Built-in Simple Executors and Reserved uses in Jina

What is Simple Executor?

In Jina we have several built-in some simple executors. They are called simple for two reasons:

  • It only requires a YAML config, no Python code is required;

  • It inherits from BaseExecutor directly, its logic thus fully relies on the drivers.

For example, the built-in _clear executor is defined as:

!BaseExecutor
with: {}
metas:
  name: clear
requests:
  on:
    [SearchRequest, TrainRequest, IndexRequest]:
      - !ReqPruneDriver {}
    ControlRequest:
      - !ControlReqDriver {}

It uses ReqPruneDriver to prune the request. That’s it.

Another example, in Flow API the join method (waiting multiple Pods to finish) is implemented with _pass Simple Executor, which is specified by

!BaseExecutor
with: {}
metas:
  name: merge
requests:
  on:
    [SearchRequest, TrainRequest, IndexRequest]:
      - !MergeDriver {}
    ControlRequest:
      - !ControlReqDriver {}

And in Flow API implementation:

def join(self, needs: Union[Tuple[str], List[str]], *args, **kwargs) -> 'Flow':
    """
    Add a blocker to the flow, wait until all peas defined in `needs` completed.

    :param needs: list of service names to wait
    :return: the modified flow
    """
    if len(needs) <= 1:
        raise FlowTopologyError('no need to wait for a single service, need len(needs) > 1')
    return self.add(name='joiner', uses='_pass', needs=needs, *args, **kwargs)

What are the reserved uses?

To help users quickly use these patterns, we reserved the following keywords for the uses. They all start with underscore.

Reserved Name Description
_clear Clear request body from a message
_logforward Like _pass, but print the message
_pass Pass the message to the downstream
_merge_topk Merge the top-k search results (both doc and chunk level) of all collected messages, often used in the tail of a Pod
_merge_topk_chunks Merge the top-k search results (chunk level only) of all collected messages, often used in the tail of a Pod
_merge_topk_docs Merge the top-k search results (doc level only) of all collected messages, often used in the tail of a Pod

How to use built-in Simple Executors?

You can directly use this executor by specifying --uses=_clear, or use it in --uses-after after collecting results from replicas.

Where ever you need to use uses in Jina, you can take any one from the table to fill in.