the example may need to change about aiozipkin HOT 11 CLOSED

Another question: Can we only send data the time is bigger than 100ms. Because there are so many request and I only care about the slow request. so the span with little delay(such as 1ms) may don't need to send to the collector (which is also a service need resource to hold all the request). The rate option may be different with what I want.

from aiozipkin.

Hi. For such specific tasks it is better to override the transport class i think. But implementation of the filtering of the spans by their parameters seems to me a good idea.
It can be combined with #39.
@jettify what do you think about this?

from aiozipkin.

it seems to me that the call asyncio.sleep at the end is also superfluous. because tracer.close() calling transport._send()

from aiozipkin.

Good catch! I have not updated this example, after implementing transport more properly, also agree on asyncio.sleep should be removed.

@jiamo would you like to create PR with fix?

from aiozipkin.

As for question, I am not sure how to implement time filtering, consider following scenario:
We have 3 chained spans with times 1ms, 1ms and 101ms, we leave only third span due to filter, as result we also loose important information a full chain of calls that leaded to this particular span.

If you want to see response time of one particular endpoint, than other statistics and monitoring tools are more suitable (like https://github.com/graphite-project). Value of distributed tracing in seeing chain of calls and all interactions with other services.

from aiozipkin.

In this case We have 3 chained spans with times 1ms, 1ms and 101ms What I wish is to collect them all because the total cast = 1 + 1 + 101 is big. (I don't looked into the source code. If it send data to collector at the close (will always be called at my on_response middleware in sanic) Then the filter is meaningful.

from aiozipkin.

from aiozipkin.

Example fixed in 326e966

from aiozipkin.

@konstantin-stepanov what do you think about context manger API around az.create?

async with az.create(zipkin_address, endpoint, sample_rate=1.0) as tracer
    with tracer.new_trace(sampled=True) as span:
        span.name('root_span')

I just wondering if it is useful outside test...

from aiozipkin.

I like this idea. This simplifies the code and it looks more readable.

As for useful, it depends on applicatuon architecture. I think It will be perfect in MVP

from aiozipkin.

Created issue to track this idea #99

from aiozipkin.