[WIP] Box complex types in AnyValue enum

[lalitb]

Fixes #1903
Design discussion issue (if applicable) #

Changes

Please provide a brief description of the changes here.

Merge requirement checklist

• CONTRIBUTING guidelines followed
• Unit tests added/updated (if applicable)
• Appropriate CHANGELOG.md files updated for non-trivial, user-facing changes
• Changes in public API reviewed (if applicable)

[lalitb]

Moving to draft - Even though AnyValue benchmarks shows improvement, the similar improvement is not reflected in tracing-appender benchmark, possibly for different reason. Looking into that.

[lalitb]

AnyValue benchmark results:

Benchmark	PR Time (ns)	Main Time (ns)
CreateOTelValueStaticString	3.7731	3.7642
CreateOTelAnyValueStaticString	4.4106	34.740
CreateOTelValueString	15.083	15.393
CreateOTelAnyValueString	16.076	41.257
CreateOTelValueInt	3.1472	3.1382
CreateOTelAnyValueInt	4.1100	33.097

[lalitb]

And the LogRecord benchmark. The benchmark for box'ed complex types has gone up slightly as expected (eg for simple-log-with-bytes/no-context , simple-log-with-map-any-value), there is visible improvement for simple types, though not to the scale of AnyValue benchmarks as in above comment.

Benchmark	PR Branch Time (ns)	Main Branch Time (ns)
simple-log/no-context	123.56	141.61
simple-log/with-context	123.84	140.62
simple-log-with-int/no-context	142.48	156.29
simple-log-with-int/with-context	143.36	155.16
simple-log-with-double/no-context	140.69	155.43
simple-log-with-double/with-context	147.05	153.54
simple-log-with-string/no-context	142.76	156.71
simple-log-with-string/with-context	143.79	154.11
simple-log-with-bool/no-context	144.85	161.52
simple-log-with-bool/with-context	144.18	160.25
simple-log-with-bytes/no-context	165.61	171.69
simple-log-with-bytes/with-context	172.59	173.45
simple-log-with-a-lot-of-bytes/no-context	168.16	175.58
simple-log-with-a-lot-of-bytes/with-context	170.43	174.53
simple-log-with-vec-any-value/no-context	204.47	219.25
simple-log-with-vec-any-value/with-context	207.93	218.56
simple-log-with-inner-vec-any-value/no-context	273.05	313.19
simple-log-with-inner-vec-any-value/with-context	273.08	287.55
simple-log-with-map-any-value/no-context	217.71	223.04
simple-log-with-map-any-value/with-context	219.62	223.06
simple-log-with-inner-map-any-value/no-context	300.22	298.58
simple-log-with-inner-map-any-value/with-context	301.72	300.53
long-log/no-context	124.05	141.42
long-log/with-context	123.79	140.95
full-log/no-context	95.383	116.95
full-log/with-context	111.75	125.99
full-log-with-4-attributes/no-context	200.20	179.25
full-log-with-4-attributes/with-context	202.43	176.71
full-log-with-9-attributes/no-context	338.03	305.05
full-log-with-9-attributes/with-context	337.06	280.23
full-log-with-attributes/no-context	430.26	419.97
full-log-with-attributes/with-context	428.65	413.89

[lalitb]

Updating benchmark with LogRecord creation for 4 attributes:

Benchmark	Main Repo (ns)	PR (box'ed) (ns)	PR (box'ed + smallVec) (ns)
full-log-with-4-attributes/no-context	165.94 - 166.51	159.96 - 160.40	110.21 - 110.36
full-log-with-4-attributes/with-context	186.79 - 187.43	176.67 - 176.97	133.45 - 133.91

[utpilla]

Updating benchmark with LogRecord creation for 4 attributes:

Benchmark Main Repo (ns) PR (box'ed) (ns) PR (box'ed + smallVec) (ns)
full-log-with-4-attributes/no-context 165.94 - 166.51 159.96 - 160.40 110.21 - 110.36
full-log-with-4-attributes/with-context 186.79 - 187.43 176.67 - 176.97 133.45 - 133.91

@lalitb I think we need to black_box the static string values provided in the log_record.add_attribute method to see the desired perf difference.

log_benchmark_group(c, "full-log-with-4-attributes", |logger| {
        let mut log_record = logger.create_log_record();
        ...
        log_record.add_attribute("name", black_box("my-event-name"));
        log_record.add_attribute("event.id", black_box(20));
        log_record.add_attribute("user.name", black_box("otel"));
        log_record.add_attribute("user.email", black_box("otel@opentelemetry.io"));
        logger.emit(log_record);
    });

I think the huge perf difference that shows up for simply creating AnyValue for static strings showed up because you wrapped the AnyValue creating with a black_box. I believe if you don't make any changes to the enum and run the create values from static string benchmarks without black_box, you would get comparable results already.

c.bench_function("CreateOTelAnyValueStaticStr", |b| {
        b.iter(|| {
            let _v = black_box(AnyValue::String("value1".into()));
        });
    });