Performance monitoring ideas?

Using Streamlit
1

Wondering if anyone in the community has had any luck with performance monitoring? We have a bunch of streamlit apps that we deploy via dokku. Some are fast, some are slow depending on what people are doing in the code. I’d like to make sure no one has stuff that takes more than a minute to load.

Has anyone had any luck tracking performance of streamlit apps? A lot of things like NewRelic track requests but with everything handled over websockets, there’s nothing to see.

The least sophisticated way I’ve tried is just using stdout and redirecting it into a log that I can parse later.

For example at the top of the streamlit app::

from datetime import datetime
start_time = datetime.now()

and at the end putting

print(f"my_app render: {(datetime.now() - start_time).total_seconds()}s")
2

Hey david! Welcome to the community !

I was going through the same problem a couple of weeks ago.
Apparently streamlit uses prometheus_client to export metrics at /metrics if you set --global.metrics=true with streamlit run like

streamlit run --global.metrics=true my_lit_app.py

you can create a init method for your prometheus metrics eg. Gauges, Summaries etc. and then use them across your application. Just make sure that you don’t create the metrics in the main file else you will keep getting Duplicate Timeseries errors. Then you can scrape the metrics using prometheus and store it in influx or the TSDB of your choice.

I visualize the metrics on grafana, you can use prometheus dashboard or something equivalent.

EDIT: Adding example,

# put this in a module like utils.py ?
class REGISTRY:
    inited = False
    @classmethod
    def get_metrics(cls):
        if cls.inited:
            return cls
        else:
            # Register your metrics here
            cls.REQUEST_TIME = Summary('some_summary', 'Time spent in processing request')
            cls.inited = True

# main script
from utils import REGISTRY

METRICS = REGISTRY.get_metrics()

@METRICS.REQUEST_TIME.time()
def i_sleep():
    sleep(1.0)

st.write("something")
i_sleep()

Load /metrics and look for some_summary_…

PS. managers love those graphs :smiley: Hope it helps ! :slight_smile:

2 Likes
Streamlit app view and statistics
3

Welcome @david!

Along with @ash2shukla’s answer, we do have an open issue for creating an easy-to-user profiler, though it hasn’t been a top priority:

4

Thanks @ash2shukla! That works well and is easy to push out across apps.