Remote Executors

JinaD enables deploying Executors on remote machines. We can pass the remote host details using hostand port_jinad arguments in the Flow syntax. Here are a few examples of using remote Executors.

Executors from Hub

Local (jinahub+docker)

from jina import Flow
f = Flow().add(uses='jinahub+docker://AdvancedExecutor')

Remote (jinahub+docker)

from jina import Flow
f = Flow().add(uses='jinahub+docker://AdvancedExecutor',
               host='1.2.3.4',
               port_jinad=8000)

Local (jinahub)

from jina import Flow
f = Flow().add(uses='jinahub://AdvancedExecutor',
               install_requirements=True)

Remote (jinahub)

from jina import Flow
f = Flow().add(uses='jinahub://AdvancedExecutor',
               install_requirements=True,
               host='1.2.3.4',
               port_jinad=8000)

Executors on different hosts

Local

from jina import Flow
f = (
    Flow()
    .add(name='encoder',
         uses='jinahub+docker://AdvancedExecutor1')
    .add(name='indexer',
         uses='jinahub+docker://AdvancedExecutor2')
)

Remote

from jina import Flow
f = (
    Flow()
    .add(name='encoder',
         uses='jinahub+docker://AdvancedExecutor1',
         host='1.2.3.4',
         port_jinad=8000)
    .add(name='indexer',
         uses='jinahub+docker://AdvancedExecutor2',
         host='2.3.4.5',
         port_jinad=8000)
)

File structures with remote Executors

Important

Read the best practices about managing repository structure.

For a simple project structure, for example,

.
├── project
│   ├── config.yml              # Defines `py_modules: my_executor.py`
│   └── my_executor.py          # Defines MyExecutor

Using YAML

from jina import Flow
PATH_TO_EXECUTOR_DIRECTORY = '/path/to/project/directory'
f = Flow().add(uses='config.yml',
               host='1.2.3.4',
               port_jinad=8000,
               upload_files=PATH_TO_EXECUTOR_DIRECTORY)

Using py_modules

from jina import Flow
PATH_TO_EXECUTOR_DIRECTORY = '/path/to/project/directory'
f = Flow().add(uses='MyExecutor',
               host='1.2.3.4',
               port_jinad=8000,
               py_modules='my_executors.py'
               upload_files=PATH_TO_EXECUTOR_DIRECTORY)

Suppose you have a complex project structure, for example,

.
├── project
│   ├── config.yml              # Defines `py_modules: executors/__init__.py`
│   └── executors
│       ├── __init__.py         # Imports MyExecutor from `my_executors.py`
│       ├── helper.py
│       ├── my_executor.py      # Defines MyExecutor
│       └── utils
│           ├── __init__.py
│           ├── data.py
│           └── io.py
│   ├── flow.yml
│   ├── requirements-index.txt  # Includes your pip dependencies
│   ├── requirements.txt

Using YAML

from jina import Flow
PATH_TO_EXECUTOR_DIRECTORY = '/path/to/project/directory'
f = Flow().add(uses='config.yml',
               host='1.2.3.4',
               port_jinad=8000,
               upload_files=PATH_TO_EXECUTOR_DIRECTORY)

Using py_modules

from jina import Flow
PATH_TO_EXECUTOR_DIRECTORY = '/path/to/project/directory'
f = Flow().add(uses='MyExecutor',
               host='1.2.3.4',
               port_jinad=8000,
               py_modules='executors/__init__.py'
               upload_files=PATH_TO_EXECUTOR_DIRECTORY)

Hint

1. Always pass the path to the directory via upload_files.

2. Always pass relative filepath wrt to the project in py_modules or uses or config.yml.

3. Keep your requirements.txt files on the root level.

Using GPU with any remote Executor

Without GPU

from jina import Flow
f = Flow().add(uses='TFExecutor',
               host='1.2.3.4',
               port_jinad=8000,
               ...)

With GPU

from jina import Flow
f = Flow().add(uses='TFExecutor',
               host='1.2.3.4',
               port_jinad=8000,
               ...,
               gpus='all')