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!
Note that this whole module is experimental, and the representations used as well as the functionality is likely to change in the future.
Types
block_histogram() = tuple()
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.
allocation_summary() =
{HistogramStart :: integer() >= 0,
UnscannedSize :: integer() >= 0,
Allocations ::
#{Origin :: atom() =>
#{Type :: atom() => block_histogram()}}}
A summary of allocated block sizes (including their headers) grouped
by their
and
.
is generally which NIF or driver that
allocated the blocks, or 'system' if it could not be determined.
is the allocation category that the blocks
belong to, e.g. db_term
, message
or binary
.
If one or more carriers could not be scanned in full without harming
the responsiveness of the system,
is the number of bytes that had to be skipped.
carrier_info_list() =
{HistogramStart :: integer() >= 0,
Carriers ::
[{AllocatorType :: atom(),
TotalSize :: integer() >= 0,
UnscannedSize :: integer() >= 0,
AllocatedSize :: integer() >= 0,
AllocatedCount :: integer() >= 0,
InPool :: boolean(),
FreeBlocks :: block_histogram()}]}
is the type of the allocator that
employs this carrier.
is the total size of the carrier,
including its header.
is the combined size of the
carrier's allocated blocks, including their headers.
is the number of allocated
blocks in the carrier.
is whether the carrier is in the
migration pool.
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,
is
the number of bytes that had to be skipped.
Functions
allocations() -> {ok, Result} | {error, Reason}
Result = allocation_summary()
Reason = not_enabled
Shorthand for
allocations(#{})
.
allocations(Options) -> {ok, Result} | {error, Reason}
Result = allocation_summary()
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:
allocator_types
The allocator types that will be searched. Note that blocks can move freely between allocator types, so restricting the search to certain allocators may return unexpected types (e.g. process heaps when searching binary_alloc), or hide blocks that were migrated out.
Defaults to all alloc_util
allocators.
scheduler_ids
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.
histogram_start
The upper bound of the first interval in the allocated block size histograms. Defaults to 128.
histogram_width
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}}}}}
carriers() -> {ok, Result} | {error, Reason}
Result = carrier_info_list()
Reason = not_enabled
Shorthand for
carriers(#{})
.
carriers(Options) -> {ok, Result} | {error, Reason}
Result = carrier_info_list()
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:
allocator_types
The allocator types that will be searched. Defaults to all
alloc_util
allocators.
scheduler_ids
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.
histogram_start
The upper bound of the first interval in the free block size histograms. Defaults to 512.
histogram_width
The number of intervals in the free block size histograms. Defaults to 14.
Example:
> instrument:carriers(#{ histogram_start => 512, histogram_width => 8 }). {ok,{512, [{ll_alloc,1048576,0,1048344,71,false,{0,0,0,0,0,0,0,0}}, {binary_alloc,1048576,0,324640,13,false,{3,0,0,1,0,0,0,2}}, {eheap_alloc,2097152,0,1037200,45,false,{2,1,1,3,4,3,2,2}}, {fix_alloc,32768,0,29544,82,false,{22,0,0,0,0,0,0,0}}, {...}|...]}}