Fix memory leak in AsyncClient

[pufit]

Summary

It's known that ThreadPoolExecutor doesn't deallocate memory without .shutdown being called. Python's GC doesn't stop running thread pools, so the code like

while True:
    client = await get_async_client(...)
    ...
    client.close()

will be leaking memory since each new AsyncClient instance will have a new executor.

Closes #424

Checklist

Delete items not relevant to your PR:

• Unit and integration tests covering the common scenarios were added
• A human-readable description of the changes was provided to include in CHANGELOG
• For significant changes, documentation in https://github.com/ClickHouse/clickhouse-docs was updated with further explanations or tutorials

[genzgd]

Thanks for digging into this! My only question is whether this can be simpler by just shutting down the executor synchronously. It looks like you added the __aenter__ and __aexit__ methods to ensure this autocloses when using it as an AsyncContext?

[genzgd]

Does this have to be async? Can we just call the executor.shutdown method synchronously?

[pufit]

Thanks for digging into this! My only question is whether this can be simpler by just shutting down the executor synchronously. It looks like you added the __aenter__ and __aexit__ methods to ensure this autocloses when using it as an AsyncContext?

This way is more Pythonic. Usually, you open and close clients inside the event loop. If you try to wait for executor shutdown in sync mode it will hang the loop. If you close the thread pool without wait=True, you won't be able to guarantee the graceful shutdown of your application.

Here is an example of how it is usually done in other libraries: https://github.com/aio-libs/aiohttp/blob/master/examples/curl.py#L11