OpenCensus and OpenTracing have merged into OpenTelemetry!



ocredis is an OpenCensus instrumented wrapper for Andy McCurdy’s popular Python Redis client redis-py

ocredis is a drop-in replacement for redis-py (which it uses underneath) and for each method that performs a network call, it creates a span and collects metrics such as latency, errors, key lengths, value lengths.

To demonstrate how much of a drop-in replacement it is for redis-py, from the getting started guide in the redis-py README

>>> import redis
>>> r = redis.Redis(host='localhost', port=6379, db=0)
>>> r.set('foo', 'bar')
>>> r.get('foo')

you can now instead just do this

>>> import ocredis
>>> r = ocredis.OcRedis(host='localhost', port=6379, db=0)
>>> r.set('foo', 'bar')
>>> r.get('foo')

which is almost identical to the original statement as seen by this diff below.

< >>> import redis
< >>> r = redis.Redis(host='localhost', port=6379, db=0)
> >>> import ocredis
> >>> r = ocredis.OcRedis(host='localhost', port=6379, db=0)


You will need the following:

Before proceeding we’ll need to firstly install the following

Requirement Guide
Redis server
Prometheus Prometheus codelab
Zipkin Zipkin codelab

Installing it

ocredis is available on PyPi at and can be installed simply by running:

pip install ocredis


ocredis provides both traces and metrics per method that invokes the network.


Each method is traced and on any exception, the underlying span.Status will be set to indicate the error.


Using OpenCensus, the following metrics are present

Metric Search suffix Unit Aggregation Tags
Latency redispy/latency ms Distribution “error”, “method”, “status”
Calls redispy/calls 1 Count “error”, “method”, “status”
Key lengths redispy/key_length By Distribution “error”, “method”, “status”
Value lengths redispy/value_length By Distribution “error”, “method”, “status”

Enabling observability

After enabling one of the Python exporters, please don’t forget to add


End to end example

In the example below, we’ll have a CLI app that:

Source code

#!/usr/bin/env python

import ocredis

from opencensus.trace.tracer import Tracer
from opencensus.trace.exporters.zipkin_exporter import ZipkinExporter
from opencensus.trace.samplers import always_on
from opencensus.common.transports import async_
from opencensus.stats import stats
from opencensus.stats.exporters import prometheus_exporter as prometheus

def create_opencensus_exporters_and_tracer():
    # Create the Prometheus metrics exporter.
    statsm = stats.Stats()
    view_manager = statsm.view_manager
    pexp = prometheus.new_stats_exporter(prometheus.Options(namespace='ocredispy', port=8000))

    # Register the defined ocredis views.

    # Create the exporter that'll upload our traces to Zipkin.
    zexp = ZipkinExporter(service_name='pysearch', transport=async_.AsyncTransport)

    tracer_init_args = dict(exporter=zexp,
            # Always sampling for demo purposes.

    return tracer_init_args

def main():
    r = ocredis.OcRedis(host='localhost', port=6379, db=10)
    tracer_init_args = create_opencensus_exporters_and_tracer()
    while True:
            query = raw_input('$ ')
            response = do_search(r, query, **tracer_init_args)
            print('> ' + response + '\n')
        except Exception as e:
            print('Caught exception %s'%(e))
        except KeyboardInterrupt as e:

def do_search(client, query, **tracer_init_kwargs):
    tracer = Tracer(**tracer_init_kwargs)
    with tracer.span('Search') as span:
        span.add_annotation('Searching', query=query)

        # Check Redis if we've already memoized the response.
        response = client.get(query)

        if response is not None: # Cache hit
            span.add_annotation('Cache hit', store='redis', client='redis-py')
            print('Cache hit! Now deleting it to make for a cache miss later')
            # Clear the response so that the next search will return a cache-miss.

        else:  # Cache miss
            span.add_annotation('Cache miss', store='redis', client='redis-py')
            print('Cache miss! Now processing and memoizing it to make for a cache hit later')

            # Now process the result and memoize it.
            response = query.upper()
            client.set(query, response)

        return response

if __name__ == '__main__':

Running the code

Having started Redis-Server, we can now get the code running.

In a fresh Python shell, we can run


which after such keyboard inputs, should produce such output

$ ./ 
No handlers could be found for logger "opencensus.stats.aggregation"
$ what is this
Cache miss! Now processing and memoizing it to make for a cache hit later

$ what is that
Cache miss! Now processing and memoizing it to make for a cache hit later

$ what is this
Cache hit! Now deleting it to make for a cache miss later


Running Zipkin

Please check out this guide for how to install and start Zipkin.

Running Prometheus

Please check out this guide for how to install and start Prometheus.

We shall use the Prometheus configuration file below, saved as prom.yaml

# Saved as prom.yaml
  - job_name: 'pysearch'

    scrape_interval: 5s
       - targets: ['localhost:8000']

and finally start Prometheus like this:

prometheus --config.file=prom.yaml

Examining traces

On navigating to the Zipkin UI at http://localhost:9411

All traces

All traces

Cache hit

Cache hit

Cache hit detail

Cache hit detail

Cache miss

Cache miss

Cache miss detail

Cache miss detail

Examining metrics

On navigating to the Prometheus UI at http://localhost:9090

All metrics

All metrics

Rate of calls


Rate of calls

Latency p95th

histogram_quantile(0.95, sum(rate(ocredispy_redispy_latency_bucket[5m])) by (method, error, le))

p95th percentile for latencies

Key lengths p95th

histogram_quantile(0.95, sum(rate(ocredispy_redispy_key_lengths_bucket[5m])) by (method, error, le))

p95th for value lengths

Value lengths p95th

histogram_quantile(0.95, sum(rate(ocredispy_redispy_value_lengths_bucket[5m])) by (method, error, le))

p95th for value lengths




Resource URL
redis-py project home on Github
ocredispy on PyPi
ocredispy on Github