// Copyright (c) 2006-2008 The Chromium Authors. All rights reserved. // Use of this source code is governed by a BSD-style license that can be // found in the LICENSE file. #ifndef BASE_TRACKED_OBJECTS_H_ #define BASE_TRACKED_OBJECTS_H_ #include <map> #include <string> #include <vector> #include "base/lock.h" #include "base/task.h" #include "base/thread_local_storage.h" #include "base/tracked.h" // TrackedObjects provides a database of stats about objects (generally Tasks) // that are tracked. Tracking means their birth, death, duration, birth thread, // death thread, and birth place are recorded. This data is carefully spread // across a series of objects so that the counts and times can be rapidly // updated without (usually) having to lock the data, and hence there is usually // very little contention caused by the tracking. The data can be viewed via // the about:objects URL, with a variety of sorting and filtering choices. // // Theese classes serve as the basis of a profiler of sorts for the Tasks // system. As a result, design decisions were made to maximize speed, by // minimizing recurring allocation/deallocation, lock contention and data // copying. In the "stable" state, which is reached relatively quickly, there // is no separate marginal allocation cost associated with construction or // destruction of tracked objects, no locks are generally employed, and probably // the largest computational cost is associated with obtaining start and stop // times for instances as they are created and destroyed. The introduction of // worker threads had a slight impact on this approach, and required use of some // locks when accessing data from the worker threads. // // The following describes the lifecycle of tracking an instance. // // First off, when the instance is created, the FROM_HERE macro is expanded // to specify the birth place (file, line, function) where the instance was // created. That data is used to create a transient Location instance // encapsulating the above triple of information. The strings (like __FILE__) // are passed around by reference, with the assumption that they are static, and // will never go away. This ensures that the strings can be dealt with as atoms // with great efficiency (i.e., copying of strings is never needed, and // comparisons for equality can be based on pointer comparisons). // // Next, a Births instance is created for use ONLY on the thread where this // instance was created. That Births instance records (in a base class // BirthOnThread) references to the static data provided in a Location instance, // as well as a pointer specifying the thread on which the birth takes place. // Hence there is at most one Births instance for each Location on each thread. // The derived Births class contains slots for recording statistics about all // instances born at the same location. Statistics currently include only the // count of instances constructed. // Since the base class BirthOnThread contains only constant data, it can be // freely accessed by any thread at any time (i.e., only the statistic needs to // be handled carefully, and it is ONLY read or written by the birth thread). // // Having now either constructed or found the Births instance described above, a // pointer to the Births instance is then embedded in a base class of the // instance we're tracking (usually a Task). This fact alone is very useful in // debugging, when there is a question of where an instance came from. In // addition, the birth time is also embedded in the base class Tracked (see // tracked.h), and used to later evaluate the lifetime duration. // As a result of the above embedding, we can (for any tracked instance) find // out its location of birth, and thread of birth, without using any locks, as // all that data is constant across the life of the process. // // The amount of memory used in the above data structures depends on how many // threads there are, and how many Locations of construction there are. // Fortunately, we don't use memory that is the product of those two counts, but // rather we only need one Births instance for each thread that constructs an // instance at a Location. In many cases, instances (such as Tasks) are only // created on one thread, so the memory utilization is actually fairly // restrained. // // Lastly, when an instance is deleted, the final tallies of statistics are // carefully accumulated. That tallying wrties into slots (members) in a // collection of DeathData instances. For each birth place Location that is // destroyed on a thread, there is a DeathData instance to record the additional // death count, as well as accumulate the lifetime duration of the instance as // it is destroyed (dies). By maintaining a single place to aggregate this // addition *only* for the given thread, we avoid the need to lock such // DeathData instances. // // With the above lifecycle description complete, the major remaining detail is // explaining how each thread maintains a list of DeathData instances, and of // Births instances, and is able to avoid additional (redundant/unnecessary) // allocations. // // Each thread maintains a list of data items specific to that thread in a // ThreadData instance (for that specific thread only). The two critical items // are lists of DeathData and Births instances. These lists are maintained in // STL maps, which are indexed by Location. As noted earlier, we can compare // locations very efficiently as we consider the underlying data (file, // function, line) to be atoms, and hence pointer comparison is used rather than // (slow) string comparisons. // // To provide a mechanism for iterating over all "known threads," which means // threads that have recorded a birth or a death, we create a singly linked list // of ThreadData instances. Each such instance maintains a pointer to the next // one. A static member of ThreadData provides a pointer to the first_ item on // this global list, and access to that first_ item requires the use of a lock_. // When new ThreadData instances is added to the global list, it is pre-pended, // which ensures that any prior acquisition of the list is valid (i.e., the // holder can iterate over it without fear of it changing, or the necessity of // using an additional lock. Iterations are actually pretty rare (used // primarilly for cleanup, or snapshotting data for display), so this lock has // very little global performance impact. // // The above description tries to define the high performance (run time) // portions of these classes. After gathering statistics, calls instigated // by visiting about:objects will assemble and aggregate data for display. The // following data structures are used for producing such displays. They are // not performance critical, and their only major constraint is that they should // be able to run concurrently with ongoing augmentation of the birth and death // data. // // For a given birth location, information about births are spread across data // structures that are asynchronously changing on various threads. For display // purposes, we need to construct Snapshot instances for each combination of // birth thread, death thread, and location, along with the count of such // lifetimes. We gather such data into a Snapshot instances, so that such // instances can be sorted and aggregated (and remain frozen during our // processing). Snapshot instances use pointers to constant portions of the // birth and death datastructures, but have local (frozen) copies of the actual // statistics (birth count, durations, etc. etc.). // // A DataCollector is a container object that holds a set of Snapshots. A // DataCollector can be passed from thread to thread, and each thread // contributes to it by adding or updating Snapshot instances. DataCollector // instances are thread safe containers which are passed to various threads to // accumulate all Snapshot instances. // // After an array of Snapshots instances are colleted into a DataCollector, they // need to be sorted, and possibly aggregated (example: how many threads are in // a specific consecutive set of Snapshots? What was the total birth count for // that set? etc.). Aggregation instances collect running sums of any set of // snapshot instances, and are used to print sub-totals in an about:objects // page. // // TODO(jar): I need to store DataCollections, and provide facilities for taking // the difference between two gathered DataCollections. For now, I'm just // adding a hack that Reset()'s to zero all counts and stats. This is also // done in a slighly thread-unsafe fashion, as the reseting is done // asynchronously relative to ongoing updates, and worse yet, some data fields // are 64bit quantities, and are not atomicly accessed (reset or incremented // etc.). For basic profiling, this will work "most of the time," and should be // sufficient... but storing away DataCollections is the "right way" to do this. // class MessageLoop; namespace tracked_objects { //------------------------------------------------------------------------------ // For a specific thread, and a specific birth place, the collection of all // death info (with tallies for each death thread, to prevent access conflicts). class ThreadData; class BirthOnThread { public: explicit BirthOnThread(const Location& location); const Location location() const { return location_; } const ThreadData* birth_thread() const { return birth_thread_; } private: // File/lineno of birth. This defines the essence of the type, as the context // of the birth (construction) often tell what the item is for. This field // is const, and hence safe to access from any thread. const Location location_; // The thread that records births into this object. Only this thread is // allowed to access birth_count_ (which changes over time). const ThreadData* birth_thread_; // The thread this birth took place on. DISALLOW_COPY_AND_ASSIGN(BirthOnThread); }; //------------------------------------------------------------------------------ // A class for accumulating counts of births (without bothering with a map<>). class Births: public BirthOnThread { public: explicit Births(const Location& location); int birth_count() const { return birth_count_; } // When we have a birth we update the count for this BirhPLace. void RecordBirth() { ++birth_count_; } // When a birthplace is changed (updated), we need to decrement the counter // for the old instance. void ForgetBirth() { --birth_count_; } // We corrected a birth place. // Hack to quickly reset all counts to zero. void Clear() { birth_count_ = 0; } private: // The number of births on this thread for our location_. int birth_count_; DISALLOW_COPY_AND_ASSIGN(Births); }; //------------------------------------------------------------------------------ // Basic info summarizing multiple destructions of an object with a single // birthplace (fixed Location). Used both on specific threads, and also used // in snapshots when integrating assembled data. class DeathData { public: // Default initializer. DeathData() : count_(0), square_duration_(0) {} // When deaths have not yet taken place, and we gather data from all the // threads, we create DeathData stats that tally the number of births without // a corrosponding death. explicit DeathData(int count) : count_(count), square_duration_(0) {} void RecordDeath(const base::TimeDelta& duration); // Metrics accessors. int count() const { return count_; } base::TimeDelta life_duration() const { return life_duration_; } int64 square_duration() const { return square_duration_; } int AverageMsDuration() const; double StandardDeviation() const; // Accumulate metrics from other into this. void AddDeathData(const DeathData& other); // Simple print of internal state. void Write(std::string* output) const; // Reset all tallies to zero. void Clear(); private: int count_; // Number of destructions. base::TimeDelta life_duration_; // Sum of all lifetime durations. int64 square_duration_; // Sum of squares in milliseconds. }; //------------------------------------------------------------------------------ // A temporary collection of data that can be sorted and summarized. It is // gathered (carefully) from many threads. Instances are held in arrays and // processed, filtered, and rendered. // The source of this data was collected on many threads, and is asynchronously // changing. The data in this instance is not asynchronously changing. class Snapshot { public: // When snapshotting a full life cycle set (birth-to-death), use this: Snapshot(const BirthOnThread& birth_on_thread, const ThreadData& death_thread, const DeathData& death_data); // When snapshotting a birth, with no death yet, use this: Snapshot(const BirthOnThread& birth_on_thread, int count); const ThreadData* birth_thread() const { return birth_->birth_thread(); } const Location location() const { return birth_->location(); } const BirthOnThread& birth() const { return *birth_; } const ThreadData* death_thread() const {return death_thread_; } const DeathData& death_data() const { return death_data_; } const std::string DeathThreadName() const; int count() const { return death_data_.count(); } base::TimeDelta life_duration() const { return death_data_.life_duration(); } int64 square_duration() const { return death_data_.square_duration(); } int AverageMsDuration() const { return death_data_.AverageMsDuration(); } void Write(std::string* output) const; void Add(const Snapshot& other); private: const BirthOnThread* birth_; // Includes Location and birth_thread. const ThreadData* death_thread_; DeathData death_data_; }; //------------------------------------------------------------------------------ // DataCollector is a container class for Snapshot and BirthOnThread count // items. It protects the gathering under locks, so that it could be called via // Posttask on any threads, or passed to all the target threads in parallel. class DataCollector { public: typedef std::vector<Snapshot> Collection; // Construct with a list of how many threads should contribute. This helps us // determine (in the async case) when we are done with all contributions. DataCollector(); // Add all stats from the indicated thread into our arrays. This function is // mutex protected, and *could* be called from any threads (although current // implementation serialized calls to Append). void Append(const ThreadData& thread_data); // After the accumulation phase, the following accessor is used to process the // data. Collection* collection(); // After collection of death data is complete, we can add entries for all the // remaining living objects. void AddListOfLivingObjects(); private: // This instance may be provided to several threads to contribute data. The // following counter tracks how many more threads will contribute. When it is // zero, then all asynchronous contributions are complete, and locked access // is no longer needed. int count_of_contributing_threads_; // The array that we collect data into. Collection collection_; // The total number of births recorded at each location for which we have not // seen a death count. typedef std::map<const BirthOnThread*, int> BirthCount; BirthCount global_birth_count_; Lock accumulation_lock_; // Protects access during accumulation phase. DISALLOW_COPY_AND_ASSIGN(DataCollector); }; //------------------------------------------------------------------------------ // Aggregation contains summaries (totals and subtotals) of groups of Snapshot // instances to provide printing of these collections on a single line. class Aggregation: public DeathData { public: Aggregation() : birth_count_(0) {} void AddDeathSnapshot(const Snapshot& snapshot); void AddBirths(const Births& births); void AddBirth(const BirthOnThread& birth); void AddBirthPlace(const Location& location); void Write(std::string* output) const; void Clear(); private: int birth_count_; std::map<std::string, int> birth_files_; std::map<Location, int> locations_; std::map<const ThreadData*, int> birth_threads_; DeathData death_data_; std::map<const ThreadData*, int> death_threads_; DISALLOW_COPY_AND_ASSIGN(Aggregation); }; //------------------------------------------------------------------------------ // Comparator is a class that supports the comparison of Snapshot instances. // An instance is actually a list of chained Comparitors, that can provide for // arbitrary ordering. The path portion of an about:objects URL is translated // into such a chain, which is then used to order Snapshot instances in a // vector. It orders them into groups (for aggregation), and can also order // instances within the groups (for detailed rendering of the instances in an // aggregation). class Comparator { public: // Selector enum is the token identifier for each parsed keyword, most of // which specify a sort order. // Since it is not meaningful to sort more than once on a specific key, we // use bitfields to accumulate what we have sorted on so far. enum Selector { // Sort orders. NIL = 0, BIRTH_THREAD = 1, DEATH_THREAD = 2, BIRTH_FILE = 4, BIRTH_FUNCTION = 8, BIRTH_LINE = 16, COUNT = 32, AVERAGE_DURATION = 64, TOTAL_DURATION = 128, // Imediate action keywords. RESET_ALL_DATA = -1, }; explicit Comparator(); // Reset the comparator to a NIL selector. Clear() and recursively delete any // tiebreaker_ entries. NOTE: We can't use a standard destructor, because // the sort algorithm makes copies of this object, and then deletes them, // which would cause problems (either we'd make expensive deep copies, or we'd // do more thna one delete on a tiebreaker_. void Clear(); // The less() operator for sorting the array via std::sort(). bool operator()(const Snapshot& left, const Snapshot& right) const; void Sort(DataCollector::Collection* collection) const; // Check to see if the items are sort equivalents (should be aggregated). bool Equivalent(const Snapshot& left, const Snapshot& right) const; // Check to see if all required fields are present in the given sample. bool Acceptable(const Snapshot& sample) const; // A comparator can be refined by specifying what to do if the selected basis // for comparison is insufficient to establish an ordering. This call adds // the indicated attribute as the new "least significant" basis of comparison. void SetTiebreaker(Selector selector, const std::string& required); // Indicate if this instance is set up to sort by the given Selector, thereby // putting that information in the SortGrouping, so it is not needed in each // printed line. bool IsGroupedBy(Selector selector) const; // Using the tiebreakers as set above, we mostly get an ordering, which // equivalent groups. If those groups are displayed (rather than just being // aggregated, then the following is used to order them (within the group). void SetSubgroupTiebreaker(Selector selector); // Translate a keyword and restriction in URL path to a selector for sorting. void ParseKeyphrase(const std::string& key_phrase); // Parse a query in an about:objects URL to decide on sort ordering. bool ParseQuery(const std::string& query); // Output a header line that can be used to indicated what items will be // collected in the group. It lists all (potentially) tested attributes and // their values (in the sample item). bool WriteSortGrouping(const Snapshot& sample, std::string* output) const; // Output a sample, with SortGroup details not displayed. void WriteSnapshot(const Snapshot& sample, std::string* output) const; private: // The selector directs this instance to compare based on the specified // members of the tested elements. enum Selector selector_; // For filtering into acceptable and unacceptable snapshot instance, the // following is required to be a substring of the selector_ field. std::string required_; // If this instance can't decide on an ordering, we can consult a tie-breaker // which may have a different basis of comparison. Comparator* tiebreaker_; // We or together all the selectors we sort on (not counting sub-group // selectors), so that we can tell if we've decided to group on any given // criteria. int combined_selectors_; // Some tiebreakrs are for subgroup ordering, and not for basic ordering (in // preparation for aggregation). The subgroup tiebreakers are not consulted // when deciding if two items are in equivalent groups. This flag tells us // to ignore the tiebreaker when doing Equivalent() testing. bool use_tiebreaker_for_sort_only_; }; //------------------------------------------------------------------------------ // For each thread, we have a ThreadData that stores all tracking info generated // on this thread. This prevents the need for locking as data accumulates. class ThreadData { public: typedef std::map<Location, Births*> BirthMap; typedef std::map<const Births*, DeathData> DeathMap; ThreadData(); // Using Thread Local Store, find the current instance for collecting data. // If an instance does not exist, construct one (and remember it for use on // this thread. // If shutdown has already started, and we don't yet have an instance, then // return null. static ThreadData* current(); // For a given about:objects URL, develop resulting HTML, and append to // output. static void WriteHTML(const std::string& query, std::string* output); // For a given accumulated array of results, use the comparator to sort and // subtotal, writing the results to the output. static void WriteHTMLTotalAndSubtotals( const DataCollector::Collection& match_array, const Comparator& comparator, std::string* output); // In this thread's data, record a new birth. Births* TallyABirth(const Location& location); // Find a place to record a death on this thread. void TallyADeath(const Births& lifetimes, const base::TimeDelta& duration); // (Thread safe) Get start of list of instances. static ThreadData* first(); // Iterate through the null terminated list of instances. ThreadData* next() const { return next_; } MessageLoop* message_loop() const { return message_loop_; } const std::string ThreadName() const; // Using our lock, make a copy of the specified maps. These calls may arrive // from non-local threads, and are used to quickly scan data from all threads // in order to build an HTML page for about:objects. void SnapshotBirthMap(BirthMap *output) const; void SnapshotDeathMap(DeathMap *output) const; // Hack: asynchronously clear all birth counts and death tallies data values // in all ThreadData instances. The numerical (zeroing) part is done without // use of a locks or atomics exchanges, and may (for int64 values) produce // bogus counts VERY rarely. static void ResetAllThreadData(); // Using our lock to protect the iteration, Clear all birth and death data. void Reset(); // Using the "known list of threads" gathered during births and deaths, the // following attempts to run the given function once all all such threads. // Note that the function can only be run on threads which have a message // loop! static void RunOnAllThreads(void (*Func)()); // Set internal status_ to either become ACTIVE, or later, to be SHUTDOWN, // based on argument being true or false respectively. // IF tracking is not compiled in, this function will return false. static bool StartTracking(bool status); static bool IsActive(); #ifdef OS_WIN // WARNING: ONLY call this function when all MessageLoops are still intact for // all registered threads. IF you call it later, you will crash. // Note: You don't need to call it at all, and you can wait till you are // single threaded (again) to do the cleanup via // ShutdownSingleThreadedCleanup(). // Start the teardown (shutdown) process in a multi-thread mode by disabling // further additions to thread database on all threads. First it makes a // local (locked) change to prevent any more threads from registering. Then // it Posts a Task to all registered threads to be sure they are aware that no // more accumulation can take place. static void ShutdownMultiThreadTracking(); #endif // WARNING: ONLY call this function when you are running single threaded // (again) and all message loops and threads have terminated. Until that // point some threads may still attempt to write into our data structures. // Delete recursively all data structures, starting with the list of // ThreadData instances. static void ShutdownSingleThreadedCleanup(); private: // Current allowable states of the tracking system. The states always // proceed towards SHUTDOWN, and never go backwards. enum Status { UNINITIALIZED, ACTIVE, SHUTDOWN, }; // A class used to count down which is accessed by several threads. This is // used to make sure RunOnAllThreads() actually runs a task on the expected // count of threads. class ThreadSafeDownCounter { public: // Constructor sets the count, once and for all. explicit ThreadSafeDownCounter(size_t count); // Decrement the count, and return true if we hit zero. Also delete this // instance automatically when we hit zero. bool LastCaller(); private: size_t remaining_count_; Lock lock_; // protect access to remaining_count_. }; #ifdef OS_WIN // A Task class that runs a static method supplied, and checks to see if this // is the last tasks instance (on last thread) that will run the method. // IF this is the last run, then the supplied event is signalled. class RunTheStatic : public Task { public: typedef void (*FunctionPointer)(); RunTheStatic(FunctionPointer function, HANDLE completion_handle, ThreadSafeDownCounter* counter); // Run the supplied static method, and optionally set the event. void Run(); private: FunctionPointer function_; HANDLE completion_handle_; // Make sure enough tasks are called before completion is signaled. ThreadSafeDownCounter* counter_; DISALLOW_COPY_AND_ASSIGN(RunTheStatic); }; #endif // Each registered thread is called to set status_ to SHUTDOWN. // This is done redundantly on every registered thread because it is not // protected by a mutex. Running on all threads guarantees we get the // notification into the memory cache of all possible threads. static void ShutdownDisablingFurtherTracking(); // We use thread local store to identify which ThreadData to interact with. static TLSSlot tls_index_; // Link to the most recently created instance (starts a null terminated list). static ThreadData* first_; // Protection for access to first_. static Lock list_lock_; // We set status_ to SHUTDOWN when we shut down the tracking service. This // setting is redundantly established by all participating threads so that we // are *guaranteed* (without locking) that all threads can "see" the status // and avoid additional calls into the service. static Status status_; // Link to next instance (null terminated list). Used to globally track all // registered instances (corresponds to all registered threads where we keep // data). ThreadData* next_; // The message loop where tasks needing to access this instance's private data // should be directed. Since some threads have no message loop, some // instances have data that can't be (safely) modified externally. MessageLoop* message_loop_; // A map used on each thread to keep track of Births on this thread. // This map should only be accessed on the thread it was constructed on. // When a snapshot is needed, this structure can be locked in place for the // duration of the snapshotting activity. BirthMap birth_map_; // Similar to birth_map_, this records informations about death of tracked // instances (i.e., when a tracked instance was destroyed on this thread). // It is locked before changing, and hence other threads may access it by // locking before reading it. DeathMap death_map_; // Lock to protect *some* access to BirthMap and DeathMap. The maps are // regularly read and written on this thread, but may only be read from other // threads. To support this, we acquire this lock if we are writing from this // thread, or reading from another thread. For reading from this thread we // don't need a lock, as there is no potential for a conflict since the // writing is only done from this thread. mutable Lock lock_; DISALLOW_COPY_AND_ASSIGN(ThreadData); }; //------------------------------------------------------------------------------ // Provide simple way to to start global tracking, and to tear down tracking // when done. Note that construction and destruction of this object must be // done when running in threaded mode (before spawning a lot of threads // for construction, and after shutting down all the threads for destruction). // To prevent grabbing thread local store resources time and again if someone // chooses to try to re-run the browser many times, we maintain global state and // only allow the tracking system to be started up at most once, and shutdown // at most once. See bug 31344 for an example. class AutoTracking { public: AutoTracking() { if (state_ != kNeverBeenRun) return; ThreadData::StartTracking(true); state_ = kRunning; } ~AutoTracking() { #ifndef NDEBUG if (state_ != kRunning) return; // Don't call these in a Release build: they just waste time. // The following should ONLY be called when in single threaded mode. It is // unsafe to do this cleanup if other threads are still active. // It is also very unnecessary, so I'm only doing this in debug to satisfy // purify (if we need to!). ThreadData::ShutdownSingleThreadedCleanup(); state_ = kTornDownAndStopped; #endif } private: enum State { kNeverBeenRun, kRunning, kTornDownAndStopped, }; static State state_; DISALLOW_COPY_AND_ASSIGN(AutoTracking); }; } // namespace tracked_objects #endif // BASE_TRACKED_OBJECTS_H_