dramatically reduce memory usage (#758)

Run all linters per package. It allows unloading package data when it's
processed. It dramatically reduces memory (and CPU because of GC) usage.

Relates: #337
This commit is contained in:
Isaev Denis 2019-09-30 16:19:41 +03:00 committed by GitHub
parent fe494af887
commit 95ec0cf21e
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
74 changed files with 2497 additions and 2439 deletions

View file

@ -4,10 +4,7 @@ import (
"context"
"fmt"
"runtime/debug"
"sort"
"strings"
"sync"
"time"
"github.com/golangci/golangci-lint/internal/errorutil"
"github.com/golangci/golangci-lint/pkg/lint/lintersdb"
@ -93,12 +90,6 @@ func NewRunner(astCache *astcache.Cache, cfg *config.Config, log logutils.Log, g
}, nil
}
type lintRes struct {
linter *linter.Config
err error
issues []result.Issue
}
func (r *Runner) runLinterSafe(ctx context.Context, lintCtx *linter.Context,
lc *linter.Config) (ret []result.Issue, err error) {
defer func() {
@ -127,148 +118,40 @@ func (r *Runner) runLinterSafe(ctx context.Context, lintCtx *linter.Context,
return issues, nil
}
func (r Runner) runWorker(ctx context.Context, lintCtx *linter.Context,
tasksCh <-chan *linter.Config, lintResultsCh chan<- lintRes, name string) {
sw := timeutils.NewStopwatch(name, r.Log)
defer sw.Print()
for {
select {
case <-ctx.Done():
return
case lc, ok := <-tasksCh:
if !ok {
return
}
if ctx.Err() != nil {
// XXX: if check it in only int a select
// it's possible to not enter to this case until tasksCh is empty.
return
}
var issues []result.Issue
var err error
sw.TrackStage(lc.Name(), func() {
issues, err = r.runLinterSafe(ctx, lintCtx, lc)
})
lintResultsCh <- lintRes{
linter: lc,
err: err,
issues: issues,
}
}
}
}
func (r Runner) logWorkersStat(workersFinishTimes []time.Time) {
lastFinishTime := workersFinishTimes[0]
for _, t := range workersFinishTimes {
if t.After(lastFinishTime) {
lastFinishTime = t
}
}
logStrings := []string{}
for i, t := range workersFinishTimes {
if t.Equal(lastFinishTime) {
continue
}
logStrings = append(logStrings, fmt.Sprintf("#%d: %s", i+1, lastFinishTime.Sub(t)))
}
r.Log.Infof("Workers idle times: %s", strings.Join(logStrings, ", "))
}
func getSortedLintersConfigs(linters []*linter.Config) []*linter.Config {
ret := make([]*linter.Config, len(linters))
copy(ret, linters)
sort.Slice(ret, func(i, j int) bool {
return ret[i].GetSpeed() < ret[j].GetSpeed()
})
return ret
}
func (r *Runner) runWorkers(ctx context.Context, lintCtx *linter.Context, linters []*linter.Config) <-chan lintRes {
tasksCh := make(chan *linter.Config, len(linters))
lintResultsCh := make(chan lintRes, len(linters))
var wg sync.WaitGroup
workersFinishTimes := make([]time.Time, lintCtx.Cfg.Run.Concurrency)
for i := 0; i < lintCtx.Cfg.Run.Concurrency; i++ {
wg.Add(1)
go func(i int) {
defer wg.Done()
name := fmt.Sprintf("worker.%d", i+1)
r.runWorker(ctx, lintCtx, tasksCh, lintResultsCh, name)
workersFinishTimes[i] = time.Now()
}(i)
}
lcs := getSortedLintersConfigs(linters)
for _, lc := range lcs {
tasksCh <- lc
}
close(tasksCh)
go func() {
wg.Wait()
close(lintResultsCh)
r.logWorkersStat(workersFinishTimes)
}()
return lintResultsCh
}
type processorStat struct {
inCount int
outCount int
}
func (r Runner) processLintResults(inCh <-chan lintRes) <-chan lintRes {
outCh := make(chan lintRes, 64)
func (r Runner) processLintResults(inIssues []result.Issue) []result.Issue {
sw := timeutils.NewStopwatch("processing", r.Log)
go func() {
sw := timeutils.NewStopwatch("processing", r.Log)
var issuesBefore, issuesAfter int
statPerProcessor := map[string]processorStat{}
var issuesBefore, issuesAfter int
statPerProcessor := map[string]processorStat{}
defer close(outCh)
var outIssues []result.Issue
if len(inIssues) != 0 {
issuesBefore += len(inIssues)
outIssues = r.processIssues(inIssues, sw, statPerProcessor)
issuesAfter += len(outIssues)
}
for res := range inCh {
if res.err != nil {
r.Log.Warnf("Can't run linter %s: %s", res.linter.Name(), res.err)
continue
}
// finalize processors: logging, clearing, no heavy work here
if len(res.issues) != 0 {
issuesBefore += len(res.issues)
res.issues = r.processIssues(res.issues, sw, statPerProcessor)
issuesAfter += len(res.issues)
outCh <- res
}
}
for _, p := range r.Processors {
p := p
sw.TrackStage(p.Name(), func() {
p.Finish()
})
}
// finalize processors: logging, clearing, no heavy work here
if issuesBefore != issuesAfter {
r.Log.Infof("Issues before processing: %d, after processing: %d", issuesBefore, issuesAfter)
}
r.printPerProcessorStat(statPerProcessor)
sw.PrintStages()
for _, p := range r.Processors {
p := p
sw.TrackStage(p.Name(), func() {
p.Finish()
})
}
if issuesBefore != issuesAfter {
r.Log.Infof("Issues before processing: %d, after processing: %d", issuesBefore, issuesAfter)
}
r.printPerProcessorStat(statPerProcessor)
sw.PrintStages()
}()
return outCh
return outIssues
}
func (r Runner) printPerProcessorStat(stat map[string]processorStat) {
@ -283,40 +166,24 @@ func (r Runner) printPerProcessorStat(stat map[string]processorStat) {
}
}
func collectIssues(resCh <-chan lintRes) <-chan result.Issue {
retIssues := make(chan result.Issue, 1024)
go func() {
defer close(retIssues)
func (r Runner) Run(ctx context.Context, linters []*linter.Config, lintCtx *linter.Context) []result.Issue {
sw := timeutils.NewStopwatch("linters", r.Log)
defer sw.Print()
for res := range resCh {
if len(res.issues) == 0 {
continue
var issues []result.Issue
for _, lc := range linters {
lc := lc
sw.TrackStage(lc.Name(), func() {
linterIssues, err := r.runLinterSafe(ctx, lintCtx, lc)
if err != nil {
r.Log.Warnf("Can't run linter %s: %s", lc.Linter.Name(), err)
return
}
for _, i := range res.issues {
retIssues <- i
}
}
}()
return retIssues
}
func (r Runner) Run(ctx context.Context, linters []*linter.Config, lintCtx *linter.Context) <-chan result.Issue {
lintResultsCh := r.runWorkers(ctx, lintCtx, linters)
processedLintResultsCh := r.processLintResults(lintResultsCh)
if ctx.Err() != nil {
// XXX: always process issues, even if timeout occurred
finishedLintersN := 0
for range processedLintResultsCh {
finishedLintersN++
}
r.Log.Errorf("%d/%d linters finished: deadline exceeded",
finishedLintersN, len(linters))
issues = append(issues, linterIssues...)
})
}
return collectIssues(processedLintResultsCh)
return r.processLintResults(issues)
}
func (r *Runner) processIssues(issues []result.Issue, sw *timeutils.Stopwatch, statPerProcessor map[string]processorStat) []result.Issue {