Skip to content

platform9/openstack-opentracing

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

35 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Build Status PyPI version

openstack-opentracing

This library provides a simple (alpha quality) implementation of a wsgi middleware that can be inserted into OpenStack APIs and can provide rudimentary tracing.

Why

Finding performance issues as at macro level is very difficult in a distributed system and it helps to find which service is to blame. Opentracing has emerged as a standard with Zipkin & Jaeger as two primary implementation.

For the purposes of this project we have explored Jaegertracing as the framework for the experiment.

Jaegertracing

More information is available here, here is a quick overview of the overall architecture:

Architecture

How to

Install Jager tracing all in one

docker run --name jaeger  -p 5775:5775/udp -p 5778:5778 -p 14268:14268 -p 6831:6831/udp -p 6832:6832/udp -p 16686:16686 jaegertracing/all-in-one:latest

Important: The current jaeger_middleware is configured to use remote controlled sampling. It will only be turned on when the jaeger collector service is running (see above), this is to make sure that no sampling is done unless there is a listener.

Install openstack-opentracing

Activate your virtualenv

source /opt/pf9/nova/bin/activate
cd /opt/pf9/nova/bin/
./python ./pip install openstack-opentracing

Configure your service

Edit your paste.ini file and insert a new filter and add the filter to the pipeline

[filter:jaeger_v1]
paste.filter_factory = openstack_opentracing.jaeger_middleware:JaegerMiddleware.factory
config.logging = False
config.local_agent.reporting_host = 127.0.0.1
config.local_agent.reporting_port = 32769
config.local_agent.sampling_port = 32770
config.local_agent.sampling_host = 127.0.0.1
service_name = Server1
validate = True

[pipeline:public_api]
# The last item in this pipeline must be public_service or an equivalent
# application. It cannot be a filter.
pipeline = sizelimit url_normalize request_id build_auth_context jaeger_v1 token_auth admin_token_auth json_body ec2_extension user_crud_extension pf9_mfa_auth_extension_v2 public_service

Restart your service

sudo service openstack-nova-api restart

Wait for a minute (default sampling refresh interval) and start making request.

Check the UI

Connect to http://:16686/ to check the UI. Make some nova-api requests and your tracing will show up automatically

I SSH into the DU and use the port forwarding

ssh -L 14268:127.0.0.1:14268 -L 16686:127.0.0.1:16686 pf9-test.platform9.net

Screenshots

Span Example1 Span Example2

Code

There is only one file which is interesting jaeger_middleware, which provides the middleware. Check the test directory for a working example of a distributed tracing example which uses the middleware and uses oslo.* libraries. This will give you a good starting point.

Build

Builds are done through travis-ci (see the build link on top of this page). Build steps include pip install, but also includes an integration test that starts 3 servers and checks for spans recorded in the jaeger server.

Ugly stuff

  • Most of the code is to make sure 'eventlet ' services like OpenStack can work with jaeger_client library primarly because jaeger_client assumes tornado. This is ugly and would need to change as jaeger_client rids itself of tornado dependency