| 1061 | } |
| 1062 | |
| 1063 | func (c *turnWorkspaceContext) getWorkspaceConn(ctx context.Context) (workspacesdk.AgentConn, error) { |
| 1064 | if c.server.agentConnFn == nil { |
| 1065 | return nil, xerrors.New("workspace agent connector is not configured") |
| 1066 | } |
| 1067 | |
| 1068 | for attempt := 0; attempt < 2; attempt++ { |
| 1069 | c.mu.Lock() |
| 1070 | currentConn, staleRelease := c.getWorkspaceConnLocked() |
| 1071 | // Capture agentID in the same lock section as |
| 1072 | // currentConn to prevent a TOCTOU race with |
| 1073 | // concurrent clearCachedWorkspaceState calls. |
| 1074 | agentID := c.agent.ID |
| 1075 | c.mu.Unlock() |
| 1076 | |
| 1077 | // Status check on cache hit: re-fetch the agent |
| 1078 | // row so we see the latest heartbeat rather than |
| 1079 | // a potentially stale cached copy. |
| 1080 | if currentConn != nil { |
| 1081 | if agentID != uuid.Nil { |
| 1082 | freshAgent, err := c.server.db.GetWorkspaceAgentByID(ctx, agentID) |
| 1083 | if err != nil { |
| 1084 | c.server.logger.Warn(ctx, "failed to re-fetch agent for status check", |
| 1085 | slog.F("agent_id", agentID), |
| 1086 | slog.Error(err), |
| 1087 | ) |
| 1088 | // On DB error the check re-runs on the |
| 1089 | // next tool call. |
| 1090 | } else if _, disconnected := agentDisconnectedFor( |
| 1091 | c.server.clock.Now(), |
| 1092 | freshAgent, |
| 1093 | c.server.agentInactiveDisconnectTimeout, |
| 1094 | ); disconnected { |
| 1095 | c.clearCachedWorkspaceState() |
| 1096 | continue |
| 1097 | } |
| 1098 | } |
| 1099 | return currentConn, nil |
| 1100 | } |
| 1101 | if staleRelease != nil { |
| 1102 | staleRelease() |
| 1103 | } |
| 1104 | |
| 1105 | chatSnapshot, agent, err := c.ensureWorkspaceAgent(ctx) |
| 1106 | if err != nil { |
| 1107 | return nil, err |
| 1108 | } |
| 1109 | if err := c.externalAgentPreflightError(ctx, chatSnapshot, agent); err != nil { |
| 1110 | return nil, err |
| 1111 | } |
| 1112 | |
| 1113 | // Wrap the dial in a timeout to bound the time spent |
| 1114 | // waiting for an unreachable agent. The timeout scopes |
| 1115 | // only dialWithLazyValidation, not ensureWorkspaceAgent |
| 1116 | // or the post-dial binding steps. |
| 1117 | dialCtx, dialCancel := context.WithTimeoutCause(ctx, c.server.dialTimeout, errChatDialTimeout) |
| 1118 | dialResult, err := dialWithLazyValidation( |
| 1119 | dialCtx, |
| 1120 | agent.ID, |