route-proxy: an ACME account belongs to the authority that issued it

autocert keeps its account key at one fixed name, `acme_account+key`, in whatever
directory it is given, and reuses it for ever. That is right while the authority
stays the same and silently wrong the moment it does not. Re-initialising the
internal CA makes a new authority with a new root: it has never heard of the
account in the cache, rejects every use of it, and autocert has no path back from
that. Nothing is re-registered, no order ever reaches the CA, and issuance stops
with nothing saying why — until somebody guesses that deleting the cache directory
by hand is the answer.

Name the directory after the authority instead of sharing one between all of them:
a digest of the ACME directory URL and the root this proxy was told to verify it
with. A re-initialised CA has a new root, the mesh delivers it as a changed bundle,
the proxy restarts on that file and lands in a directory with no account in it, so
autocert registers afresh and orders again. The healing is that "is this account
still valid" never has to be asked — an account is only ever found where it is
still valid, which needs no error codes, no probe at startup and no network call
that can itself fail.

It closes a latent one of the same shape: pointing ACME_DIRECTORY at production
after testing against staging reused the staging account, because the cache had no
idea the two were different.

Trailing whitespace around the delivered root is not a new authority — the mesh
writes that file, and a newline coming or going must not throw away an account.
Old directories are left on disk, unused: they hold the only copy of certificates
that may still be valid, and this program is not the thing that should decide a
certificate is finished with.

A proxy already holding certificates orders them once more on the first start after
this, because its account moves. Free against the lab's own CA and against staging;
one issuance per name against a public authority.

novox/hq ADR 0056

Claude-Session: https://claude.ai/code/session_01LrgweAeERJYBg88c5cKDzF

examples/route-proxy/main.go

@@ -20,9 +20,12 @@
 package main
 
 import (
+	"bytes"
 	"context"
+	"crypto/sha256"
 	"crypto/tls"
 	"crypto/x509"
+	"encoding/hex"
 	"encoding/json"
 	"fmt"
 	"log"
@@ -31,6 +34,7 @@ import (
 	"net/http/httputil"
 	"net/url"
 	"os"
+	"path/filepath"
 	"sort"
 	"strings"
 	"sync"
@@ -198,20 +202,22 @@ func run() error {
 	// nothing trusts yet — the lab's, or an internal step-ca. Trusting it is a deliberate act and
 	// names a file, rather than the client being told to skip verification: *skip* would also
 	// apply on the day this points at a public issuer, and nothing would say so.
+	var root []byte
 	if bundle := strings.TrimSpace(os.Getenv("ACME_CA_BUNDLE")); bundle != "" {
-		pem, err := os.ReadFile(bundle)
+		read, err := os.ReadFile(bundle)
 		if err != nil {
 			return fmt.Errorf("ACME_CA_BUNDLE names %s and it cannot be read: %w", bundle, err)
 		}
+		root = read
 		// An empty bundle means the issuer's root is already in the system trust store — a public
 		// authority whose root ships with the OS, pointed at by a provider that serves an empty
 		// root (novox/hq ADR 0056). The mesh always writes the bundle file, so it exists and holds
 		// nothing; that is the signal to fall back to the system roots, the same as if nothing had
 		// named a bundle at all. A file that holds bytes but no certificate is still a
 		// misconfiguration and is refused, because there the operator meant to trust something.
-		if strings.TrimSpace(string(pem)) != "" {
+		if strings.TrimSpace(string(root)) != "" {
 			pool := x509.NewCertPool()
-			if !pool.AppendCertsFromPEM(pem) {
+			if !pool.AppendCertsFromPEM(root) {
 				return fmt.Errorf("%s holds no certificate this can trust", bundle)
 			}
 			client.HTTPClient = &http.Client{
@@ -220,13 +226,16 @@ func run() error {
 			}
 		}
 	}
+	// Where this authority's account and certificates are kept. Per authority, not per proxy — see
+	// forThisAuthority, which is what makes a re-initialised CA heal itself.
+	mine := forThisAuthority(cache, issuer(), root)
 	manager := &autocert.Manager{
-		Cache:      autocert.DirCache(cache),
+		Cache:      autocert.DirCache(mine),
 		Prompt:     autocert.AcceptTOS,
 		HostPolicy: onlyWhatTheMeshSaid(held),
 		Client:     client,
 	}
-	log.Printf("issuing from %s, for whatever the mesh routes here", issuer())
+	log.Printf("issuing from %s into %s, for whatever the mesh routes here", issuer(), mine)
 
 	// Port 80 answers the HTTP-01 challenge and goes on proxying everything else. The challenge
 	// must be answered *at the name being certified*, which is why issuance happens on the node
@@ -245,6 +254,35 @@ func run() error {
 	return server.ListenAndServeTLS("", "")
 }
 
+// forThisAuthority is where one ACME authority's account and certificates are kept.
+//
+// **A cached ACME account belongs to the authority that issued it, and nothing in the cache says
+// so** (novox/hq ADR 0056). autocert keeps its account key at one fixed name — `acme_account+key` —
+// in whatever directory it is given, and reuses it for ever. That is right while the authority stays
+// the same and silently wrong the moment it does not: an internal CA that is re-initialised is a new
+// authority with a new root, it has never heard of the account in the cache, and every attempt to
+// use it is rejected. autocert has no path back from that. Nothing is retried, nothing is
+// re-registered, no order ever reaches the CA — issuance simply stops, with no error anybody sees,
+// until a person deletes the directory by hand and finds out that was the answer.
+//
+// So the directory is named after the authority instead of being shared by all of them. The name is
+// a digest of the two things that identify one: the directory URL, and the root this proxy was told
+// to verify it with. Re-initialising the CA produces a new root; the mesh delivers it as a changed
+// bundle; this proxy restarts on that file and lands in a directory with no account in it, so
+// autocert registers afresh and orders again. **The healing is that the question "is this account
+// still valid" never has to be asked** — an account is only ever found where it is still valid.
+//
+// It also fixes a latent one of the same shape: pointing ACME_DIRECTORY at production after testing
+// against staging reused the staging account, because the cache had no idea they were different.
+//
+// The old directories stay on disk, unused. Left rather than deleted: they are the only copy of
+// certificates that may still be valid, and this program is not the thing that should decide a
+// certificate is finished with.
+func forThisAuthority(cache, directory string, root []byte) string {
+	sum := sha256.Sum256([]byte(directory + "\x00" + string(bytes.TrimSpace(root))))
+	return filepath.Join(cache, hex.EncodeToString(sum[:])[:16])
+}
+
 // newTable is an empty routing table.
 func newTable() *table {
 	return &table{to: map[string]*httputil.ReverseProxy{}, targets: map[string]string{}}