impl(bigquery): job level retry for queries

[alvarowolfx]

When a query job fails with a retryable BigQuery error (for example backendError or rateLimitExceeded), the client now automatically creates a new job attempt with backoff instead of failing immediately. Move all logic for query execution away from builder layer (RunQuery) to the execution layer.

Towards #5844 #6218

[gemini-code-assist]

Code Review

This pull request introduces a RetryContext to handle job-level retries and re-issuances for BigQuery queries, refactoring the query execution path to delegate to this context and updating Query::until_done to retry on transient failures. Feedback on the changes highlights a critical compilation error in until_done where self is partially moved inside a loop, and suggests avoiding destructuring self at the loop's start. Additionally, it is recommended to pass self by value in reissue to avoid an unnecessary clone of the context.

[codecov]

Codecov Report

❌ Patch coverage is 98.99329% with 3 lines in your changes missing coverage. Please review.
✅ Project coverage is 96.27%. Comparing base (4ff2683) to head (49eaaf9).

Files with missing lines	Patch %	Lines
src/bigquery/src/query/execution.rs	97.50%	2 Missing ⚠️
src/bigquery/src/query/query_handle.rs	99.43%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #6318      +/-   ##
==========================================
+ Coverage   96.26%   96.27%   +0.01%     
==========================================
  Files         282      282              
  Lines       73304    73508     +204     
==========================================
+ Hits        70563    70772     +209     
+ Misses       2741     2736       -5     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[alvarowolfx]

/gemini review

[gemini-code-assist]

Code Review

This pull request introduces a RetryContext to manage job-level retries and re-issuances for BigQuery queries, refactoring PostQueryExecutor and InsertJobExecutor to return raw responses and updating Query::until_done to handle retries in a loop. The review feedback focuses on minor Rust idiomatic improvements and optimizations, such as avoiding an unnecessary double reference (&&str) when passing project_id, removing an unnecessary .to_string() allocation on set_project_id, and using pattern matching instead of .take() on the owned retry_context option.

[alvarowolfx]

/gemini review

[gemini-code-assist]

Code Review

This pull request introduces a RetryContext to manage job-level retries and re-issuances for BigQuery queries, refactoring PostQueryExecutor and InsertJobExecutor to return raw responses and updating Query::until_done to handle retryable errors. Feedback on the changes suggests deriving Debug for the public RunQuery struct to adhere to the repository style guide, and addresses an off-by-one error in RetryableJobErrors that allows an extra retry attempt beyond the configured limit.

[gemini-code-assist]

According to the repository style guide, all public types should implement both Debug and Clone to ensure interoperability. Since RunQuery is a public struct, please derive Debug for it (e.g., by changing #[derive(Clone)] to #[derive(Clone, Debug)] on line 31).

References

1. Public types should implement both Debug and Clone to ensure interoperability. ^(link)
2. Public types, such as request builders, should implement both Debug and Clone to ensure interoperability.

[alvarowolfx]

will do that in a separated PR. Need to figure out how to add #[derive(Debug)] on the generated types

[gemini-code-assist]

The attempt_limit in RetryableJobErrors actually allows attempt_limit + 1 total attempts due to an off-by-one check in its on_error implementation (which checks state.attempt_count >= self.attempt_limit). Since attempt_count starts at 0 for the first failure, an attempt_limit of 3 will allow attempts 0, 1, 2, and 3 (total of 4 attempts). Consider changing the check to state.attempt_count + 1 >= self.attempt_limit to strictly adhere to the attempt limit while keeping attempt_count 0-indexed for backoff calculations.

[alvarowolfx]

will fix in a separated PR