Skip to content

Commit

Permalink
docs: spike of info about how to use deployer topology tests (#19576)
Browse files Browse the repository at this point in the history
Initial developer facing docs on how to write deployer (topology) integration tests.
  • Loading branch information
rboyer committed Nov 8, 2023
1 parent da2a572 commit e823d6a
Showing 1 changed file with 199 additions and 1 deletion.
200 changes: 199 additions & 1 deletion test-integ/README.md
Original file line number Diff line number Diff line change
@@ -1,3 +1,201 @@
# test-integ

Go integration tests for consul. `/test/integration` also holds integration tests; they need migrating.
Go integration tests for consul. `/test/integration` also holds integration tests; they need migrating.

These should use the [testing/deployer framework](../testing/deployer) to bring
up some local testing infrastructure and fixtures to run test assertions against.

Where reasonably possible, try to bring up infrastructure interesting enough to
be able to run many related sorts of test against it, rather than waiting for
many similar clusters to be provisioned and torn down. This will help ensure
that the integration tests do not consume CPU cycles needlessly.

## Prerequisites

Before you can run these tests, a fresh dev build should be created and tagged
for automatic use by tests using `utils.TargetImages()`. This can be done by
the toplevel consul make target:

make test-deployer-setup

You can run the entire set of deployer integration tests using:

make test-deployer # everything except peering_commontopo
make test-deployer-peering # just peering_commontopo

You can also run them one by one if you like:

go test ./catalogv2 -run TestBasicL4ExplicitDestinations -v

You can have the logs stream unbuffered directly to your terminal which can
help diagnose stuck tests that would otherwise need to fully timeout before the
logs would print by also setting the `NOLOGBUFFER=1` environment variable.

## Getting started

Deployer tests have three main parts:

1. Declarative topology description.
2. Launching the infrastructure defined by that description.
3. Making test assertions about the infrastructure.

Some tests may also have an optional _mutation_ phase followed by additional
assertions. These are only needed if the test needs to observe a reaction in
the system to a change in the environment or configuration.

### Topology description

Test authors craft a declarative description of the infrastructure necessary to exist for the test.
These are also referred to as a "topology".

These are comprised of 4 main parts:

- **Images**: The set of docker images and specific versions that will be used
by default if not overridden on each Cluster or Node.

- Consul CE
- Consul Enterprise
- Consul Dataplane
- Envoy Proxy

- **Networks**: The non-overlapping networks that should exist for use by the Clusters.

- **Clusters**: The unique Consul clusters that should exist.

- **Nodes**: A "box with ip address(es)". This should feel a bit like a VM or
a Kubernetes Pod as an enclosing entity.

- **Services/Workloads**: The list of service instances (v1) or workloads
(v2) that will execute on the given node. v2
Services will be implied by similarly named
workloads here unless opted out. This helps
define a v1-compatible topology and repurpose it
for v2 without reworking it.

- **Services** (v2): v2 Service definitions to define explicitly, in addition
to the inferred ones.

- **InitialConfigEntries** (v1): Config entries that should be created as
part of the fixture and that make sense to
include as part of the test definition,
rather than something created during the
test assertion phase.

- **InitialResources** (v2): v2 Resources that should be created as part of
the fixture and that make sense to include as
part of the test definition, rather than
something created during the test assertion
phase.

- **Peerings**: The peering relationships between Clusters to establish.

In the [topoutil](./topoutil) package there are some helpers for defining
common sets of nodes or services like Consul Servers, Mesh Gateways, or [fortio
servers](https://github.com/fortio/fortio)

#### Useful topology concepts

Consul has a lot of independent configurables that can greatly increase the
testing configuration space required to flush out any bugs. The topology
definition was designed to be easily "exploded" to create testing microcosms on
a variety of axes:

- agentful (clients) vs agentless (dataplane)
- tenancies (partitions, namespaces)
- locally or across a peering
- catalog v1 or v2 object model

Since the topology is just a declarative struct, a test author could rewrite
any one of these attributes with a single field (such as `Node.Kind` or
`Node.Version`) and cause the identical test to run against the other
configuration. With the addition of a few `if enterprise {}` blocks and `for`
loops, a test author could easily write one test of a behavior and execute it
to cover agentless, agentful, non-default tenancy, and v1/v2 in a few extra
lines of code.

#### Non-optional security settings

The test framework always enables ACLs in default deny mode and provisions
minimal-permission tokens automatically to the various containers that need
them.

TLS certificates are similarly minted and distributed to all components that
need them.

### Launching a topology

There is a [sprawltest](../testing/deployer/sprawl/sprawltest) package that has
utilities to bring up a topology in the context of a Go test. This is basically a one-liner:

sp := sprawltest.Launch(t, config)

After this line returns you will have a handle (`sp`) to the running cluster
and can use it to get ready-made api clients, http clients, gRPC resource
client, or test sockets open to a variety of the topology components for use in
authoring test code.

This helper will rig up a `t.Cleanup` handler that will destroy all resources
created during the test. This can be opted-out of by setting the
`SPRAWL_KEEP_RUNNING=1` environment variable before running the tests.

### Test assertions

Typical service mesh tests want to ensure that use of a service from another
service behaves in a certain way. Because the entire set of components is known
declaratively, we can process it into a flat list of known source/destination
relationships:

ships := topology.ComputeRelationships()

This works hand-in-hand with the topology concepts mentioned above to
programmatically verify independent subunits of a topology that may exist (this
is helpful for things like testing multiple tenancy configurations without
duplicating all of the assertion code).

This can also be pretty printed to the log for diagnostic purposes with:

t.Log(topology.RenderRelationships(ships))

Which looks like this:

$ NOLOGBUFFER=1 go test ./catalogv2/ -run TestBasicL4ExplicitDestinations -v
...(skipping a bunch of output)...
2023-11-08T11:48:04.395-0600 [INFO] TestBasicL4ExplicitDestinations: topology is ready for use: elapsed=33.510298357s
explicit_destinations_test.go:55: DOWN |node |service |port |UP |service |
dc1 |default/dc1-box2 |default/default/single-client |5000 |dc1 |default/default/single-server |
dc1 |default/dc1-box4 |default/default/multi-client |5000 |dc1 |default/default/multi-server |
dc1 |default/dc1-box4 |default/default/multi-client |5001 |dc1 |default/default/multi-server |
| | | | | |

=== RUN TestBasicL4ExplicitDestinations/relationship:_default/default/single-client_on_default/dc1-box2_in_dc1_via_:5000_=>_default/default/single-server_in_dc1_port_http
service.go:224: making call to http://10.238.170.5:5000
service.go:245: ...got response code 200
=== RUN TestBasicL4ExplicitDestinations/relationship:_default/default/multi-client_on_default/dc1-box4_in_dc1_via_:5000_=>_default/default/multi-server_in_dc1_port_http
service.go:224: making call to http://10.238.170.7:5000
service.go:245: ...got response code 200
=== RUN TestBasicL4ExplicitDestinations/relationship:_default/default/multi-client_on_default/dc1-box4_in_dc1_via_:5001_=>_default/default/multi-server_in_dc1_port_http-alt
service.go:224: making call to http://10.238.170.7:5001
service.go:245: ...got response code 200
2023-11-08T11:48:04.420-0600 [INFO] TestBasicL4ExplicitDestinations.tfgen: Running 'terraform destroy'...
--- PASS: TestBasicL4ExplicitDestinations (40.60s)
--- PASS: TestBasicL4ExplicitDestinations/relationship:_default/default/single-client_on_default/dc1-box2_in_dc1_via_:5000_=>_default/default/single-server_in_dc1_port_http (0.01s)
--- PASS: TestBasicL4ExplicitDestinations/relationship:_default/default/multi-client_on_default/dc1-box4_in_dc1_via_:5000_=>_default/default/multi-server_in_dc1_port_http (0.01s)
--- PASS: TestBasicL4ExplicitDestinations/relationship:_default/default/multi-client_on_default/dc1-box4_in_dc1_via_:5001_=>_default/default/multi-server_in_dc1_port_http-alt (0.01s)
PASS
ok github.com/hashicorp/consul/test-integ/catalogv2 40.612s

There is a ready-made helper to assist with making common inquiries to Consul
and Envoy that you can create in your test:

asserter := topoutil.NewAsserter(sp)

asserter.UpstreamEndpointStatus(t, svc, clusterPrefix+".", "HEALTHY", 1)

## Examples

- `catalogv2`
- [Explicit L4 destinations](./catalogv2/explicit_destinations_test.go)
- [Implicit L4 destinations](./catalogv2/implicit_destinations_test.go)
- [Explicit L7 destinations with traffic splits](./catalogv2/explicit_destinations_l7_test.go)
- [`peering_commontopo`](./peering_commontopo)
- A variety of extensive v1 Peering tests.

0 comments on commit e823d6a

Please sign in to comment.