Highlights 🌟

Monitoring 💻

Now Jina enables you to monitor the performance of your Executors and the complete Flow with the help of Prometheus and allows you to build a Grafana dashboard to gain better insights on how your system works.

from jina import Flow

f = Flow(monitoring=True).add().add()

with f:
	f.block()

#4636 #4635 #4638 #4672 #4689 #4687 #4735 #4738 #4737 #4758 #4752 #4768

Robustness 💪

We have improved the robustness of the system and the network communication between microservices in a Flow and have put a lot of work on getting better information of potential failures in the system.

Changes include:

• Adding a timeout parameter for the Gateway for its grpc communication with Executors #4660
• Ensuring a graceful shutdown of Runtimes in Kubernetes to avoid requests loss #4682
• Adding an extensive test for handling Pod failues in Kubernetes #4743
• Better error reporting of errors happening in the Head of a sharded Executor #4640
• Better error reporting when Flow fails to start #4652

UI improvements 💇

Jina comes with a better UI and a better logging system to help the user get better information from the logs in the best format based on rich.

#4614 #4697 #4724 #4736 #4739

Other changes

• Remove unnecessary argument from Executors runtime_backend (#4644)
• Pass environment variables to Gateway and Head Runtimes (#4646)
• Add resource layer with gpu in docker-compose files when to_docker_compose is called for Services requiring it (#4718)
• You can have inmutable tags when pushing Executors into the Hub (#4676)
• target_executor parameter is considered a regex pattern and not a exact string. All Executors matching this regex will receive the request. (#4756)
• Gateway only sends request to Executor if the endpoint is exposed by it (#4756)

Bug fixes 🐛

• Avoid leaking file descriptors when client sends requests (#4734)
• Fix flow plotting when private Executor is used (#4643)
• Executor could not start processes internally (#4692)
• Deadlock on Executor start due to ModuleLock error (#4693)
• Flow early stop when startup failure (#4726)

Breaking changes ⚠️

target_executor is treated as a regex pattern and not as a string.

Before:

from docarray import DocumentArray
from jina import Flow, Executor, requests


class Exec(Executor):

   @requests
   def foo(self, docs, **kwargs):
       for doc in docs:
           doc.text += 'calling foo '


f = Flow().add(name='exec', uses=Exec).add(name='exec_after', uses=Exec)

with f:
   res = f.post(on='/', inputs=DocumentArray.empty(1), target_executor='exec')
   print(res.texts)

Before, it would give:

['calling foo ']

Now:

['calling foo calling foo ']

❤️ Thanks to the community for all the support and hard work!