Cloud networks
A connector inside a VPC or VNet gives you reachability into it without
peering, transit gateways, or a conversation about who re-addresses their
10.0.0.0/8.
Why not peering#
Peering joins two address spaces. That means a CIDR negotiation, a routing change, a security-group review, and reachability far broader than the one service you wanted. It also fails outright when both sides use the same private range.
A connector sidesteps all of it: the VPC keeps its addressing, nothing is routed between estates, and the only reachable things are the services you publish.
Deployment pattern#
Run the connector as a small instance pair in a private subnet with a NAT gateway or egress-only route for outbound 443. Give the instances an instance profile with no permissions beyond what your own tooling needs — the connector does not call cloud APIs.
1 2 3 4 5 6 7 | |
Use attested identity from the cloud instance identity document so the connector's identity is tied to the actual instance rather than to a token that was copied into user-data. See Identity.
Kubernetes#
In-cluster, run the connector as a Deployment with two replicas and publish services by cluster DNS name:
1 2 3 4 | |
The connector resolves the target through the cluster's own DNS, so Service and Endpoint changes are picked up without touching Proxility.
Multi-account and multi-cloud#
One realm per account or subscription keeps delegation clean and matches how most organisations already separate environments. Cross-account access is then a cross-realm rule, which is explicit and shows up in review — see Tenants and realms.
What this does not solve#
Proxility publishes services. It does not give a VPC arbitrary L3 reachability into another network, and it is not a substitute for a transit gateway if you genuinely need routed connectivity between many subnets. See Overview.