class Sampler

Defined at line 59 of file ../../src/performance/cpu_profiler/sampler.h

Protected Members

async_dispatcher_t * dispatcher_

TargetTree targets_

vector sample_specs_

vector inspecting_durations_

function sample_cb_

unordered_map process_watchers_

unordered_map job_watchers_

map contexts_

unordered_set pending_module_caches_

Public Methods

void Sampler (async_dispatcher_t * dispatcher, TargetTree targets, std::vector<fuchsia_cpu_profiler::SamplingConfig> sample_specs, SampleCallback sample_cb)

Defined at line 61 of file ../../src/performance/cpu_profiler/sampler.h

fxl::WeakPtr<Sampler> GetWeakPtr ()

Defined at line 85 of file ../../src/performance/cpu_profiler/sampler.h

std::optional<std::string_view> GetProcessName (zx_koid_t pid)

Defined at line 87 of file ../../src/performance/cpu_profiler/sampler.h

std::optional<std::string_view> GetThreadName (zx_koid_t tid)

Defined at line 90 of file ../../src/performance/cpu_profiler/sampler.h

std::vector<zx::ticks> SamplingDurations ()

Defined at line 93 of file ../../src/performance/cpu_profiler/sampler.h

zx::result<> Start (size_t buffer_size_mb)

Defined at line 237 of file ../../src/performance/cpu_profiler/sampler.cc

zx::result<> Stop ()

Defined at line 248 of file ../../src/performance/cpu_profiler/sampler.cc

void GetContexts (fit::callback<void (zx::result<profiler::SymbolizationContext>)> callback)

Asynchronously gathers the information needed to symbolize the samples.

Process modules are read on a background thread so that `dispatcher_` keeps servicing the

process exception channels in the meantime: reading a process's memory may fault, and

resolving that fault may in turn require one of those exceptions to be handled.

`callback` is invoked on `dispatcher_` with the result, after every module scan scheduled

before this call has completed. If the sampler is destroyed first, `callback` is dropped

without being invoked. Only one call may be outstanding at a time.

Defined at line 296 of file ../../src/performance/cpu_profiler/sampler.cc

zx::result<> AddTarget (JobTarget && target)

Defined at line 108 of file ../../src/performance/cpu_profiler/sampler.cc

void ~Sampler ()

Defined at line 291 of file ../../src/performance/cpu_profiler/sampler.cc

Protected Methods

zx::result<> WatchTargets ()

Defined at line 181 of file ../../src/performance/cpu_profiler/sampler.cc

zx::result<> WatchTarget (const JobTarget & target)

Defined at line 117 of file ../../src/performance/cpu_profiler/sampler.cc

void AddThread (std::vector<zx_koid_t> job_path, zx_koid_t pid, zx_koid_t tid, zx::thread t)

Defined at line 369 of file ../../src/performance/cpu_profiler/sampler.cc

void RemoveThread (std::vector<zx_koid_t> job_path, zx_koid_t pid, zx_koid_t tid)

Defined at line 385 of file ../../src/performance/cpu_profiler/sampler.cc

void CollectSamples (async_dispatcher_t * dispatcher, async::TaskBase * task, zx_status_t status)

Defined at line 255 of file ../../src/performance/cpu_profiler/sampler.cc

void CacheModules (const ProcessTarget & p)

Defined at line 394 of file ../../src/performance/cpu_profiler/sampler.cc

void ScheduleCacheModules (std::vector<zx_koid_t> job_path, zx_koid_t pid)

Cache the modules of process `pid` on a dedicated background worker loop so that reading a

process's modules (e.g. for Starnix processes with large address spaces) never blocks

`dispatcher_`, which services process/thread exception channels.

Defined at line 443 of file ../../src/performance/cpu_profiler/sampler.cc

void RunUntilModuleCachesDoneForTesting (async::Loop & loop)

Runs `loop`, which must own `dispatcher_`, until every task already posted to the module cache

loop has run and the results it posted back to `dispatcher_` have been delivered.

`loop` keeps running in the meantime rather than blocking on the module cache loop: a module

cache task (or the module cache thread starting) can wait on an exception that only

`dispatcher_` handles.

Defined at line 423 of file ../../src/performance/cpu_profiler/sampler.cc

void AddProcessesStartedSinceConfigure ()

The target tree is populated when the session is configured, which can take a long time for

large targets (e.g. the whole system), and processes and threads keep starting in the meantime.

These add the ones that started since then, so that they aren't silently left out of the

session.

Call AddProcessesStartedSinceConfigure after watching the target jobs, and

AddThreadsStartedSinceConfigure after watching `process`, so that anything starting later is

reported by the watchers instead.

Defined at line 491 of file ../../src/performance/cpu_profiler/sampler.cc

void AddThreadsStartedSinceConfigure (std::span<const zx_koid_t> job_path, const ProcessTarget & process)

Defined at line 592 of file ../../src/performance/cpu_profiler/sampler.cc