go-agent

New Relic agent for Go runtime. It collect a lot of metrics about scheduler, garbage collector and memory allocator and
send them to NewRelic.

Requirements

Go 1.1 or higher
github.com/yvasiyarov/gorelic
github.com/yvasiyarov/newrelic_platform_go
github.com/yvasiyarov/go-metrics

You have to install manually only first two dependencies. All other dependencies will be installed automatically
by Go toolchain.

Installation

go get github.com/yvasiyarov/gorelic

and add to the initialization part of your application following code:

import (
    "github.com/yvasiyarov/gorelic"
)
....
agent := gorelic.NewAgent()
agent.Verbose = true
agent.NewrelicLicense = "YOUR NEWRELIC LICENSE KEY THERE"
agent.Run()

Middleware

If you using Beego, Martini, Revel, Kami or Gin framework you can hook up gorelic with your application by using the following middleware:

Configuration

NewrelicLicense - its the only mandatory setting of this agent.
NewrelicName - component name in NewRelic dashboard. Default value: “Go daemon”
NewrelicPollInterval - how often metrics will be sent to NewRelic. Default value: 60 seconds
Verbose - print some usefull for debugging information. Default value: false
CollectGcStat - should agent collect garbage collector statistic or not. Default value: true
CollectHTTPStat - should agent collect HTTP metrics. Default value: false
CollectHTTPStatuses - should agent collect metrics on HTTP status codes. Default value: false
CollectMemoryStat - should agent collect memory allocator statistic or not. Default value: true
GCPollInterval - how often should GC statistic collected. Default value: 10 seconds. It has performance impact. For more information, please, see metrics documentation.
MemoryAllocatorPollInterval - how often should memory allocator statistic collected. Default value: 60 seconds. It has performance impact. For more information, please, read metrics documentation.

Metrics reported by plugin

This agent use functions exposed by runtime or runtime/debug packages to collect most important information about Go runtime.

General metrics

Runtime/General/NOGoroutines - number of runned go routines, as it reported by NumGoroutine() from runtime package
Runtime/General/NOCgoCalls - number of runned cgo calls, as it reported by NumCgoCall() from runtime package

Garbage collector metrics

Runtime/GC/NumberOfGCCalls - Nuber of GC calls, as it reported by ReadGCStats() from runtime/debug
Runtime/GC/PauseTotalTime - Total pause time diring GC calls, as it reported by ReadGCStats() from runtime/debug (in nanoseconds)
Runtime/GC/GCTime/Max - max GC time
Runtime/GC/GCTime/Min - min GC time
Runtime/GC/GCTime/Mean - GC mean time
Runtime/GC/GCTime/Percentile95 - 95% percentile of GC time

All this metrics are measured in nanoseconds. Last 4 of them can be inaccurate if GC called more often then once in GCPollInterval.
If in your workload GC is called more often - you can consider decreasing value of GCPollInterval.
But be careful, ReadGCStats() blocks mheap, so its not good idea to set GCPollInterval to very low values.

Memory allocator

Component/Runtime/Memory/SysMem/Total - number of bytes/minute allocated from OS totally.
Component/Runtime/Memory/SysMem/Stack - number of bytes/minute allocated from OS for stacks.
Component/Runtime/Memory/SysMem/MSpan - number of bytes/minute allocated from OS for internal MSpan structs.
Component/Runtime/Memory/SysMem/MCache - number of bytes/minute allocated from OS for internal MCache structs.
Component/Runtime/Memory/SysMem/Heap - number of bytes/minute allocated from OS for heap.
Component/Runtime/Memory/SysMem/BuckHash - number of bytes/minute allocated from OS for internal BuckHash structs.
Component/Runtime/Memory/Operations/NoFrees - number of memory frees per minute
Component/Runtime/Memory/Operations/NoMallocs - number of memory allocations per minute
Component/Runtime/Memory/Operations/NoPointerLookups - number of pointer lookups per minute
Component/Runtime/Memory/InUse/Total - total amount of memory in use
Component/Runtime/Memory/InUse/Heap - amount of memory in use for heap
Component/Runtime/Memory/InUse/MCacheInuse - amount of memory in use for MCache internal structures
Component/Runtime/Memory/InUse/MSpanInuse - amount of memory in use for MSpan internal structures
Component/Runtime/Memory/InUse/Stack - amount of memory in use for stacks

Process metrics

Component/Runtime/System/Threads - number of OS threads used
Runtime/System/FDSize - number of file descriptors, used by process
Runtime/System/Memory/VmPeakSize - VM max size
Runtime/System/Memory/VmCurrent - VM current size
Runtime/System/Memory/RssPeak - max size of resident memory set
Runtime/System/Memory/RssCurrent - current size of resident memory set

All this metrics collected once in MemoryAllocatorPollInterval. In order to collect this statistic agent use ReadMemStats() routine.
This routine calls stoptheworld() internally and it block everything. So, please, consider this when you change MemoryAllocatorPollInterval value.

HTTP metrics

throughput (requests per second), calculated for last minute
mean throughput (requests per second)
mean response time
min response time
max response time
75%, 90%, 95% percentiles for response time

In order to collect HTTP metrics, handler functions must be wrapped using WrapHTTPHandlerFunc:

http.HandleFunc("/", agent.WrapHTTPHandlerFunc(handler))

Tracing Metrics

You can collect metrics for blocks of code or methods.

func anyMethod() {
  // Trace the whole method.
  t := agent.Tracer.BeginTrace("My traced method")
  defer t.EndTrace()
  ...Code here
  // Trace a block of code
  agent.Tracer.Trace("block trace", func() {
     .. Code here
  })
}

TODO

Collect per-size allocation statistic
Collect user defined metrics