond to the number of GC cycles that have run, but is guaranteed to be monotonically increasing. Function Properties of the `Debugger.Memory.prototype` Object ------------------------------------------------------------- ## `drainAllocationsLog()` When `trackingAllocationSites` is `true`, this method returns an array of recent `Object` allocations within the set of debuggees. *Recent* is defined as the `maxAllocationsLogLength` most recent `Object` allocations since the last call to `drainAllocationsLog`. Therefore, calling this method effectively clears the log. Objects in the array are of the form: ``` { "timestamp": timestamp, "frame": allocationSite, "class": className, "size": byteSize, "inNursery": inNursery, } ``` Where * `timestamp` is the [timestamp][timestamps] of the allocation event. * `allocationSite` is an allocation site (as a [captured stack][saved-frame]). Note that this property can be null if the object was allocated with no JavaScript frames on the stack. * `className` is the string name of the allocated object's internal `[[Class]]` property, for example "Array", "Date", "RegExp", or (most commonly) "Object". * `byteSize` is the size of the object in bytes. * `inNursery` is true if the allocation happened inside the nursery. False if the allocation skipped the nursery and started in the tenured heap. When `trackingAllocationSites` is `false`, `drainAllocationsLog()` throws an `Error`. ## `takeCensus(options)` Carry out a census of the debuggee compartments' contents. A *census* is a complete traversal of the graph of all reachable memory items belonging to a particular `Debugger`'s debuggees. The census produces a count of those items, broken down by various criteria. The options argument is an object whose properties specify how the census should be carried out. If options has a `breakdown` property, that determines how the census categorizes the items it finds, and what data it collects about them. For example, if `dbg` is a `Debugger` instance, the following performs a simple count of debuggee items: dbg.memory.takeCensus({ breakdown: { by: 'count' } }) That might produce a result like: { "count": 1616, "bytes": 93240 } Here is a breakdown that groups JavaScript objects by their class name, non-string, non-script items by their C++ type name, and DOM nodes with their node name: { by: "coarseType", objects: { by: "objectClass" }, other: { by: "internalType" }, domNode: { by: "descriptiveType" } } which produces a result like this: { "objects": { "Function": { "count": 404, "bytes": 37328 }, "Object": { "count": 11, "bytes": 1264 }, "Debugger": { "count": 1, "bytes": 416 }, "ScriptSource": { "count": 1, "bytes": 64 }, // ... omitted for brevity... }, "scripts": { "count": 1, "bytes": 0 }, "strings": { "count": 701, "bytes": 49080 }, "other": { "js::Shape": { "count": 450, "bytes": 0 }, "js::BaseShape": { "count": 21, "bytes": 0 }, "js::ObjectGroup": { "count": 17, "bytes": 0 } }, "domNode": { "#text": { "count": 1, "bytes": 12 } } } In general, a `breakdown` value has one of the following forms: * { by: "count", count:count, bytes:bytes } The trivial categorization: none whatsoever. Simply tally up the items visited. If count is true, count the number of items visited; if bytes is true, total the number of bytes the items use directly. Both count and bytes are optional; if omitted, they default to `true`. In the result of the census, this breakdown produces a value of the form: { "count": n, "bytes": b } where the `count` and `bytes` properties are present as directed by the count and bytes properties on the breakdown. Note that the census can produce byte sizes only for the most common types. When the census cannot find the byte size for a given type, it returns zero. * { by: "bucket" } Do not do any filtering or categorizing. Instead, accumulate a bucket of each node's ID for every node that matches. The resulting report is an array of the IDs. For example, to find the ID of all nodes whose internal object `[[class]]` property is named "RegExp", you could use the following code: const report = dbg.memory.takeCensus({ breakdown: { by: "objectClass", then: { by: "bucket" } } }); doStuffWithRegExpIDs(report.RegExp); * { by: "allocationStack", then:breakdown, noStack:noStackBreakdown } Group items by the full JavaScript stack trace at which they were allocated. Further categorize all the items allocated at each distinct stack using breakdown. In the result of the census, this breakdown produces a JavaScript `Map` value whose keys are `SavedFrame` values, and whose values are whatever sort of result breakdown produces. Objects allocated on an empty JavaScript stack appear under the key `null`. SpiderMonkey only tracks allocation sites for items if requested via the [`trackingAllocationSites`][tracking-allocs] flag; even then, it does not record allocation sites for every kind of item that appears in the heap. Items that lack allocation site information are counted using noStackBreakdown. These appear in the result `Map` under the key string `"noStack"`. * { by: "objectClass", then:breakdown, other:otherBreakdown } Group JavaScript objects by their ECMAScript `[[Class]]` internal property values. Further categorize JavaScript objects in each class using breakdown. Further categorize items that are not JavaScript objects using otherBreakdown. In the result of the census, this breakdown produces a JavaScript object with no prototype whose own property names are strings naming classes, and whose values are whatever sort of result breakdown produces. The results for non-object items appear as the value of the property named `"other"`. * { by: "coarseType", objects:objects, scripts:scripts, strings:strings, domNode:domNode, other:other } Group items by their coarse type. Use the breakdown value objects for items that are JavaScript objects. Use the breakdown value scripts for items that are representations of JavaScript code. This includes bytecode, compiled machine code, and saved source code. Use the breakdown value strings for JavaScript strings. Use the breakdown value domNode for DOM nodes. Use the breakdown value other for items that don't fit into any of the above categories. In the result of the census, this breakdown produces a JavaScript object of the form: ``` { "objects": result, "scripts": result, "strings": result, "domNode:" result, "other": result, } ``` where each result is a value of whatever sort the corresponding breakdown value produces. All breakdown values are optional, and default to `{ type: "count" }`. * `{ by: "filename", then:breakdown, noFilename:noFilenameBreakdown }` For scripts only, group by the filename of the script. Further categorize all of the scripts from each distinct filename using breakdown. Scripts that lack a filename are counted using noFilenameBreakdown. These appear in the result `Map` under the key string `"noFilename"`. * `{ by: "internalType", then: breakdown }` Group items by the names given their types internally by SpiderMonkey. These names are not meaningful to web developers, but this type of breakdown does serve as a catch-all that can be useful to Firefox tool developers. For example, a census of a pristine debuggee global broken down by internal type name typically looks like this: { "JSString": { "count": 701, "bytes": 49080 }, "js::Shape": { "count": 450, "bytes": 0 }, "JSObject": { "count": 426, "bytes": 44160 }, "js::BaseShape": { "count": 21, "bytes": 0 }, "js::ObjectGroup": { "count": 17, "bytes": 0 }, "JSScript": { "count": 1, "bytes": 0 } } In the result of the census, this breakdown produces a JavaScript object with no prototype whose own property names are strings naming types, and whose values are whatever sort of result breakdown produces. * [ breakdown, ... ] Group each item using all the given breakdown values. In the result of the census, this breakdown produces an array of values of the sort produced by each listed breakdown. To simplify breakdown values, all `then` and `other` properties are optional. If omitted, they are treated as if they were `{ type: "count" }`. Breakdown groupings cannot be nested within themselves. This would not be useful, and forbidding this prevents infinite recursion. If the `options` argument has no `breakdown` property, `takeCensus` defaults to the following: ```js { by: "coarseType", objects: { by: "objectClass" }, domNode: { by: "descriptiveType" }, other: { by: "internalType" } } ``` which produces results of the form: ``` { objects: { class: count, ... }, scripts: count, strings: count, domNode: { node name:count, ... }, other: { type name:count, ... } } ``` where each `count` has the form: ```js { "count": count, bytes: bytes } ``` Because performing a census requires traversing the entire graph of objects in debuggee compartments, it is an expensive operation. On developer hardware in 2014, traversing a memory graph containing roughly 130,000 nodes and 410,000 edges took roughly 100ms. The traversal itself temporarily allocates one hash table entry per node (roughly two address-sized words) in addition to the per-category counts, whose size depends on the number of categories. Memory Use Analysis Exposes Implementation Details -------------------------------------------------- Memory analysis may yield surprising results, because browser implementation details that are transparent to content JavaScript often have visible effects on memory consumption. Web developers need to know their pages' actual memory consumption on real browsers, so it is correct for the tool to expose these behaviors, as long as it is done in a way that helps developers make decisions about their own code. This section covers some areas where Firefox's actual behavior deviates from what one might expect from the specified behavior of the web platform. ## Objects SpiderMonkey objects usually use less memory than the naïve "table of properties with attributes" model would suggest. For example, it is typical for many objects to have identical sets of properties, with only the properties' values varying from one object to the next. To take advantage of this regularity, SpiderMonkey objects with identical sets of properties may share their property metadata; only property values are stored directly in the object. Array objects may also be optimized, if the set of live indices is dense. ## Strings SpiderMonkey has three representations of strings: - Normal: the string's text is counted in its size. - Substring: the string is a substring of some other string, and points to that string for its storage. This representation may result in a small string retaining a very large string. However, the memory consumed by the string itself is a small constant independent of its size, since it is simply a reference to the larger string, a start position, and a length. - Concatenations: When asked to concatenate two strings, SpiderMonkey may elect to delay copying the strings' data, and represent the result simply as a pointer to the two original strings. Again, such a string retains other strings, but the memory consumed by the string itself is a small constant independent of its size, since it is simply a pair of pointers. SpiderMonkey converts strings from the more complex representations to the simpler ones when it pleases. Such conversions usually increase memory consumption. SpiderMonkey shares some strings amongst all web pages and browser JS. These shared strings, called *atoms*, are not included in censuses' string counts. ## Scripts SpiderMonkey has a complex, hybrid representation of JavaScript code. There are four representations kept in memory: - _Source code_. SpiderMonkey retains a copy of most JavaScript source code. - _Compressed source code_. SpiderMonkey compresses JavaScript source code, and de-compresses it on demand. Heuristics determine how long to retain the uncompressed code. - _Bytecode_. This is SpiderMonkey's parsed representation of JavaScript. Bytecode can be interpreted directly, or used as input to a just-in-time compiler. Source is parsed into bytecode on demand; functions that are never called are never parsed. - _Machine code_. SpiderMonkey includes several just-in-time compilers, each of which translates JavaScript source or bytecode to machine code. Heuristics determine which code to compile, and which compiler to use. Machine code may be dropped in response to memory pressure, and regenerated as needed. Furthermore, SpiderMonkey's just-in-time compilers generate inline caches for type specialization. This information is dropped periodically to reduce memory usage. In a census, all the various forms of JavaScript code are placed in the `"scripts"` category. [debugger]: Debugger-API.md [debugger-object]: Debugger.md [tracking-allocs]: #trackingallocationsites [bernoulli-trial]: https://en.wikipedia.org/wiki/Bernoulli_trial [alloc-sampling-probability]: #allocsamplingprobability [object]: Debugger.Object.md [allocation-site]: Debugger.Object.md#allocationsite [saved-frame]: ../SavedFrame/index [drain-alloc-log]: #drainAllocationsLog [max-alloc-log]: #maxAllocationsLogLength [take-census]: #takecensus-options [timestamps]: ./Conventions.md#timestamps