When I'm using the following technologies:
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.
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
.
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.
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)
- TBA
- added an option to metricize the queries
- added support for MockTracker
- a warning will be shown with unsupported tracing mechanism
- improved reporting when arguments is a dict