Review of 082: a store killed mid-conversation is an outage and a wrong password is not; one report holds the queue at most two minutes, then is let go loudly; shutdown does not wait out the pause

internal/inventory/reachable.go

@@ -2,7 +2,8 @@ package inventory
 
 import (
 	"errors"
-	"strings"
+	"io"
+	"net"
 
 	"github.com/jackc/pgx/v5/pgconn"
 )
@@ -11,25 +12,41 @@ import (
 // that it answered no.
 //
 // The difference decides whether something worth keeping is kept or lost. The store is recreated
-// when the foundation is adopted (ADR 0078), and for those seconds every write fails with "the
-// database system is starting up" or a refused connection; a caller that treats that like a
-// refusal throws away what it was writing — a node's report of the apply that caused the restart
-// was lost exactly so, and the mesh never heard from the node again (novox/hq issue 082).
+// when the foundation is adopted (ADR 0078), and for those seconds every write fails; a caller
+// that treats that like a refusal throws away what it was writing — a node's report of the apply
+// that caused the restart was lost exactly so, and the mesh never heard from the node again
+// (novox/hq issue 082).
 //
-// Unreachable is a connection that failed, and the server classes that mean "not now": 08
-// (connection exception) and 57 (operator intervention — starting up, shutting down, admin
-// shutdown, cannot connect now). Everything else is an answer, and asking again gets the same one.
+// Unreachable is the connection failing and the server saying "not now". The connection failing
+// is a network error, a connection that ended mid-conversation (a store killed rather than
+// stopped gives a bare unexpected EOF), a timeout, or anything pgx marks safe to retry. The server
+// saying "not now" is a connection exception (class 08) or it shutting down or starting up (57P01,
+// 57P02, 57P03). Everything else is an answer, and asking again gets the same one: a wrong
+// password, a database that does not exist, a statement cancelled, a constraint. Those are
+// checked first, even inside a failed connection, because a connection that failed on a wrong
+// password would otherwise read as a network fault and be asked again for ever.
 func Unreachable(err error) bool {
 	if err == nil {
 		return false
 	}
-	var connect *pgconn.ConnectError
-	if errors.As(err, &connect) {
-		return true
-	}
 	var pg *pgconn.PgError
 	if errors.As(err, &pg) {
-		return strings.HasPrefix(pg.Code, "08") || strings.HasPrefix(pg.Code, "57")
+		switch pg.Code {
+		case "57P01", "57P02", "57P03":
+			return true
+		}
+		return len(pg.Code) == 5 && pg.Code[:2] == "08"
 	}
-	return false
+	if errors.Is(err, io.ErrUnexpectedEOF) || errors.Is(err, io.EOF) {
+		return true
+	}
+	if pgconn.Timeout(err) || pgconn.SafeToRetry(err) {
+		return true
+	}
+	var network net.Error
+	if errors.As(err, &network) {
+		return true
+	}
+	var connect *pgconn.ConnectError
+	return errors.As(err, &connect)
 }