Storage: Add sensible connect (and maybe read) timeouts to avoid hanging

[crwilcox]

After some spelunking, started from reports like #5909 where users have reported hanging on requests, I have uncovered that we don't have read or connect timeouts for our default requests session on storage.

I think adding a connect timeout is warranted for sure, and possible having a read timeout, makes sense? urllib3 via request.session is what is controlling that which by default doesn't specify anything. https://urllib3.readthedocs.io/en/latest/reference/urllib3.util.html#module-urllib3.util.timeout

I propose a connection timeout of 1 minute. I am less sure of read timeout as that would vary a bit between crud and streaming operations.

cc: @frankyn

[tseaver]

Upload / download retries should tie into the RetryStrategy from google-resumable-media.

[tswast]

Retry support tracked in #7907.

There's an open PR at #8177, but not complete.

[tswast]

Per Tres's comment, BigQuery uses RetryStrategy here:

google-cloud-python/bigquery/google/cloud/bigquery/client.py

Lines 1603 to 1606 in c2df95b

	if num_retries is not None:
	upload._retry_strategy = resumable_media.RetryStrategy(
	max_retries=num_retries
	)

[crwilcox]

@tswast to some extent this isn't about exposing retries. Currently, there are no timeouts for connecting. This is problematic and can cause clients to hang. Is this handled special in retry strategy? The way the storage client is configured, we are using the defaults from urllib3 which has infinite timeout for both connect and read.

[tswast]

At least in GAPIC, the retry config sets a timeout as part of the request, though I'm not seeing it in the linked RetryStrategy class.

In BigQuery, we do have a separate timeout parameter for some operations. We should probably be more consistent about these.

google-cloud-python/bigquery/google/cloud/bigquery/client.py

Lines 1052 to 1054 in d72161d

	timeout_ms (int):
	(Optional) number of milliseconds the the API call should
	wait for the query to complete before the request times out.

[tseaver]

Notes to future self: docs for requests timeout handling.

[crwilcox]

Closing as I believe the PR in August resolved this.