Profiles the given file or expression using Erlang's tprof tool.

Requires Erlang/OTP27 or above.

:tprof is an experimental module introduced in Erlang/OTP 27 which provides a unified API for measuring call count, time, and allocation, and aims to replace :eprof and :cprof. It can be useful when you want to discover the bottlenecks related to any of these measurements.

Before running the code, it invokes the app.start task which compiles and loads your project. After that, the target expression is profiled together with all matching function calls using the Erlang trace BIFs. The tracing of the function calls for that is enabled when the profiling is begun, and disabled when profiling is stopped.

To profile the code, you can use syntax similar to the mix run task:

$ mix profile.tprof -e Hello.world
$ mix profile.tprof -e "[1, 2, 3] |> Enum.reverse |> Enum.map(&Integer.to_string/1)"
$ mix profile.tprof my_script.exs arg1 arg2 arg3

By default, tprof uses the time type, but you can profile memory too:

$ mix profile.tprof -e "Enum.map([1, 2, 3], &Integer.to_string/1)" --type memory

Call count is present with both type time and memory, but if you only need the call count information, you can use the type calls which has the lowest footprint:

$ mix profile.tprof -e "Enum.map([1, 2, 3], &Integer.to_string/1)" --type calls

This task is automatically re-enabled, so you can profile multiple times in the same Mix invocation.

Command line options

• --matching - only profile calls matching the given Module.function/arity pattern
• --type - the type of profiling, calls, time or memory (default: time)
• --calls - filters out any results with a call count lower than this
• --time - filters out any results that took lower than specified (in µs), the type needs to be time
• --memory - filters out any results that used less memory than specified (in words), the type needs to be memory
• --sort - sorts the results by calls, per_call or by the value of type (default: the value of type)
• --report - returns the per-process breakdown when process, or the total for all processes when total (default: process). Always total when type is calls.
• --eval, -e - evaluates the given code
• --require, -r - requires pattern before running the command
• --parallel, -p - makes all requires parallel
• --no-warmup - skips the warmup step before profiling
• --no-compile - does not compile even if files require compilation
• --no-deps-check - does not check dependencies
• --no-archives-check - does not check archives
• --no-halt - does not halt the system after running the command
• --no-start - does not start applications after compilation
• --no-elixir-version-check - does not check the Elixir version from mix.exs

Profile output

Example output (time type):

Profile results of #PID<0.107.0>
#                                               CALLS      % TIME µS/CALL
Total                                              20 100.00    2    0.10
String.Chars.Integer.to_string/1                    5   0.00    0    0.00
anonymous fn/0 in :elixir_compiler_1.__FILE__/1     1   0.00    0    0.00
Enum.each/2                                         1   0.00    0    0.00
Enum.reduce_range/5                                 3   0.00    0    0.00
:erlang.integer_to_binary/1                         5  50.00    1    0.20
anonymous fn/3 in Enum.each/2                       5  50.00    1    0.20

Profile done over 6 matching functions

Example output (memory type):

Profile results of #PID<0.107.0>
#                           CALLS      % WORDS PER CALL
Total                           6 100.00    19     3.17
Enum.each/2                     1  21.05     4     4.00
:erlang.integer_to_binary/1     5  78.95    15     3.00

Profile done over 2 matching functions

Example output (calls type)

Profile results over all processes
#                                               CALLS      %
Total                                              20 100.00
anonymous fn/0 in :elixir_compiler_1.__FILE__/1     1   5.00
Enum.each/2                                         1   5.00
Enum.reduce_range/5                                 3  15.00
:erlang.integer_to_binary/1                         5  25.00
String.Chars.Integer.to_string/1                    5  25.00
anonymous fn/3 in Enum.each/2                       5  25.00

Profile done over 6 matching functions

The default output contains data gathered from all matching functions. The first row after the header contains the sums of the partial results and the average time or memory usage for all the function calls listed. The following rows contain the function call, followed by the number of times that the function was called, then by the percentage of time/memory that the call uses, then the total time/memory for that function in microseconds/words, and, finally, the average time/memory per call in microseconds/words.

When --matching option is specified, call count tracing will be started only for the functions matching the given pattern:

Profile results of #PID<0.106.0>
#                                CALLS      % TIME µS/CALL
Total                                5 100.00    1    0.20
String.Chars.Integer.to_string/1     5 100.00    1    0.20

Profile done over 1 matching functions

The pattern can be a module name, such as String to count all calls to that module, a call without arity, such as String.split, to count all calls to that function regardless of arity, or a call with arity, such as String.split/3, to count all calls to that exact module, function and arity.

Caveats

You should be aware that the code being profiled is running in an anonymous function which is invoked by :tprof module. Thus, you'll see some additional entries in your profile output. It is also important to note that the profiler is stopped as soon as the code has finished running, and this may need special attention, when: running asynchronous code as function calls which were called before the profiler stopped will not be counted; running synchronous code as long running computations and a profiler without a proper MFA trace pattern or filter may lead to a result set which is difficult to comprehend.

You should expect a slowdown in your code execution using this tool since :tprof has some performance impact on the execution, but the impact is considerably lower than Mix.Tasks.Profile.Fprof. If you have a large system try to profile a limited scenario or focus on the main modules or processes. The calls type can also be used, which is more limited but has a lower footprint.