Envoy uses the Google Test framework for tests, including Google Mock for mocks and matchers. See the documentation on those pages for information on the various classes, macros, and matchers that Envoy uses from those frameworks.
Envoy contains an integration testing framework, for testing downstream-Envoy-upstream communication. See the framework's README for more information.
Envoy includes some custom Google Mock matchers to make test expectation statements simpler to write and easier to understand.
Tests that a HeaderMap argument contains exactly one header with the given key, whose value satisfies the given expectation. The expectation can be a matcher, or a string that the value should equal.
Examples:
EXPECT_THAT(response->headers(), HeaderValueOf(Headers::get().Server, "envoy"));
using testing::HasSubstr;
EXPECT_THAT(request->headers(),
HeaderValueOf(Headers::get().AcceptEncoding, HasSubstr("gzip")));
Tests that a HeaderMap argument has the provided HTTP status code. The status code can be passed in as a string or an integer.
Example:
EXPECT_THAT(response->headers(), HttpStatusIs("200"));
HeaderMapEqualRef
tests that two HeaderMap
arguments are equal.
HeaderMapEqual
is the same, but it compares two pointers to HeaderMap
s, and
the matcher's argument must be a HeaderMapPtr
.
Example:
EXPECT_THAT(response->headers(), HeaderMapEqualRef(expected_headers));
Tests that one HeaderMap
argument contains every header in another
HeaderMap
.
Examples:
EXPECT_THAT(response->headers(), IsSubsetOfHeaders(allowed_headers));
EXPECT_THAT(response->headers(), IsSupersetOfHeaders(required_headers));
In Envoy production code, time and timers are managed via
Event::TimeSystem
,
which provides a mechanism for querying the time and setting up time-based
callbacks. Bypassing this abstraction in Envoy code is flagged as a format
violation in CI.
In tests we use a derivation
Event::TestTimeSystem
which adds the ability
to sleep or do a blocking timed wait on a condition variable. There are two
implementations of the Event::TestTimeSystem
interface:
Event::TestRealTimeSystem
, and Event::SimulatedTimeSystem
. The latter is
recommended for all new tests, as it helps avoid flaky tests on slow machines,
and makes tests run faster.
Typically we do not want to have both real-time and simulated-time in the same
test; that could lead to hard-to-reproduce results. Thus both implementations
have a mechanism to enforce that only one of them can be instantiated at once.
A runtime assertion occurs if an Event::TestRealTimeSystem
and
Event::SimulatedTimeSystem
are instantiated at the same time. Once the
time-system goes out of scope, usually at the end of a test method, the slate
is clean and a new test-method can use a different time system.
There is also Event::GlobalTimeSystem
, which can be instantiated in shared
test infrastructure that wants to be agnostic to which TimeSystem
is used in a
test. When no TimeSystem
is instantiated in a test, the Event::GlobalTimeSystem
will lazy-initialize itself into a concrete TimeSystem
. Currently this is
TestRealTimeSystem
but will be changed in the future to SimulatedTimeSystem
.