GopherCon IL 2020 - Web Application Profiling 101

Web Application
Proﬁling 101
Yinon Avraham
@yinonavraham

OVERVIEW
Go's standard library comes with built-in tools to
support monitoring and proﬁling applications.
To enable it in a web application, it needs to register
some built-in HTTP endpoints.
We will now review some of the tools.

AGENDA
Exposing Applicative Information
Proﬁling CPU, Memory and More
Tracing Application Execution
Summary

EXPOSING
APPLICATIVE INFORMATION

EXPVAR PACKAGE
● Provides information on exposed application
variables
● Predeﬁned variables: command line, memory
statistics
● Supports adding custom variables
● Useful for monitoring a running application

EXPVAR EXAMPLE
GET /debug/vars
{
"cmdline": ["/path/to/myapp"],
"memstats": {
"Alloc": 7891752,
"PauseTotalNs": 316582784,
"NumGC": 670,
...
}
}

EXPVAR HOW-TO
import "expvar" // omitted other required imports
var calls expvar.Int
func main() {
expvar.Publish("hello.calls", &calls)
http.HandleFunc("/hello", ServeHello)
log.Fatal(http.ListenAndServe(":7777", nil))
}
func ServeHello(w http.ResponseWriter, req *http.Request) {
calls.Add(1)
fmt.Fprintf(w, "Hello JFrog!")
}

EXPVAR EXAMPLE
GET /debug/vars
{
"cmdline": ["/path/to/myapp"],
"hello.calls": 4,
"memstats": {
...
}
}

NET/HTTP/PPROF PACKAGE
Provides a web page with several built-in proﬁling
information, including:
● Memory allocations
● Synchronization points (blocks)
● Active goroutines
● Locks (mutex)

PROFILING HOW-TO
Register the pprof endpoints, e.g. implicitly:
import _ "net/http/pprof"
Browse to the pprof web page (by default at):
/debug/pprof
Analyze using the pprof tool, for example:
go tool pprof -http :3000
http://localhost:7777/debug/pprof/profile

CPU PROFILING
/debug/pprof/profile
● Provides information on "hot" paths
● Call stack sample is taken every 10 ms (default)
● Sums the CPU time every sampled function spends
● Has some performance impact (non-neglectable),
but only on-demand

CPU PROFILE DIAGRAMS
Flame Graph
Call Graph

MEMORY PROFILING (sampling)
/debug/pprof/heap
Memory allocations of live objects
/debug/pprof/allocs
All past memory allocations
● Collected by sampling, based on the GC information
● Helps to identify suspects for GC exhaustion
● Use the pprof tool to analyze

GOROUTINES STACK TRACES
/debug/pprof/goroutine
All current goroutines
/debug/pprof/threadcreate
Led to the creation of new OS threads
/debug/pprof/block
Led to blocking on synchronization primitives (inc. channels)
/debug/pprof/mutex
holders of contended mutexes

CUSTOM PROFILES
● Anyone can add custom deﬁned proﬁles
● Usually used to track resources and identify leaks
● There are some requirements from the managed
resource - read the runtime/pprof package's GoDoc

CUSTOM PROFILES HOW-TO
import "runtime/pprof"
var myProfile = pprof.NewProfile("my.profile")
func New() *Resource {
r := &Resource{}
myProfile.Add(r, 1)
return r
}
func (r *Resource) Close() {
myProfile.Remove(r)
}

TRACE TOOL
● Used to trace the execution of a running application
● Provides visual information on:
○ Goroutines Scheduling
○ CPU Utilization
○ Heap Memory Allocation
○ GC

TRACE HOW-TO
1. Collect trace information for e.g. 5 seconds:
GET /debug/pprof/trace?seconds=5
And save the output to a ﬁle
(e.g. using: curl <url> -o trace.out)
2. Use the trace tool to open a web browser:
go tool trace trace.out

SUMMARY
● Make the debug endpoints (vars, pprof, etc.)
available at runtime
● Use the expvar package to expose applicative
information and metrics
● DON'T use net/http package's DefaultServeMux,
create your own, explicitly add debug endpoints
● MUST restrict access to the debug endpoints
e.g. put behind auth, or use non-public IP & port

REFERENCES
● Go Tooling in Action (Francesc Campoy)
https://youtu.be/uBjoTxosSys
● Profiling and Optimizing Go (Prashant Varanasi)
https://youtu.be/N3PWzBeLX2M
● Profiling & Optimizing in Go (Brad Fitzpatrick)
https://youtu.be/xxDZuPEgbBU
● Profiling Go programs with pprof (Julia Evans)
https://jvns.ca/blog/2017/09/24/profiling-go-with-pprof/
● Profiling a go service in production (Alexander Else)
https://youtu.be/19bxBMPOlyA
● Two Go Programs, Three Different Profiling Techniques (Dave Cheney)
https://youtu.be/nok0aYiGiYA
● Custom pprof profiles (Jaana B. Dogan)
https://rakyll.org/custom-profiles/

GopherCon IL 2020 - Web Application Profiling 101

Recommended

More Related Content

What's hot (20)

Similar to GopherCon IL 2020 - Web Application Profiling 101 (20)

Recently uploaded (20)

GopherCon IL 2020 - Web Application Profiling 101