1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
|
# API reference
## Functions
#### `start(builtins=False, profile_threads=True, profile_greenlets=True)`
Starts profiling all threads in the current interpreter instance.
This function can be called from any thread at any time.
Resumes profiling if stop() is called previously.
`builtins` enables profiling of builtin functions.
`profile_threads` enables profiling of all threads. If this flag is true, all current threads and the ones that are generated
in the future will be profiled.
`profile_greenlets` enables profiling of multiple greenlets. This argument is only respected when context backend is
'greenlet' and ignored otherwise.
#### `stop()`
Stop the profiler.
Same profiling session might be resumed later by calling `start()`.
#### `clear_stats()`
Clears the profiler results.
All results stay in memory unless application (all threads including the main thread) exits or `clear_stats()` is explicitly called.
#### `get_func_stats(tag=None, ctx_id=None, filter_callback=None)`
Returns the function stats as a list of [`YFuncStat`](#yfuncstat) object. As Yappi is a C extension, it catches the profile data in C API.
Thus, the profile data collected is buffered until `clear_stats` is called. `get_func_stats` function enumerates the underlying
buffered data and aggregates the information there. The functions that contain same index id will be aggregated in a single `YFuncStat`
object. So, if you want to select a specific `tag` or `ctx_id`, you need to select by providing them as arguments to `get_func_stats`.
Otherwise, data with different `tag`/`ctx_id` will be combined into one `YFuncStat` object. If you really would like to enumerate
buffered stats in raw, you can use an undocumented function: `_yappi.enum_func_stats(enum_callback, filter_dict)`. You can see
the usage in `get_func_stats` function.
---
**Note:**
Filtering `tag` and `ctx_id` are very fast compared to using `filter_callback` since the filtering is completely done on the C extension
with an internal hash table.
---
`tag` retrieves the `YFuncStat` objects having the same `tag` as specified.
`ctx_id` retrieves the `YFuncStat` objects having the same `ctx_id` as specified.
`filter_callback` is a callback which takes a `YFuncStat` object as an argument and returns a boolean value to indicate
to include or exclude it. As the object is directly passed to the `filter_callback` you can easily filter on any attribute
that `YFuncStat` has.
---
**Note:**
The `filter` dict is deprecated. Please do not use it anymore. We still support that for backward compatability but
it is not recommended anymore.
---
An example demonstrating how `filter_callback` can be used to filter on a function name having `foo`:
```python
stats = yappi.get_func_stats(
filter_callback=lambda x: x.name == 'foo'
).print_all()
```
There are handy functions that can be used with `filter_callback` to match multiple functions or modules easily.
See [func_matches](#func_matchesstat-funcs) and [module_matches](#module_matchesstat-modules).
#### `get_thread_stats()`
Returns the thread stats as a [`YThreadStat`](#ythreadstat) object.
#### `get_greenlet_stats()`
Returns the greenlet stats as a [`YGreenletStats`](#ygreenletstats) object.
#### `is_running()`
Returns a boolean indicating whether profiler is running or not.
#### `get_clock_type()`
Returns information about the underlying clock type Yappi should use to measure timing.
Read [Clock Types](./clock_types.md) for more information.
#### `set_clock_type(type)`
Sets the underlying clock type. `type` must be one of `"wall"` or `"cpu"`.
Read [Clock Types](./clock_types.md) for more information.
#### `func_matches(stat, funcs)`
This function returns `True` if the `stat`(`YStat`) object is in a given list of `funcs`(`callable`) list.
An example usage is when filtering stats based on actual function objects:
```python
def a():
pass
def b():
pass
...
stats = yappi.get_func_stats(
filter_callback=lambda x: yappi.func_matches(x, [a, b])
)
```
---
**Note:**
Once a profile session is saved or loaded from a file, you cannot use
`func_matches` on the items as the mapping between the stats and the functions are
not serialized.
---
#### `module_matches(stat, modules)`
This function returns `True` if the `stat`(`YStat`) object is in a given list of `modules`(`ModuleType`) list.
An example usage is when filtering stats based on actual module objects:
```python
import collections
...
stats = yappi.get_func_stats(
filter_callback=lambda x: yappi.module_matches(x, [collections])
)
```
---
**Note:**
Once a profile session is saved or loaded from a file, you cannot use
`func_matches` on the items as the mapping between the stats and the functions are
not serialized.
---
#### `set_context_backend(type='native_thread')`
Sets the internal context backend used to track execution context. Type must be one of 'greenlet' or 'native_thread'.
Default is `native_thread` and there is no need to call this function for initialization.
Setting the context backend will reset any callbacks configured via:
- `set_context_id_callback`
- `set_context_name_callback`
#### `set_ctx_id_callback(callback)`
`callback` is a simple callable with no arguments that returns an integer.
`callback` is called for every profile event to get the current context id of a running context.
In Yappi terminology a `context` means a construct that has its own callstack. It defaults to uniquely identifying a threading.Thread object, but there might be some advanced cases for this one, like a different threading library is used.
This is an internally used function, so please do not play with unless you have a good reason.
---
**Note:**
The context id callback can be called from the `threading.Thread` initialization code and thus can hold some related
locks in `threading` library(e.x: _active_limbo_lock). So, it is not safe to use threading APIs like `threading.current_thread()`
which can also use these locks and lead to deadlocks. However, it is safe to use `threading.local()` or global variables
using your own locks.
---
#### `set_tag_callback(callback)` _New in v1.2_
`callback` is a simple callable with no arguments that returns an integer.
`callback` is called for every profile event to get the current tag id of a running function.
In Yappi, every profiled function is associated with a tag. By default, this tag is same for all stats collected. You can change this behavior and aggregate different function stat data in different tags.
This will give additional name-spacing on what kind of data to collect.
A recent use case for this functionality is aggregating of single request/response cycle in an ASGI application via `contextvar` module. See [here](https://github.com/sumerc/yappi/issues/21) for details. It can also be used for profiling multithreaded WSGI applications, too.
A simple example demonstrating on how it can be used to profile request/response cycles in a multithreaded WSGI application where every request/response is handled by a different thread. See [here](https://modwsgi.readthedocs.io/en/develop/user-guides/processes-and-threading.html#the-unix-worker-mpm) for more details.
```python
_req_counter = 0
tlocal = threading.local()
def _worker_tag_cbk():
global _req_counter
if not hasattr(tlocal, '_request_id'):
_req_counter += 1 # protect this with mutex
tlocal._request_id = _req_counter
return tlocal._request_id
yappi.set_tag_callback(_worker_tag_cbk)
yappi.start()
...
# code that starts a server serving request with different threads
...
yappi.stop()
# get per-request/response cycle profiling info
for i in range(_req_counter):
req_stats = yappi.get_func_stats(filter={'tag': i})
req_stats.print_all()
```
---
**Note:**
Relevant request id is held in a thread local storage and we simply increment the counter when we encounter a new thread.
Above code will aggregate all stats into a single tag for every function called from the same thread.
---
#### `yappi.get_mem_usage()`
Returns the internal memory usage of the profiler itself.
#### `convert2pstats(stats)`
Converts the internal stat type of Yappi (as returned by `YFuncStat.get()`) to a [`pstats`](https://docs.python.org/3/library/profile.html#module-pstats) object.
# Classes
## `YFuncStat`
This holds the stat items as a list of `YFuncStat` objects.
| Attribute | Description |
|------------- |--------------------------------------------------------------------------------- |
| name | name of the executed function |
| module | module name of the executed function |
| lineno | line number of the executed function |
| ncall | number of times the executed function is called. |
| nactualcall | number of times the executed function is called, excluding the recursive calls. |
| builtin | bool, indicating whether the executed function is a builtin |
| ttot | total time spent in the executed function |
| tsub | total time spent in the executed function, excluding subcalls |
| tavg | per-call average total time spent in the executed function. |
| index | unique id for the YFuncStat object |
| children | list of [YChildFuncStat](#ychildfuncstat) objects |
| ctx_id | id of the underlying context(thread) |
| ctx_name | name of the underlying context(thread) |
| full_name | unique full name of the executed function |
| tag | tag of the executed function. (set via `set_tag_callback`<br>) |
#### `get()`
This method retrieves the current profiling stats.
[`yappi.get_func_stats()`](#get_func_statstagnone-ctx_idnone-filter_callbacknone) is actually just a wrapper for this function.
#### `add(path, type="ystat")`
This method loads the saved profile stats stored in file at `path`.
`type` indicates the type of the saved profile stats.
Currently, only loading from `"ystat"` format is possible. `"ystat"` is the current Yappi internal format.`
#### `save(path, type="ystat")`
This method saves the current profile stats to file at `path`.
`type` indicates the target type that the profile stats will be saved in.
Can be either
[`"pstat"`](http://docs.python.org/3.3/library/profile.html?highlight=pstat#pstats.Stats.print_stats) or
[`"callgrind"`](http://kcachegrind.sourceforge.net/html/CallgrindFormat.html).
#### `print_all(out=sys.stdout)`
This method prints the current profile stats to `out`.
#### `sort(sort_type, sort_order="desc")`
This method sorts the current profile stats.
The `sort_type` must be one of the following:
- `ncall`
- `ttot`
- `tsub`
- `tavg`
`sort_order` must be either `"desc"` or `"asc"`
#### `clear()`
Clears the stats.
---
**Note:**
This method only clears the current object. You need to explicitly call [`yappi.clear_stats()`](#clear_stats) to clear the current profile session stats.
---
#### `empty()`
Returns a boolean indicating whether we have any stats available or not.
#### `strip_dirs()`
Strip the directory information from the results. Affects the child function stats too.
#### `debug_print()`
This method _debug_ prints the current profile stats to stdout.
Debug print prints out callee functions and more detailed info than the [`print_all()`](#print_alloutsysstdout-1) function call.
## `YThreadStat`
`YThreadStat` object has following attributes:
| Attribute | Description |
|------------- |------------------------------------------------------------------------------------------------ |
| name | class name of the current thread object which is derived from the `threading.Thread`<br> class |
| id | a unique id given by Yappi (ctx_id) |
| tid | the real OS thread id |
| ttot | total time spent in the thread |
| sched_count | number of times this thread is scheduled. |
## `YGreenletStats`
`YGreenletStats` object has following attributes:
| Attribute | Description |
|------------- |------------------------------------------------------------------------------------------------ |
| name | class name of the current greenlet |
| id | a unique id given by Yappi (ctx_id) |
| ttot | total time spent in the thread |
| sched_count | number of times this thread is scheduled. |
#### `get()`
This method retrieves the current thread stats.
[`yappi.get_thread_stats()`](#get_thread_stats) is actually just a wrapper for this function.
#### `print_all(out=sys.stdout)`
This method prints the current profile stats to the file `out`.
#### `sort(sort_type, sort_order="desc")`
This method sorts the current profile stats.
`sort_type` must be either `"ttot"` or `"scnt"`
`sort_order` must be either `"desc"` or `"asc"`
#### `clear()`
Clears the retrieved stats.
---
**Note:**
This method only clears the current object. You need to explicitly call [`yappi.clear_stats()`](#clear_stats) to clear the current profile session stats.
---
#### `empty()`
Returns a `bool` indicating whether we have any stats available or not.
## `YChildFuncStat`
This holds a list of child functions called from the main executed function as a `YChildFuncStat` object.
This class holds a list of `YChildFuncStat` objects.
For example, if `a` calls `b`, then `a.children` will hold `b` as a `YChildFuncStat` object.
`YChildFuncStat` object has following attributes:
| Attribute | Description |
|------------- |--------------------------------------------------------------------------------- |
| name | name of the executed function |
| module | module name of the executed function |
| lineno | line number of the executed function |
| ncall | number of times the executed function is called. |
| nactualcall | number of times the executed function is called, excluding the recursive calls. |
| builtin | bool, indicating whether the executed function is a builtin |
| ttot | total time spent in the executed function |
| tsub | total time spent in the executed function, excluding subcalls |
| tavg | per-call average total time spent in the executed function. |
| index | unique id for the YFuncStat object |
| full_name | unique full name of the executed function |
|
