- Low level plumbing for resources is still retained for now.
- Retain "Workload" terminology over "Service".
- Revert "Destination" terminology back to "Upstream".
- Remove TPROXY support as it only worked for v2.
Conceptually renaming the following topology terms to avoid confusion with v2 and to better align with it:
- ServiceID -> ID
- Service -> Workload
- Upstream -> Destination
This updates the testing/deployer (aka "topology test") framework to allow for a
v2-oriented topology to opt services into enabling TransparentProxy. The restrictions
are similar to that of #19046
The multiport Ports map that was added in #19046 was changed to allow for the
protocol to be specified at this time, but for now the only supported protocol is TCP
as only L4 functions currently on main.
As part of making transparent proxy work, the DNS server needed a new zonefile
for responding to virtual.consul requests, since there is no Kubernetes DNS and
the Consul DNS work for v2 has not happened yet. Once Consul DNS supports v2 we should switch over. For now the format of queries is:
<service>--<namespace>--<partition>.virtual.consul
Additionally:
- All transparent proxy enabled services are assigned a virtual ip in the 10.244.0/24
range. This is something Consul will do in v2 at a later date, likely during 1.18.
- All services with exposed ports (non-mesh) are assigned a virtual port number for use
with tproxy
- The consul-dataplane image has been made un-distroless, and gotten the necessary
tools to execute consul connect redirect-traffic before running dataplane, thus simulating
a kubernetes init container in plain docker.
This updates the testing/deployer (aka "topology test") framework to conditionally
configure and launch catalog constructs using v2 resources. This is controlled via a
Version field on the Node construct in a topology.Config. This only functions for a
dataplane type and has other restrictions that match the rest of v2 (no peering, no
wanfed, no mesh gateways).
Like config entries, you can statically provide a set of initial resources to be synced
when bringing up the cluster (beyond those that are generated for you such as
workloads, services, etc).
If you want to author a test that can be freely converted between v1 and v2 then that
is possible. If you switch to the multi-port definition on a topology.Service (aka
"workload/instance") then that makes v1 ineligible.
This also adds a starter set of "on every PR" integration tests for single and multiport
under test-integ/catalogv2
R.B. Boyer