[9.0.0] Classify UnknownHostException as retriable for repository

[bazel-io]

UnknownHostException can represent a transient failure in DNS.
Permit this exception to trigger the retry logic for the download loop.

Closes #28023.

PiperOrigin-RevId: 854154072
Change-Id: I84bca74b00af1bbbb322ecc3f8697ed252dbfe64

Commit 8fe8d55

[gemini-code-assist]

Code Review

This pull request aims to make UnknownHostException retryable during repository downloads, which is a good improvement for handling transient DNS failures. The approach is to classify UnknownHostException as retryable in the DownloadManager and adjust exception handling in HttpConnector.

However, the implementation in HttpConnector incorrectly uses addSuppressed instead of initCause to wrap the original exception. This prevents the DownloadManager's retry logic from detecting the UnknownHostException as it only checks the exception cause chain. I've left a comment with a suggested fix.

[gemini-code-assist]

The retry logic in DownloadManager.shouldRetryDownload inspects the exception's cause chain to find a retryable exception. Using addSuppressed(e) will not place the original UnknownHostException in the cause chain, so it won't be detected for a retry.

To ensure the exception can be caught by the retry logic, you should use initCause(e) to properly chain the exceptions.

Suggested change

	IOException httpException = new UnrecoverableHttpException(message);
	httpException.addSuppressed(e);
	throw httpException;
	IOException httpException = new UnrecoverableHttpException(message);
	httpException.initCause(e);
	throw httpException;