Erlang Doctor

Lightweight tracing, debugging and profiling tool, which collects traces in an ETS table, putting minimal impact on your system. After collecting the traces, you can query and analyse them. By separating data collection from analysis, this tool helps you limit unnecessary repetition and guesswork. There is ExDoctor for Elixir as well.

Quick start

To quickly try it out right now, copy & paste the following to your Erlang shell:

P = "/tmp/tr.erl", ssl:start(), inets:start(), {ok, {{_, 200, _}, _, Src}} = httpc:request("https://git.io/fj024"), file:write_file(P, Src), {ok, tr, B} = compile:file(P, binary), code:load_binary(tr, P, B), rr(P), tr:start().

This snippet downloads, compiles and starts the tr module from the master branch. Your Erlang Doctor is now ready to use!

The easiest way to use it is the following:

tr:trace([your_module]).
your_module:some_function().
tr:select().

You should see the collected traces for the call and return of your_module:some_function/0.

This compact tool is capable of much more - see below.

Include it as a dependency

To avoid copy-pasting the snippet shown above, you can include erlang_doctor in your dependencies in rebar.config. There is a Hex package as well.

Use it during development

You can make Erlang Doctor available in the Erlang/Rebar3 shell during development by cloning it to ERLANG_DOCTOR_PATH, calling rebar3 compile, and loading it in your ~/.erlang file:

code:add_path("ERLANG_DOCTOR_PATH/erlang_doctor/_build/default/lib/erlang_doctor/ebin").
code:load_file(tr).

Tracing: data collection

The test suite helpers from tr_SUITE.erl are used here as examples. You can follow these examples on your own - just call rebar3 as test shell in ERLANG_DOCTOR_PATH.

Setting up: start, start_link

The first thing to do is to start the tracer with tr:start/0.

There is also tr:start/1, which accepts a map of options, including:

• tab: collected traces are stored in an ETS table with this name (default: trace),
• limit: maximum number of traces in the table - when it is reached, tracing is stopped (default: no limit).

There are tr:start_link/0 and tr:start_link/1 as well, and they are intended for use with the whole erlang_doctor application.

For this tutorial we start the tr module in the simplest way:

1> tr:start().
{ok, <0.218.0>}

Tracing with trace

To trace function calls for given modules, use tr:trace/1, providing a list of traced modules:

2> tr:trace([tr_SUITE]).
ok

You can provide {Module, Function, Arity} tuples in the list as well. The function tr:trace_app/1 traces an application, and tr:trace_apps/1 traces multiple ones. If you need to trace an application and some additional modules, use tr:app_modules/1 to get the list of modules for an application:

tr:trace([Module1, Module2 | tr:app_modules(YourApp)]).

If you want to trace selected processes instead of all of them, you can use tr:trace/2:

tr:trace([Module1, Module2], [Pid1, Pid2]).

The tr:trace/1 function accepts a map of options, which include:

• modules: a list of module names or {Module, Function, Arity} tuples. The list is empty by default.
• pids: a list of Pids of processes to trace, or the atom all (default) to trace all processes.
• msg: none (default), all, send or recv. Specifies which message events will be traced. By default no messages are traced.
• msg_trigger: after_traced_call (default) or always. By default, traced messages in each process are stored after the first traced function call in that process. The goal is to limit the number of traced messages, which can be huge in the entire Erlang system. If you want all messages, set it to always.

Calling the traced function

Now we can call some functions - let's trace the following function call. It calculates the factorial recursively and sleeps 1 ms between each step.

3> tr_SUITE:sleepy_factorial(3).
6

Stopping tracing

You can stop tracing with the following function:

4> tr:stop_tracing().
ok

It's good to stop it as soon as possible to avoid accumulating too many traces in the ETS table. Usage of tr on production systems is risky, but if you have to do it, start and stop the tracer in the same command, e.g. for one second with:

tr:trace(Modules), timer:sleep(1000), tr:stop_tracing().

Debugging: data analysis

The collected traces are stored in an ETS table (default name: trace). They are stored as tr records with the following fields:

• index: trace identifier, auto-incremented for each received trace.
• pid: process identifier associated with the trace.
• event: call, return or exception for function traces; send or recv for messages.
• mfa: {Module, Function, Arity} for function traces; no_mfa for messages.
• data: argument list (for calls), returned value (for returns) or class and value (for exceptions).
• timestamp in microseconds.
• info: For function traces and recv events it is no_info. For send events it is a {To, Exists} tuple, where To is the recipient pid, and Exists is a boolean indicating if the recipient process existed.

It's useful to read the record definitions before trace analysis:

5> rr(tr).
[node,tr]

The snippet shown at the top of this page includes this already.

Trace selection: select

Use tr:select/0 to select all collected traces.

6> tr:select().
[#tr{index = 1,pid = <0.395.0>,event = call,
     mfa = {tr_SUITE,sleepy_factorial,1},
     data = [3],
     ts = 1705475521743239,info = no_info},
 #tr{index = 2,pid = <0.395.0>,event = call,
     mfa = {tr_SUITE,sleepy_factorial,1},
     data = [2],
     ts = 1705475521744690,info = no_info},
 #tr{index = 3,pid = <0.395.0>,event = call,
     mfa = {tr_SUITE,sleepy_factorial,1},
     data = [1],
     ts = 1705475521746470,info = no_info},
 #tr{index = 4,pid = <0.395.0>,event = call,
     mfa = {tr_SUITE,sleepy_factorial,1},
     data = [0],
     ts = 1705475521748499,info = no_info},
 #tr{index = 5,pid = <0.395.0>,event = return,
     mfa = {tr_SUITE,sleepy_factorial,1},
     data = 1,ts = 1705475521750451,info = no_info},
 #tr{index = 6,pid = <0.395.0>,event = return,
     mfa = {tr_SUITE,sleepy_factorial,1},
     data = 1,ts = 1705475521750453,info = no_info},
 #tr{index = 7,pid = <0.395.0>,event = return,
     mfa = {tr_SUITE,sleepy_factorial,1},
     data = 2,ts = 1705475521750454,info = no_info},
 #tr{index = 8,pid = <0.395.0>,event = return,
     mfa = {tr_SUITE,sleepy_factorial,1},
     data = 6,ts = 1705475521750455,info = no_info}]

The tr:select/1 function accepts a fun that is passed to ets:fun2ms/1. This way you can limit the selection to specific items and select only some fields from the tr record:

7> tr:select(fun(#tr{event = call, data = [N]}) -> N end).
[3, 2, 1, 0]

Use tr:select/2 to further filter the results by searching for a term in #tr.data (recursively searching in lists, tuples and maps).

8> tr:select(fun(T) -> T end, 2).
[#tr{index = 2,pid = <0.395.0>,event = call,
     mfa = {tr_SUITE,sleepy_factorial,1},
     data = [2],
     ts = 1705475521744690,info = no_info},
 #tr{index = 7,pid = <0.395.0>,event = return,
     mfa = {tr_SUITE,sleepy_factorial,1},
     data = 2,ts = 1705475521750454,info = no_info}]

Trace filtering: filter

Sometimes it might be easier to use tr:filter/1, because it can accept any function as the argument. You can use tr:contains_data/2 to search for the same term as in the example above.

9> Traces = tr:filter(fun(T) -> tr:contains_data(2, T) end).
[#tr{index = 2,pid = <0.395.0>,event = call,
     mfa = {tr_SUITE,sleepy_factorial,1},
     data = [2],
     ts = 1705475521744690,info = no_info},
 #tr{index = 7,pid = <0.395.0>,event = return,
     mfa = {tr_SUITE,sleepy_factorial,1},
     data = 2,ts = 1705475521750454,info = no_info}]

The provided function is a predicate, which has to return true for the matching traces. For other traces it can return another value, or even raise an exception:

10> tr:filter(fun(#tr{data = [2]}) -> true end).
[#tr{index = 2,pid = <0.395.0>,event = call,
     mfa = {tr_SUITE,sleepy_factorial,1},
     data = [2],
     ts = 1705475521744690,info = no_info}]

There is also tr:filter/2, which can be used to search in a different table than the current one - or in a list:

11> tr:filter(fun(#tr{event = call}) -> true end, Traces).
[#tr{index = 2,pid = <0.395.0>,event = call,
     mfa = {tr_SUITE,sleepy_factorial,1},
     data = [2],
     ts = 1705475521744690,info = no_info}]

Tracebacks for filtered traces: tracebacks

To find the tracebacks (stack traces) for matching traces, use tr:tracebacks/1:

12> tr:tracebacks(fun(#tr{data = 1}) -> true end).
[[#tr{index = 3,pid = <0.395.0>,event = call,
      mfa = {tr_SUITE,sleepy_factorial,1},
      data = [1],
      ts = 1705475521746470,info = no_info},
  #tr{index = 2,pid = <0.395.0>,event = call,
      mfa = {tr_SUITE,sleepy_factorial,1},
      data = [2],
      ts = 1705475521744690,info = no_info},
  #tr{index = 1,pid = <0.395.0>,event = call,
      mfa = {tr_SUITE,sleepy_factorial,1},
      data = [3],
      ts = 1705475521743239,info = no_info}]]

Note, that by specifying data = 1 we are only matching return traces, as call traces always have a list in data. Only one traceback is returned. It starts with a call that returned 1. What follows is the stack trace for this call.

One can notice that the call for 0 also returned 1, but the call tree got pruned - whenever two tracebacks overlap, only the shorter one is left. You can change this by returning tracebacks for all matching traces even if they overlap, setting the output option to all. All options are specified in the second argument, which is a map:

13> tr:tracebacks(fun(#tr{data = 1}) -> true end, #{output => all}).
[[#tr{index = 4,pid = <0.395.0>,event = call,
      mfa = {tr_SUITE,sleepy_factorial,1},
      data = [0],
      ts = 1705475521748499,info = no_info},
  #tr{index = 3,pid = <0.395.0>,event = call,
      mfa = {tr_SUITE,sleepy_factorial,1},
      data = [1],
      ts = 1705475521746470,info = no_info},
  #tr{index = 2,pid = <0.395.0>,event = call,
      mfa = {tr_SUITE,sleepy_factorial,1},
      data = [2],
      ts = 1705475521744690,info = no_info},
  #tr{index = 1,pid = <0.395.0>,event = call,
      mfa = {tr_SUITE,sleepy_factorial,1},
      data = [3],
      ts = 1705475521743239,info = no_info}],
 [#tr{index = 3,pid = <0.395.0>,event = call,
      mfa = {tr_SUITE,sleepy_factorial,1},
      data = [1],
      ts = 1705475521746470,info = no_info},
  #tr{index = 2,pid = <0.395.0>,event = call,
      mfa = {tr_SUITE,sleepy_factorial,1},
      data = [2],
      ts = 1705475521744690,info = no_info},
  #tr{index = 1,pid = <0.395.0>,event = call,
      mfa = {tr_SUITE,sleepy_factorial,1},
      data = [3],
      ts = 1705475521743239,info = no_info}]]

The third possibility is output => longest which does the opposite of pruning, leaving only the longest tracabecks when they overlap:

14> tr:tracebacks(fun(#tr{data = 1}) -> true end, #{output => longest}).
[[#tr{index = 4,pid = <0.395.0>,event = call,
      mfa = {tr_SUITE,sleepy_factorial,1},
      data = [0],
      ts = 1705475521748499,info = no_info},
  #tr{index = 3,pid = <0.395.0>,event = call,
      mfa = {tr_SUITE,sleepy_factorial,1},
      data = [1],
      ts = 1705475521746470,info = no_info},
  #tr{index = 2,pid = <0.395.0>,event = call,
      mfa = {tr_SUITE,sleepy_factorial,1},
      data = [2],
      ts = 1705475521744690,info = no_info},
  #tr{index = 1,pid = <0.395.0>,event = call,
      mfa = {tr_SUITE,sleepy_factorial,1},
      data = [3],
      ts = 1705475521743239,info = no_info}]]

Possible options for tr:tracebacks/2 include:

• tab is the table or list which is like the second argument of tr:filter/2,
• output - shortest (default), all, longest - see above.
• format - list (default), tree - returns a call tree instead of a list of tracebacks. Trees don't distinguish between all and longest output formats.
• order - top_down (default), bottom_up - call order in each tracaback; only for the list format.
• limit - positive integer or infinity (default) - limits the number of matched traces. The actual number of tracebacks returned can be smaller unless output => all

There are also functions tr:traceback/1 and tr:traceback/2. They set limit to one and return only one trace if it exists. The options for tr:traceback/2 are the same as for tr:traceback/2 except limit and format. Additionally, it is possible to pass a tr record (or an index) directly to tr:traceback/1 to obtain the traceback for the provided trace event.

Trace ranges for filtered traces: ranges

To get a list of traces between each matching call and the corresponding return, use tr:ranges/1:

15> tr:ranges(fun(#tr{data = [1]}) -> true end).
[[#tr{index = 3,pid = <0.395.0>,event = call,
      mfa = {tr_SUITE,sleepy_factorial,1},
      data = [1],
      ts = 1705475521746470,info = no_info},
  #tr{index = 4,pid = <0.395.0>,event = call,
      mfa = {tr_SUITE,sleepy_factorial,1},
      data = [0],
      ts = 1705475521748499,info = no_info},
  #tr{index = 5,pid = <0.395.0>,event = return,
      mfa = {tr_SUITE,sleepy_factorial,1},
      data = 1,ts = 1705475521750451,info = no_info},
  #tr{index = 6,pid = <0.395.0>,event = return,
      mfa = {tr_SUITE,sleepy_factorial,1},
      data = 1,ts = 1705475521750453,info = no_info}]]

There is also tr:ranges/2 - it accepts a map of options, including:

• tab is the table or list which is like the second argument of tr:filter/2,
• max_depth is the maximum depth of nested calls. A message event also adds 1 to the depth. You can use #{max_depth => 1} to see only the top-level call and the corresponding return.

There are two additional function: tr:range/1 and tr:range/2, which return only one range if it exists. It is possible to pass a tr record or an index to tr:range/1 as well.

Calling function from a trace: do

It is easy to replay a particular function call with tr:do/1:

16> [T] = tr:filter(fun(#tr{data = [3]}) -> true end).
[#tr{index = 1,pid = <0.395.0>,event = call,
     mfa = {tr_SUITE,sleepy_factorial,1},
     data = [3],
     ts = 1705475521743239,info = no_info}]
17> tr:do(T).
6

This is useful e.g. for checking if a bug has been fixed without running the whole test suite. This function can be called with an index as the argument.

Getting a single trace for the index: lookup

Use tr:lookup/1 to obtain the trace for an index.

Profiling

You can quickly get a hint about possible bottlenecks and redundancies in your system with function call statistics.

Call statistics: call_stat

The argument of tr:call_stat/1 is a function that returns a key by which the traces are grouped. The simplest way to use this function is to look at the total number of calls and their time. To do this, we group all calls under one key, e.g. total:

18> tr:call_stat(fun(_) -> total end).
#{total => {4,7216,7216}}

Values of the returned map have the following format (time is in microseconds):

{call_count(), acc_time(), own_time()}

In the example there are four calls, which took 7216 microseconds in total. For nested calls we only take into account the outermost call, so this means that the whole calculation took 7.216 ms. Let's see how this looks like for individual steps - we can group the stats by the function argument:

19> tr:call_stat(fun(#tr{data = [N]}) -> N end).
#{0 => {1,1952,1952},
  1 => {1,3983,2031},
  2 => {1,5764,1781},
  3 => {1,7216,1452}}

You can use the provided function to do filtering as well:

20> tr:call_stat(fun(#tr{data = [N]}) when N < 3 -> N end).
#{0 => {1,1952,1952},1 => {1,3983,2031},2 => {1,5764,1781}}

Sorted call statistics: sorted_call_stat

You can sort the call stat by accumulated time (descending) with tr:sorted_call_stat/1:

21> tr:sorted_call_stat(fun(#tr{data = [N]}) -> N end).
[{3,1,7216,1452},
 {2,1,5764,1781},
 {1,1,3983,2031},
 {0,1,1952,1952}]

The first element of each tuple is the key, the rest is the same as above. To pretty-print it, use tr:print_sorted_call_stat/2. The second argument limits the table row number, e.g. we can only print the top 3 items:

22> tr:print_sorted_call_stat(fun(#tr{data = [N]}) -> N end, 3).
3  1  7216  1452
2  1  5764  1781
1  1  3983  2031
ok

Call tree statistics: top_call_trees

The function tr:top_call_trees/0 makes it possible to detect complete call trees that repeat several times, where corresponding function calls and returns have the same arguments and return values, respectively. When such functions take a lot of time and do not have useful side effects, they can be often optimized.

As an example, let's trace the call to a function which calculates the 4th element of the Fibonacci Sequence in a recursive way. The trace table should be empty, so let's clean it up first:

23> tr:clean().
ok
24> tr:trace([tr_SUITE]).
ok
25> tr_SUITE:fib(4).
3
26> tr:stop_tracing().
ok

Now it is possible to print the most time consuming call trees that repeat at least twice:

27> tr:top_call_trees().
[{13,2,
  #node{module = tr_SUITE,function = fib,
        args = [2],
        children = [#node{module = tr_SUITE,function = fib,
                          args = [1],
                          children = [],
                          result = {return,1}},
                    #node{module = tr_SUITE,function = fib,
                          args = [0],
                          children = [],
                          result = {return,0}}],
        result = {return,1}}},
 {5,3,
  #node{module = tr_SUITE,function = fib,
        args = [1],
        children = [],
        result = {return,1}}}]

The resulting list contains tuples {Time, Count, Tree} where Time is the accumulated time (in microseconds) spent in the tree, and Count is the number of times the tree repeated. The list is sorted by Time, descending. In the example above fib(2) was called twice and fib(1) was called 3 times, what already shows that the recursive implementation is suboptimal.

There is also tr:top_call_trees/1, which takes a map of options, including:

• output is reduced by default, but it can be set to complete where subtrees of already listed trees are also listed.
• min_count is the minimum number of times a tree has to occur to be listed, the default is 2.
• min_time is the minimum accumulated time for a tree, by default there is no minimum.
• max_size is the maximum number of trees presented, the default is 10.

As an exercise, try calling tr:top_call_trees(#{min_count => 1000}) for fib(20).

Exporting and importing traces

To get the current table name, use tr:tab/0:

28> tr:tab().
trace

To switch to a new table, use tr:set_tab/1. The table need not exist.

29> tr:set_tab(tmp).
ok

Now you can collect traces to the new table without changing the original one.

30> tr:trace([lists]), lists:seq(1, 10), tr:stop_tracing().
ok
31> tr:select().
[#tr{index = 1, pid = <0.175.0>, event = call,
     mfa = {lists, ukeysort, 2},
     data = [1,
             [{'Traces', [#tr{index = 2, pid = <0.175.0>, event = call,
                             mfa = {tr_SUITE, sleepy_factorial, 1},
                             data = [2],
(...)

You can dump a table to file with tr:dump/1 - let's dump the tmp table:

32> tr:dump("tmp.ets").
ok

In a new Erlang session we can load the data with tr:load/1. This will set the current table name to tmp.

1> tr:start().
{ok, <0.181.0>}
2> tr:load("tmp.ets").
{ok, tmp}
3> tr:select().
(...)
4> tr:tab().
tmp

Finally, you can remove all traces from the ETS table with tr:clean/0.

5> tr:clean().
ok

To stop tr, just call tr:stop/0.

Example use cases

Debugging a vague error

While reworking the LDAP connection layer in MongooseIM, the following error occured in the logs:

14:46:35.002 [warning] lager_error_logger_h dropped 79 messages in the last second that exceeded the limit of 50 messages/sec
14:46:35.002 [error] gen_server 'wpool_pool-mongoose_wpool$ldap$global$bind-1' terminated with reason: no case clause matching {badkey,handle} in wpool_process:handle_info/2 line 123
14:46:35.003 [error] CRASH REPORT Process 'wpool_pool-mongoose_wpool$ldap$global$bind-1' with 1 neighbours crashed with reason: no case clause matching {badkey,handle} in wpool_process:handle_info/2 line 123
14:46:35.003 [error] Supervisor 'wpool_pool-mongoose_wpool$ldap$global$bind-process-sup' had child 'wpool_pool-mongoose_wpool$ldap$global$bind-1' started with wpool_process:start_link('wpool_pool-mongoose_wpool$ldap$global$bind-1', mongoose_ldap_worker, [{port,3636},{encrypt,tls},{tls_options,[{verify,verify_peer},{cacertfile,"priv/ssl/cacert.pem"},...]}], [{queue_manager,'wpool_pool-mongoose_wpool$ldap$global$bind-queue-manager'},{time_checker,'wpool_pool-mongoose_wpool$ldap$global$bind-time-checker'},...]) at <0.28894.0> exit with reason no case clause matching {badkey,handle} in wpool_process:handle_info/2 line 123 in context child_terminated
14:46:35.009 [info] Connected to LDAP server
14:46:35.009 [error] gen_server 'wpool_pool-mongoose_wpool$ldap$global$default-1' terminated with reason: no case clause matching {badkey,handle} in wpool_process:handle_info/2 line 123
14:46:35.009 [error] CRASH REPORT Process 'wpool_pool-mongoose_wpool$ldap$global$default-1' with 1 neighbours crashed with reason: no case clause matching {badkey,handle} in wpool_process:handle_info/2 line 123

As this messages appear every 10 seconds (on each attempt to reconnect to LDAP), we can start tracing. The most lkely culprit is the mongoose_ldap_worker module, so let's trace it:

(mongooseim@localhost)16> tr:trace([mongoose_ldap_worker]).
ok

A few seconds (and error messages) later we can check the traces for the badkey value we saw in the logs:

(mongooseim@localhost)17> tr:filter(fun(T) -> tr:contains_data(badkey, T) end).
[#tr{index = 255, pid = <0.8118.1>, event = exception,
     mfa = {mongoose_ldap_worker, connect, 1},
     data = {error, {badkey, handle}},
     ts = 1557838064073778},
     (...)

This means that the key handle was missing from a map. Let's see the traceback to find the exact place in the code:

(mongooseim@localhost)18> tr:traceback(fun(T) -> tr:contains_data(badkey, T) end).
[#tr{index = 254, pid = <0.8118.1>, event = call,
     mfa = {mongoose_ldap_worker, connect, 1},
     data = [#{connect_interval => 10000, encrypt => tls, password => <<>>,
               port => 3636, root_dn => <<>>,
               servers => ["localhost"],
               tls_options =>
                   [{verify, verify_peer},
                    {cacertfile, "priv/ssl/cacert.pem"},
                    {certfile, "priv/ssl/fake_cert.pem"},
                    {keyfile, "priv/ssl/fake_key.pem"}]}],
     ts = 1557838064052121}, ...]

We can see that the handle key is missing from the map passed to mongoose_ldap_worker:connect/1. After looking at the source code of this function and searching for handle we can see only one matching line:

                    State#{handle := Handle};

The := operator assumes that the key is already present in the map. The solution would be to either change it to => or ensure that the map already contains that key.

Loading traces to the trace table after tracing to file

It's possible to use tr with a file generated by dbg:trace_port/2 tracing. The file may be generated on another system.

1> {ok, St} = tr:init({}).
{ok, #{index => 0, traced_modules => []}}
2> dbg:trace_client(file, "/Users/erszcz/work/myproject/long-pong.dbg.trace", {fun tr:handle_trace/2, St}).
<0.178.0>
3> tr:select().
[#tr{index = 1, pid = <14318.7477.2537>, event = call,
     mfa = {mod_ping, user_ping_response_metric, 3},
     data = [{jid, <<"user1">>, <<"myproject.com">>, <<"res1">>,
              <<"user1">>, <<"myproject.com">>, <<"res1">>},
             {iq, <<"EDC1944CF88F67C6">>, result, <<>>, <<"en">>, []},
             5406109],
     ts = 1553517330696515},
...