instrument

instrument

instrument
Analysis and Utility Functions for Instrumentation

The module instrument contains support for studying the resource usage in an Erlang runtime system. Currently, only the allocation of memory can be studied.

Note

Since this module inspects internal details of the runtime system it may differ greatly from one version to another. We make no compatibility guarantees in this module.

A histogram of block sizes where each interval's upper bound is twice as high as the one before it.

The upper bound of the first interval is provided by the function that returned the histogram, and the last interval has no upper bound.

For example, the histogram below has 40 (message) blocks between 256-512 bytes in size, 78 blocks between 512-1024 bytes,2 blocks between 1-2KB, and 2 blocks between 2-4KB.

> instrument:allocations(#{ histogram_start => 128, histogram_width => 15 }).
{ok, {128, 0, #{ message => {0,40,78,2,2,0,0,0,0,0,0,0,0,0,0}, ... } }}

A summary of allocated block sizes (including their headers) grouped by their Origin and Type.

Origin is generally which NIF or driver that allocated the blocks, or 'system' if it could not be determined.

Type is the allocation category that the blocks belong to, e.g. db_term, message or binary. The categories correspond to those in erl_alloc.types.

If one or more carriers could not be scanned in full without harming the responsiveness of the system, UnscannedSize is the number of bytes that had to be skipped.

AllocatorType is the type of the allocator that employs this carrier.

InPool is whether the carrier is in the migration pool.

TotalSize is the total size of the carrier, including its header.

Allocations is a summary of the allocated blocks in the carrier. Note that carriers may contain multiple different block types when carrier pools are shared between different allocator types (see the erts_alloc documentation for more details).

FreeBlocks is a histogram of the free block sizes in the carrier.

If the carrier could not be scanned in full without harming the responsiveness of the system, UnscannedSize is the number of bytes that had to be skipped.

Types

Reason = not_enabled
Options =
    #{scheduler_ids => [integer() >= 0],
      allocator_types => [atom()],
      histogram_start => integer() >= 1,
      histogram_width => integer() >= 1}

Returns a summary of all tagged allocations in the system, optionally filtered by allocator type and scheduler id.

Only binaries and allocations made by NIFs and drivers are tagged by default, but this can be configured an a per-allocator basis with the +M<S>atags emulator option.

If the specified allocator types are not enabled, the call will fail with {error, not_enabled}.

The following options can be used:

The allocator types that will be searched.

Specifying a specific allocator type may lead to strange results when carrier migration between different allocator types has been enabled: you may see unexpected types (e.g. process heaps when searching binary_alloc), or fewer blocks than expected if the carriers the blocks are on have been migrated out to an allocator of a different type.

Defaults to all alloc_util allocators.

The scheduler ids whose allocator instances will be searched. A scheduler id of 0 will refer to the global instance that is not tied to any particular scheduler. Defaults to all schedulers and the global instance.

The upper bound of the first interval in the allocated block size histograms. Defaults to 128.

The number of intervals in the allocated block size histograms. Defaults to 18.

Example:

> instrument:allocations(#{ histogram_start => 128, histogram_width => 15 }).
{ok,{128,0,
     #{udp_inet =>
           #{driver_event_state => {0,0,0,0,0,0,0,0,0,1,0,0,0,0,0}},
       system =>
           #{heap => {0,0,0,0,20,4,2,2,2,3,0,1,0,0,1},
             db_term => {271,3,1,52,80,1,0,0,0,0,0,0,0,0,0},
             code => {0,0,0,5,3,6,11,22,19,20,10,2,1,0,0},
             binary => {18,0,0,0,7,0,0,1,0,0,0,0,0,0,0},
             message => {0,40,78,2,2,0,0,0,0,0,0,0,0,0,0},
             ... }
       spawn_forker =>
           #{driver_select_data_state =>
                 {1,0,0,0,0,0,0,0,0,0,0,0,0,0,0}},
       ram_file_drv => #{drv_binary => {0,0,0,0,0,0,1,0,0,0,0,0,0,0,0}},
       prim_file =>
           #{process_specific_data => {2,0,0,0,0,0,0,0,0,0,0,0,0,0,0},
             nif_trap_export_entry => {0,4,0,0,0,0,0,0,0,0,0,0,0,0,0},
             monitor_extended => {0,1,0,0,0,0,0,0,0,0,0,0,0,0,0},
             drv_binary => {0,0,0,0,0,0,1,0,3,5,0,0,0,1,0},
             binary => {0,4,0,0,0,0,0,0,0,0,0,0,0,0,0}},
       prim_buffer =>
           #{nif_internal => {0,4,0,0,0,0,0,0,0,0,0,0,0,0,0},
             binary => {0,4,0,0,0,0,0,0,0,0,0,0,0,0,0}}}}}

Types

Reason = not_enabled
Options =
    #{scheduler_ids => [integer() >= 0],
      allocator_types => [atom()],
      histogram_start => integer() >= 1,
      histogram_width => integer() >= 1}

Returns a summary of all carriers in the system, optionally filtered by allocator type and scheduler id.

If the specified allocator types are not enabled, the call will fail with {error, not_enabled}.

The following options can be used:

The allocator types that will be searched. Defaults to all alloc_util allocators.

The scheduler ids whose allocator instances will be searched. A scheduler id of 0 will refer to the global instance that is not tied to any particular scheduler. Defaults to all schedulers and the global instance.

The upper bound of the first interval in the free block size histograms. Defaults to 512.

The number of intervals in the free block size histograms. Defaults to 14.

Example:

> instrument:carriers(#{ histogram_start => 512, histogram_width => 8 }).
{ok,{512,
     [{driver_alloc,false,262144,0,
                    [{driver_alloc,1,32784}],
                    {0,0,0,0,0,0,0,1}},
      {binary_alloc,false,32768,0,
                    [{binary_alloc,15,4304}],
                    {3,0,0,0,1,0,0,0}},
      {...}|...]}}