(google-cloud-spanner) Built-in metrics fail when one process uses databases in different projects

Environment details

• OS type and version: Linux (also happens on macOS)
• Python version: 3.13
• google-cloud-spanner version: 3.68.0

Steps to reproduce

1. In one process, create two clients for databases in two different projects.
2. Run a query on each.
3. Wait a minute for the metrics export.

Code example

from google.cloud import spanner

a = spanner.Client(project="project-a")
b = spanner.Client(project="project-b")

a.instance("i").database("d").snapshot().execute_sql("SELECT 1")
b.instance("i").database("d").snapshot().execute_sql("SELECT 1")

What happens

The metrics export fails, and no metrics show up in Cloud Monitoring for either project:

InvalidArgument: 400 Field resource.labels.project_id had an invalid value of
"project-b": if present, must be the project number or ID in the request name
(projects/project-a).

Whichever client is created first decides where all metrics go, so everything from the other project is rejected — and it takes the first project's metrics down with it, since they're sent in the same request.

What I expected

Metrics for each database go to that database's project. Running against several projects is pretty normal, and right now it means no built-in metrics at all.

What I tried

SPANNER_DISABLE_BUILTIN_METRICS=true and Client(disable_builtin_metrics=True) both work, but only by giving up the metrics. As far as I can tell there's no way to pass in my own exporter or meter provider, so there's no way to keep them. Java has this (googleapis/google-cloud-java#13679) and there's an open request for Go (googleapis/google-cloud-go#11204), but I couldn't find anything similar here.

Comece executando o exemplo fornecido com dois projetos usando google-cloud-spanner 3.68.0 e inspecione o caminho de métricas integradas usado por spanner.Client. Compare o comportamento padrão com SPANNER_DISABLE_BUILTIN_METRICS e Client(disable_builtin_metrics=True) e, em seguida, rastreie como o projeto do primeiro cliente é reutilizado durante a exportação. O trabalho estará concluído quando as métricas de cada banco de dados chegarem ao seu próprio projeto sem desabilitar as métricas integradas.