We have an issue where integration between this package and OpenTelemetry is generating a lot of warnings.

In google.cloud.sqlalchemy_spanner sqlalchemy_spanner.py when it executes do_execute it calls trace_call but before doing so it sets the trace_attributes including db.params which are the values of the SQL query.

In trace_call, it checks if HAS_OPENTELEMETRY_INSTALLED and if so it will add these db.params atttributes to the trace and performs the trace.

But, in the OpenTelemetry Python package, it will check all attributes. One of the check that is performed on the attributes is that if an attribute is a Sequence type, all elements of the sequence must be of the same type. If not, it will throw a warning. This checks happens in the Python OpenTelemetry (external to Google) package in file attributes/__init__.py in the function _clean_attribute (see checks and _logger.warning calls).

The issue is that the Google Spanner SQLAlchemy package will store it’s db.params in the various types that they represent (which is good on it's own). It will then send these to the OpenTelemetry package who will check they’re all of the same type and throw a warning if not. So for all SQL queries that contain multiple datatype, OpenTelemetry will generate a warning.

This results in a huge number of these warnings in our GCP Log Explorer.
This could be fixed in a number of ways including sending these values as strings when sending them to OpenTelemetry.

I've made some tests and if I change this package sqlalchemy_spanner/_opentelemetry_tracing.py file to use the following code:

if extra_attributes:
        if isinstance(extra_attributes, dict):
            for k, v in extra_attributes.items():
                if isinstance(v, collections.abc.Sequence):
                    extra_attributes[k] = [str(e) for e in v]

        attributes.update(extra_attributes)

I'm no longer receiving any warnings as all sequences values are converted to string before tracing.