fix(reporting): Add retry logic and configurable chunk size for BigQuery writes

[erinlimbogoogle]

Summary

This PR addresses transient BigQuery rate limit errors (specifically 429 TooManyRequests: too many table update operations) that occur when running large evaluation datasets or executing multiple evaluations concurrently.
It implements two main changes in BigQueryReporter:

1. Configurable Chunk Size: Replaces the hardcoded chunk size of 250 with a configurable chunk_size parameter in the reporter config (defaulting to 500). Increasing the chunk size reduces the number of load jobs (table updates) generated per run.
2. Transient Error Retries: Wraps the BigQuery load job in a retry loop with exponential backoff (up to 5 attempts, backing off by a factor of 2 starting at 2 seconds). It catches TooManyRequests (429) and ServiceUnavailable (503) errors, preventing temporary rate-limit spikes from failing the entire evaluation.

Changes

• evalbench/reporting/bqstore.py:
  • Defined _DEFAULT_CHUNK_SIZE = 500.
  • Initialized self.chunk_size from reporting_config in BigQueryReporter.__init__.
  • Updated store to use self.chunk_size when splitting the results dataframe.
  • Added a retry loop around self.client.load_table_from_dataframe and job.result().

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[prernakakkar-google]

/gcbrun

[helloeve]

nit: I wonder if we can use retry.Retry?

from google.api_core import retry
api_retry_strategy = retry.Retry(
    predicate=retry.if_exception_type(
        InternalServerError, 
        ServiceUnavailable, 
        TooManyRequests
    ),
    initial=2.0,   # Seconds to wait before the first retry
    multiplier=2.0, # Exponential backoff multiplier
    maximum=10.0,   # Maximum delay between retries
    timeout=60.0    # Stop retrying after this many total seconds
)
job.result(retry=api_retry_strategy)