Ignore disappeared datasets during listing tables in BigQuery

[ebyhr]

No description provided.

[hashhar]

LGTM % nits

[hashhar]

Maybe a comment like in

trino/plugin/trino-base-jdbc/src/main/java/io/trino/plugin/jdbc/DefaultJdbcMetadata.java

Lines 600 to 603 in 1a7d82e

	catch (TableNotFoundException | AccessDeniedException e) {
	// table disappeared during listing operation or user is not allowed to access it
	// these exceptions are ignored because listTableColumns is used for metadata queries (SELECT FROM information_schema)
	}

would be helpful here otherwise it seems "incorrect" to do.

[hashhar]

BTW this seems to be a pre-existing case in JDBC metadata too. There we ignore only missing tables, not schemas.

[ebyhr]

As far as I checked JDBC module before sending this PR, it doesn't have the same issue. Can you point out the exact issue place? The DefaultJdbcMetadata snippet condition is little misleading (donwstream method won't throw TableNotFoundException), but the entire listing logic looks correct.

[hashhar]

It throws JDBC_ERROR wrapping the original exception that was encountered instead of ignoring failures during listing.

See BaseJdbcClient#getTableNames -

trino/plugin/trino-base-jdbc/src/main/java/io/trino/plugin/jdbc/BaseJdbcClient.java

Lines 226 to 228 in 338425d

	catch (SQLException e) {
	throw new TrinoException(JDBC_ERROR, e);
	}

[ebyhr]

The getTables() in the try block calls DatabaseMetaData.getTables() method, so it won't throw an exception even if the schema doesn't exist in my understanding. IdentifierMapping might lead to the failure though.

[hashhar]

Sorry if it was unclear. I meant that the existing JDBC code can fail if DatabaseMetaData#getTAbles fails which we know to be the case with MemSQL for example. It's purely theoretical but it can happen IIUC?

[hashhar]

Maybe we can use ifPresentOrElse here to avoid assignment and the if...else? Would it look better?