tg4perfetto 0.0.4

A python project

tg4perfetto

Simple python library for generating your own perfetto traces for your application.

Python application tracing

Example code (see tg4perfetto/example_profile.py for the code)

import tg4perfetto
import threading

# A "stat" counter.  This can be used to log integers or floating-point stats.
count_stats = tg4perfetto.count("num_active_threads")

# Normally, tracks are assigned to its thread's default track.  This creates a custom track
custom_track = tg4perfetto.track("TOP_TRACK")

# Log the event on a thread-default track.  trace_func_args logs the arguments put on the
# function, so be careful
@tg4perfetto.trace_func_args
def merge(x, x1, x2):
    # ... (omitted)

def merge_sort_wrapper(x, flow_id):
    # Instant events
    tg4perfetto.instant("START_THREAD", incoming_flow_ids = [flow_id[0]])
    custom_track.instant("START_THREAD", incoming_flow_ids = [flow_id[1]])

    return merge_sort(x)

def merge_sort_threaded(x):
    # Instant event.  Has two "flow ID" which can be used for connecting two events.
    flow_ids = tg4perfetto.instant("INVOKE_THREAD", num_outgoing_flow_ids = 2)

    t = threading.Thread(target=merge_sort_wrapper, args=(x,flow_ids))
    t.start()
    return (x, t)


# Log the event on a thread-default track.  Unlike trace_func_args, trace_func
# only logs the call in/out events.
@tg4perfetto.trace_func
def merge_sort(x):
    l = len(x)

    if l < 4096: return sorted(x)

    if l < 40000:
        x1 = merge_sort(x[:int(l/2)])
        x2 = merge_sort(x[int(l/2):])
    else:
        # Using count_stats to increment and decrement
        count_stats.increment(1)
        x1, t1 = merge_sort_threaded(x[:int(l/2)])
        count_stats.increment(1)
        x2, t2 = merge_sort_threaded(x[int(l/2):])
        t1.join()
        count_stats.increment(-1)
        t2.join()
        count_stats.increment(-1)
    
    return merge(x, x1, x2)

# Log the event on the given specific track.
@tg4perfetto.trace_func(custom_track)
def validate(xarray):
    # ... (omitted)

if __name__ == "__main__":
    # Start logging.  Logging stops when this goes out of scope.
    with tg4perfetto.open("tg4p.perfetto-trace"):

        # Use a custom track.
        # to put into the default track. just use tg4perfetto.trace(...) instead.
        with custom_track.trace('SORT').get_outgoing_flow_ids(1) as out_flow_id:
            xarray = [ (17 * x + 8) % 100 for x in range(100000) ]
            xarray = merge_sort(xarray)

            # Create one flow ID from the current track.  We can create flow IDs before closing the track.
            p = out_flow_id[0]

        # Set the incoming flow ID (optional, set only if there are any).
        with custom_track.trace('VALIDATE').set_incoming_flow_ids([p]):
            tg4perfetto.instant("CHECKING", {"final_result": xarray})
            validate(xarray)
            print("Done")

This will generate a trace file named "tg4p.perfetto-trace" which can be read from perfetto.

Custom packet generation

Example code (see tg4perfetto/example.py for the code)

# Packets can be created out-of-order.  This is because perfetto is designed to process out-of-order traces
# and reads all packets at once, rearranges them, and then visualizes it at once.
tgen = TraceGenerator(sys.argv[1])
pid = tgen.create_group("aaa", "example_track")
pid.open(100, "SOME_TRACK")
# "Flow" packet.  this will create an arrow from here to "open" event down there (400ns)
pid.close(250, [4])

# Global counter track
tid = tgen.create_counter_track("bbb")
tid.count(0, 3)
tid.count(200, 5)
tid.count(400, 7)
tid.count(700, 2)

# Counter track within the "aaa" group"
tid = pid.create_counter_track("bbb")
tid.count(0, 2)
tid.count(200, 4)
tid.count(400, 5)
tid.count(700, 1)

tid = pid.create_track("ddd")
tid.open(100, "WXX")
# another "flow" packet.
tid.close(300, [3])

tgen.flush()

pid = tgen.create_group("vvv")
tid = pid.create_counter_track("bbb2")
tid.count(0, 2)
tid.count(300, 400)
tid.count(400, 500)
tid.count(700, 1000)

tid = pid.create_track("ddd2")
tid2 = pid.create_track("ddd3")

tid2.instant(200, "WXYZ")
tid.open(222, "XXX")
tid2.open(300, "WXX3", {"aaa":"bbb", "ccc":"ddd"})
tid2.instant(300, "ABCDE", {"aaa": "bbb", "ccc": "xxx"})
tid.close(333)
# receives an arrow from the packet above.  this can be either from an instant event or a normal event.
tid2.open(400, "WXX4", {"aaa":"bbb", "ccc":"ddd"}, [3, 4])
tid2.instant(400, "ABCDE")

# Some annotation on instant event
tid2.instant(600, "ADE", {"aaa": "abc", "ccc": "xxx", "eee" : {"aaa": "abc", "ccc": "ddd"}})
tid2.close(670, [2])

# very complex annotations!
tid2.instant(700, "ADE2", {
    "aaa": "abc",
    "ccc": [1, 2, 3, 4, "a", "b", {"abcdef" : "fdsa", "ggg": True}],
    "eee" : {
        "aaa": "abc",
        "ccc": True,
        "eee": {
            "fff": "ggg",
            "hhh": 0x1234567
        }
    },
    "jjj": "kkk"
}, [2])
tid2.close(900, [1])
tid.open(900, "WXX2", {"aaa":"bbb", "ccc":"ddd"}, [1])
tid.close(1000)

pid4 = tgen.create_group("abc.2")
tid4 = pid4.create_group("XX")
t1 = tid4.create_track()
t2 = tid4.create_track()
t1.open(100, "X")
t2.open(300, "Y")
t1.close(500)
t2.close(600)

Example output: