python-cassandra-jaeger

When do I use it?

When I'm using the following technologies:

• cassandra-driver
• cassandra-jaeger-tracing

And you want to attach your traces to Cassandra's requests.

Note: so far only Jaeger tracing and MockTracer are supported. I mean, theoretically every opentracing-compatible framework is supported, but only Jaeger's tracing enables to select which query to trace or not.

In case that you're not using Jaeger or MockTracer, every single query will be sent with a tracing request, which may negatively impact your performance, Cassandra-wise.

If you have an issue with that, please file an issue with a description of what it is that you're using for tracing.

A relevant warning will be shown as well.

Usage

Just do the following:

from python_cassandra_jaeger import SessionTracer
from cassandra.cluster import Cluster

from .tracing import tracer

c = Cluster(['127.0.0.1'])
s = c.connect('keyspace')
st = SessionTracer(s, tracer)

st.execute('SELECT * FROM table')

And you keep on utilizing st instead of s. This will automatically execute tracing when your span is being traced.

Just remember to match your tracing key with Cassandra's JAEGER_TRACE_KEY.

What does it do?

It hijacks session.execute, grabs the current span, determines if sampling is active for this span (currently works only with Jaeger) and if it is so, starts a new child span, attaches it's ID to custom_payload and enables tracing for the request.

Additionally metricizing your queries

Added in v0.4: You can also metrify how long did your queries take to execute, say:

from satella.instrumentation.metrics import getMetric

met_sum = getMetric('cassandra.query.time.summary', 'summary')
s = c.connect('keyspace')
st = SessionTracer(s, tracer, metric=met_sum)

History

v0.4

• added an option to metricize the queries
• added support for MockTracker

v0.3

• a warning will be shown with unsupported tracing mechanism

v0.2

• improved reporting when arguments is a dict