fix(database): drop .failed state so clean health-monitor teardown stops alerting connection lost

CHANGELOG.md

@@ -41,6 +41,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Keychain reads no longer collapse a cancelled Touch ID prompt, a failed biometric auth, or any unknown OSStatus into "not found". The `KeychainResult` enum now distinguishes `.userCancelled`, `.authFailed`, and `.error(OSStatus)` from `.notFound`, and the read paths in connection passwords, SSH profile secrets, AI provider keys, and the license key log each case with its own message. Previously a cancelled prompt looked identical to a missing entry, so the caller would treat the password as gone and silently re-save with an empty string on the next write, producing duplicate keychain entries or a connection saved with a blank password.
 - Terminal PTY writes retry on `EINTR` instead of treating any non-positive return as "we're done". A signal mid-write previously truncated the input the user typed; the loop would exit silently and the keystrokes were partially sent. The new path retries on `EINTR`, logs the byte position and errno on any other non-recoverable failure, and reports a return value of zero distinctly so the cause is visible in Console.
 - MCP HTTP transport no longer writes an empty body when JSON encoding of the response envelope fails. Five sites in `MCPInboundExchange` and `MCPHttpRequestRouter` previously fell back to `Data()`, sending zero bytes to the client which then saw a protocol violation and either disconnected or hung. The encode-failure paths now log and substitute a static `{"jsonrpc":"2.0","id":null,"error":{"code":-32603,"message":"internal_error"}}` envelope; the pairing-exchange success path falls back to `internalServerError` with a small JSON error body.
+- Closing the last window for a connection no longer flashes "Connection lost" and clears that session's cached schema. The health monitor's reconnect loop previously transitioned to `.failed` when its task was cancelled (clean teardown), and the session-level observer treated `.failed` as a real error: it overwrote `session.status` with the lost-connection alert and called `clearCachedData()`. The `.failed` state was never reachable through any non-cancellation path, so it has been removed from `HealthState` along with the orphaned `resetAfterManualReconnect` reset method that only existed to reset from it. Cancellation now logs cleanly and returns without touching session state.
 - Result-grid cells on rows marked for deletion keep their dropdown / date / JSON / blob chevron visible at reduced opacity instead of hiding it, so the cell type is still legible while clearly inactive. Click on the dimmed chevron is a no-op; FK arrow navigation is unchanged. Matches the macOS HIG "disabled appearance" guideline.
 - Foreign key navigation from a table with unsaved edits opens the referenced table in a new window tab to preserve the edit buffer. Closing that new tab no longer wipes the original tab's data grid. Previously the new tab's teardown broadcast a connection-scoped event that other coordinators on the same connection received, causing them to release their cell data.
 - Tables sidebar refreshes automatically after a successful SQL import; the refresh notification now fires after the success sheet's dismissal animation, so the main window is key when the observer runs (#1114)

TablePro/Core/Database/ConnectionHealthMonitor.swift

@@ -17,7 +17,6 @@ extension ConnectionHealthMonitor {
         case healthy
         case checking
         case reconnecting(attempt: Int) // 1-based attempt number
-        case failed
     }
 }
 
@@ -122,15 +121,6 @@ actor ConnectionHealthMonitor {
         await task?.value
     }
 
-    /// Resets the monitor to `.healthy` after the user manually reconnects.
-    ///
-    /// Call this when an external reconnection succeeds so the monitor resumes
-    /// normal periodic pings instead of staying in `.failed` state.
-    func resetAfterManualReconnect() async {
-        Self.logger.info("Manual reconnect succeeded, resetting to healthy for connection \(self.connectionId)")
-        await transitionTo(.healthy)
-    }
-
     // MARK: - Health Check
 
     /// Performs a single health check cycle.
@@ -176,9 +166,10 @@ actor ConnectionHealthMonitor {
     /// Attempts to reconnect with exponential backoff.
     ///
     /// Uses initial delays of 2s, 4s, 8s, then continues doubling up to a
-    /// 120-second cap. Transitions to `.failed` only when the task is cancelled
-    /// (i.e., the monitor is stopped). This ensures the monitor keeps trying
-    /// indefinitely rather than giving up after a fixed number of attempts.
+    /// 120-second cap. Loops indefinitely until either a reconnect succeeds
+    /// (transitions to `.healthy` and returns) or the monitoring task is
+    /// cancelled (returns without a state transition, since cancellation is
+    /// clean teardown initiated by `stopMonitoring`).
     private func attemptReconnect() async {
         var attempt = 0
 
@@ -187,7 +178,7 @@ actor ConnectionHealthMonitor {
 
             let delay = backoffDelay(for: attempt)
 
-            Self.logger.warning("Reconnect attempt \(attempt) for connection \(self.connectionId) — waiting \(delay)s")
+            Self.logger.warning("Reconnect attempt \(attempt) for connection \(self.connectionId), waiting \(delay)s")
             await transitionTo(.reconnecting(attempt: attempt))
 
             try? await Task.sleep(for: .seconds(delay))
@@ -208,8 +199,7 @@ actor ConnectionHealthMonitor {
             Self.logger.warning("Reconnect attempt \(attempt) failed for connection \(self.connectionId)")
         }
 
-        Self.logger.error("Reconnect cancelled after \(attempt) attempts for connection \(self.connectionId)")
-        await transitionTo(.failed)
+        Self.logger.debug("Reconnect loop cancelled after \(attempt) attempts for connection \(self.connectionId)")
     }
 
     /// Computes the backoff delay for a given attempt number (1-based).
@@ -249,8 +239,6 @@ actor ConnectionHealthMonitor {
             return .debug
         case .reconnecting:
             return .default
-        case .failed:
-            return .error
         }
     }
 }

TablePro/Core/Database/DatabaseManager+Health.swift

@@ -82,19 +82,12 @@ extension DatabaseManager {
                     case .reconnecting(let attempt):
                         Self.logger.info("Reconnecting session \(id) (attempt \(attempt))")
                         if case .connecting = self.activeSessions[id]?.status {
-                            // Already .connecting — skip redundant write
+                            // Already .connecting, skip redundant write
                         } else {
                             self.updateSession(id) { session in
                                 session.status = .connecting
                             }
                         }
-                    case .failed:
-                        Self.logger.error(
-                            "Health monitoring failed for session \(id)")
-                        self.updateSession(id) { session in
-                            session.status = .error(String(localized: "Connection lost"))
-                            session.clearCachedData()
-                        }
                     case .checking:
                         break  // No UI update needed
                     }