How to use logstream
This is a guide to start using logstream
in your e2e testing.
Requirements
-
The SYSTEM_NAMESPACE
environment variable must be configured. Many of the
knative test scripts already define this, and in some places (e.g. serving)
randomize it. However, to facilitate usage outside of CI, you should consider
including a package like
this
and linking it like
this
-
Test resources must be named with
test.ObjectNameForTest(t)
-
At the start of your test add: t.Cleanup(logstream.Start(t))
-
To enable logcapture from containers across multiple namespaces configure
SYSTEM_NAMESPACE to contains a csv list of namespaces
(knative-serving,knative-test ??????{}
). Specific, well known containers
that do not produce key decorated logs (see detailed description below) need
to be enumerated in WellKnownContainers in stream.go.
With that, you will start getting logs from the processes in the system
namespace interleaved into your test output via t.Log
.
How it works
In Knative we use zap.Logger
for all of our logging, and most of those loggers
(e.g. in the context of a reconcile) have been decorated with the "key" of the
resource being processed. logstream
simply decodes these structured log
messages and when the key
matches the naming prefix that ObjectNameForTest
uses, it includes it into the test's output.
Integrating in Libraries.
When a shared component is set up and called from reconciliation, it may have
it's own logger. If that component is dealing with individual resources, it can
scope individual log statements to that resource by decorating the logger with
its key like so:
logger := logger.With(zap.String(logkey.Key, resourceKey))
Now, any log statements that the library prints through this logger will appear
in the logstream!
For an example of this pattern, see
the knative/networking prober library.