feat: register db.sql metrics

[deniseli]

Fixes #2285

Register the standard otel database metrics using: https://github.com/XSAM/otelsql
Found call sites by grepping for sql.Open and skipping all test code.

Sample:

ScopeMetrics #2
ScopeMetrics SchemaURL: 
InstrumentationScope github.com/XSAM/otelsql 0.32.0

Metric #0
Descriptor:
     -> Name: db.sql.latency
     -> Description: The latency of calls in milliseconds
     -> Unit: ms
     -> DataType: Histogram
     -> AggregationTemporality: Cumulative

HistogramDataPoints #0
Data point attributes:
     -> method: Str(sql.conn.begin_tx)
     -> status: Str(ok)
StartTimestamp: 2024-08-08 23:18:01.011762 +0000 UTC
Timestamp: 2024-08-08 23:18:11.012177 +0000 UTC
Count: 23
Sum: 17.082795
Min: 0.137584
Max: 3.233917

[alecthomas]

LGTM with some comments that need to be addressed.

[alecthomas]

That is a nice clean API.

[deniseli]

Looks like this is breaking tests right now because modulecontext's Build() calls internal/reflect's DeepCopy, which doesn't expect any values of kind=string that can't be cast to a string. The databases now include a bunch of metric.descOpt{...}s wrapping the string descriptions of each metric, so that's confusing the cloning. I'll pick this back up in the morning

[deniseli]

The alias in question: https://github.com/open-telemetry/opentelemetry-go/blob/c609b12d9815bbad0810d67ee0bfcba0591138ce/metric/instrument.go#L39

[deniseli]

New changes: 2 minor additions to internal/reflect/reflect.go:

• Added support for string aliases. Before, this would cast the src string as long as its Kind was string, but that would cause a panic when the true type was an alias. In this case, it was metric.descOpt.
• Added support for the list.List type, which is used by the global.meter type inside otelsql.otConnector. Before, DeepCopy would panic whenever a list.List was given as input with reflect.Set: value of type *list.List is not assignable to type *list.Element. This adds a special case for that type and performs a standard linked-list deep copy, which is easier than managing consistency across all the backpointers and parent pointers in that structure.

[alecthomas]

Why'd these get removed?

[deniseli]

oh what the heck! I'll undo all the changes to that file. Thanks for catching that