// containing any run metadata information accumulated so far and clears this
// information.
// If async mode is enabled, this call blocks till all currently pending ops are
// done.
TF_CAPI_EXPORT extern void TFE_ContextExportRunMetadata(TFE_Context* ctx,
                                                        TF_Buffer* buf,

  ~DeviceThread();

  // Requests that the worker thread execute the specified operation. Blocks
  // until the previously pending operation (a StartExecute without a Join) has
  // finished, if any.
  //
  // `cancellation_manager` must live until after `Join` finishes and pending
  // `is_async` operations finish. In addition to allowing the caller to cancel
  // the operation, its `StartCancel` method will be called if op execution

  //
  // If `is_async=false` (constructor argument), `cancellation_manager` must
  // live until `Join` finishes. If `is_async=true` it must live until `Join` is
  // followed by `TFE_ContextAsyncWait` to clear pending operations. It will be
  // used to cancel all other operations if any fails.
  //
  // Set step_id to configure the step id used for rendezvous creation. step id

2. Running the CI scripts locally, as explained below
3. **Google employees only**: Google employees can use an internal-only tool
called "MLCI" that makes testing more convenient: it can execute any full CI job
against a pending change. Search for "MLCI" internally to find it.

You may invoke a CI script of your choice by following these instructions:

```bash
cd tensorflow-git-dir

// This call may not block for execution of ops enqueued concurrently with this
// call.
TF_CAPI_EXPORT extern void TFE_ExecutorWaitForAllPendingNodes(
    TFE_Executor*, TF_Status* status);

// When an error happens, any pending operations are discarded, and newly issued
// ops return an error. This call clears the error state and re-enables
// execution of newly issued ops.
//

TensorFlow in a multitenant design mixes the risks described above with the
inherent ones from multitenant configurations. The primary areas of concern are
tenant isolation, resource allocation, model sharing and hardware attacks.

### Tenant isolation

Since any tenants or users providing models, graphs or checkpoints can execute
code in context of the TensorFlow service, it is important to design isolation

  // will take ownership and maintain devices' lifetime.
  virtual Status AddDevices(std::vector<std::unique_ptr<Device>> devices) = 0;

  // Block until all pending nodes are finished.
  virtual Status AsyncWait() = 0;

  // Add a function (serialized FunctionDef protocol buffer) so that it can
  // be executed as an op. Return error if the function with the same name