feat: use millis in RPC metrics

[LesnyRumcajs]

Summary of changes

Changes introduced in this pull request:

Based on the feedback from Glif:

• changed the unit of RPC metrics to milliseconds, same as is reported in Lotus. This will facilitate cross-implementation performance testing,
• added unit of measure to the metric description.

Based on my subjective opinion (and so I'm open to reverting that):

• changed the histogram characteristics a bit, otherwise the majority of the methods would fall into all buckets. Should we change the bucket count to 5 and increase the factor to 10 @lemmih ?

Reference issue to close (if applicable)

Closes

Other information and links

# HELP rpc_processing_time Duration of RPC method call in milliseconds.
# TYPE rpc_processing_time histogram
rpc_processing_time_sum{method="Filecoin.ChainGetTipSetByHeight"} 0.039538
rpc_processing_time_count{method="Filecoin.ChainGetTipSetByHeight"} 1
rpc_processing_time_bucket{le="0.1",method="Filecoin.ChainGetTipSetByHeight"} 1
rpc_processing_time_bucket{le="1.0",method="Filecoin.ChainGetTipSetByHeight"} 1
rpc_processing_time_bucket{le="10.0",method="Filecoin.ChainGetTipSetByHeight"} 1
rpc_processing_time_bucket{le="100.0",method="Filecoin.ChainGetTipSetByHeight"} 1
rpc_processing_time_bucket{le="1000.0",method="Filecoin.ChainGetTipSetByHeight"} 1
rpc_processing_time_bucket{le="+Inf",method="Filecoin.ChainGetTipSetByHeight"} 1
rpc_processing_time_sum{method="Filecoin.ChainExport"} 35033.210394999995
rpc_processing_time_count{method="Filecoin.ChainExport"} 1
rpc_processing_time_bucket{le="0.1",method="Filecoin.ChainExport"} 0
rpc_processing_time_bucket{le="1.0",method="Filecoin.ChainExport"} 0
rpc_processing_time_bucket{le="10.0",method="Filecoin.ChainExport"} 0
rpc_processing_time_bucket{le="100.0",method="Filecoin.ChainExport"} 0
rpc_processing_time_bucket{le="1000.0",method="Filecoin.ChainExport"} 0
rpc_processing_time_bucket{le="+Inf",method="Filecoin.ChainExport"} 1
rpc_processing_time_sum{method="Filecoin.StateNetworkName"} 0.43903800000000006
rpc_processing_time_count{method="Filecoin.StateNetworkName"} 5
rpc_processing_time_bucket{le="0.1",method="Filecoin.StateNetworkName"} 3
rpc_processing_time_bucket{le="1.0",method="Filecoin.StateNetworkName"} 5
rpc_processing_time_bucket{le="10.0",method="Filecoin.StateNetworkName"} 5
rpc_processing_time_bucket{le="100.0",method="Filecoin.StateNetworkName"} 5
rpc_processing_time_bucket{le="1000.0",method="Filecoin.StateNetworkName"} 5
rpc_processing_time_bucket{le="+Inf",method="Filecoin.StateNetworkName"} 5
rpc_processing_time_sum{method="Filecoin.ChainHead"} 0.058253
rpc_processing_time_count{method="Filecoin.ChainHead"} 1
rpc_processing_time_bucket{le="0.1",method="Filecoin.ChainHead"} 1
rpc_processing_time_bucket{le="1.0",method="Filecoin.ChainHead"} 1
rpc_processing_time_bucket{le="10.0",method="Filecoin.ChainHead"} 1
rpc_processing_time_bucket{le="100.0",method="Filecoin.ChainHead"} 1
rpc_processing_time_bucket{le="1000.0",method="Filecoin.ChainHead"} 1
rpc_processing_time_bucket{le="+Inf",method="Filecoin.ChainHead"} 1

Change checklist

• I have performed a self-review of my own code,
• I have made corresponding changes to the documentation,
• I have added tests that prove my fix is effective or that my feature works (if possible),
• I have made sure the CHANGELOG is up-to-date. All user-facing changes should be reflected in this document.

[lemmih]

changed the histogram characteristics a bit, otherwise the majority of the methods would fall into all buckets. Should we change the bucket count to 5 and increase the factor to 10 @lemmih ?

That sounds great to me.

[lemmih]

Too bad .as_millis_f64() isn't stable.

[LesnyRumcajs]

Yeah, that is disappointing.

[ruseinov]

Could still use a millisecond constant from Duration instead of the magic number.

Suggested change

	.observe(start_time.elapsed().as_secs_f64() * 1000.0);
	.observe(start_time.elapsed().as_secs_f64() * Duration::MILLISECOND as f64);

[LesnyRumcajs]

No, we couldn't. That constant is nightly-only. https://doc.rust-lang.org/std/time/struct.Duration.html#associatedconstant.MILLISECOND

Even if it wasn't nightly, the suggested code would still not work. Conceptually, you'd convert, e.g., 3s to 3ms.

We could introduce a named constant, but given it's used exclusively here and it's pretty trivial, combined with the comment above, I'm not convinced it'd bring any value.

Do you agree?

[ruseinov]

Oops, yeah, this was a bit rushed. Not a blocker, however still not a fan of magic numbers. IMHO it's best to introduce a self-documenting constant than a comment, both are exactly one line of code.

[ruseinov]

Looks good to me, but would like to see the 1000.0 magic number gone.