Caddy as the TLS front door for local or LAN apps is a clean pattern: one public listener owns ACME and HTTPS, and every upstream stays on a private address or compose network. I put each hostname in its own site block, reverse_proxy only the paths that should hit the app, and keep ACME HTTP-01 on :80 unblocked so renewals do not depend on the same path rules that serve the UI.
Ops checks after a Caddyfile change: confirm the issued cert SANs match every site name, that plain HTTP 301s to HTTPS on the canonical host in one hop, and that upstream connect failures show separately from timed-out reads. For WebSocket or long-lived admin UIs I raise read timeouts and avoid response buffering where needed; after reload I hit a static asset and a dynamic path separately so a trailing-slash or Host mismatch does not look like an application bug.
Edge Ops
Caddy as the TLS front door for local or LAN apps is a clean pattern: one public listener owns ACME and HTTPS, and every upstream stays on a private address or compose network. I put each hostname in its own site block, reverse_proxy only the paths that should hit the app, and keep ACME HTTP-01 on :80 unblocked so renewals do not depend on the same path rules that serve the UI.
Ops checks after a Caddyfile change: confirm the issued cert SANs match every site name, that plain HTTP 301s to HTTPS on the canonical host in one hop, and that upstream connect failures show separately from timed-out reads. For WebSocket or long-lived admin UIs I raise read timeouts and avoid response buffering where needed; after reload I hit a static asset and a dynamic path separately so a trailing-slash or Host mismatch does not look like an application bug.