can we avoid these local retry loops and build the txn aborted retry generically into our retry logic ?

there are several other retry codes that probably got deleted with a refactor/change ?

com.google.spanner.v1.ResultSet resultSet =
runWithRetries( // lost this ?
new Callable<com.google.spanner.v1.ResultSet>() {
@OverRide
public com.google.spanner.v1.ResultSet call() throws Exception {
return rpc.executeQuery(builder.build(), session.options);
}
});
These are all the codes we should be retrying on:
private static boolean isRetryable(ErrorCode code, @nullable Throwable cause) {
switch (code) {
case INTERNAL:
return hasCauseMatching(cause, Matchers.isRetryableInternalError);
case UNAVAILABLE:
return true;
case RESOURCE_EXHAUSTED:
return SpannerException.extractRetryDelay(cause) > 0;
default:
return false;
}
}

The above mentioned method was removed during the refactor to use Gax for retries. Retries of single gRPC calls are now automatically handled and retried by Gax (i.e. an rpc call that temporarily returns UNAVAILABLE will automatically be retried by Gax without the need for any custom code in the Spanner library).

Transaction retries are still handled by custom loops in the Spanner library, as these need to retry application logic and possibly multiple gRPC calls. We could refactor these to use the standard Gax com.google.cloud.RetryHelper instead of our own custom retry loops. I've changed this PR for PDML to use the RetryHelper.

I'll submit a separate PR for using RetryHelper for normal transaction retries.