v4.0.0

• Unforks and upgrades Prometheus
• Migrates to Go modules

v3.4.2

• Allow passing in net.ListenerConfig for TUDPTransport
• Decrease M3 reporter time resolution from 1ms to 100ms

v3.4.1

• Further improve the memory efficiency of M3 histograms
• Extend shared metric storage to non-histogram metric types as well
• Add internal metrics
• Permit nested subscope closing (requires conversion to io.Closer)
• Do not drop metrics when reporting channel is full

v3.4.0

This release improves the memory efficiency of histograms. In particular, high-cardinality, histogram-heavy workloads should see noticeable reductions to heap size and allocated objects, as well as a corresponding (though application-dependent) reduction to CPU spent on GC.

• Make histogram allocations more efficient, general cleanup (#146)
• Thrift rework for allocations, more histogram improvements (#149) (#152)
• Fix Histogram bucket counter metric sizing (#154)

v3.3.17

• Don't change Histogram values during Snapshot (#142)

v3.3.15

• Optimize Scope.cacheReport (#121)

v3.3.14

• m3: Speed up calculateSize and reduce allocations (#116)
• Add HistogramBucketTagPrecision to m3 config (#118)
• Fix license headers for 2020 (#122)
• Update go versions used during CI (#123)
• m3: Reduce allocations when emitting metrics (#120)
• m3: Remove atomic from TCalcTransport (#119)
• Use raw concat instead of fmt.Sprintf for fullyQualifiedName (#124)

v3.3.13

• Fix lock usage to avoid deadlocks when allocating metrics (#112)
• Add context/test for Tally<>Prometheus usage discrepancies (#114)

v3.3.12

• Flush reporter from root scope only (#108)

v3.3.11

• Vendor apache thrift due to build incompatibility