Add timeout for I/O waitgroups

crosbymichael · crosbymichael · commit d74e8a9081a0 · 2019-09-26T14:34:48.000-04:00
Backport for #3361 This and a combination of a couple Docker changes are needed to fully resolve the issue on the Docker side. However, this ensures that after processes exit, we still leave some time for the I/O to fully flush before closing. Without this timeout, the delete methods would block forever. Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
diff --git a/linux/proc/exec.go b/linux/proc/exec.go
@@ -94,7 +94,7 @@ func (e *execProcess) setExited(status int) {
 }
 
 func (e *execProcess) delete(ctx context.Context) error {
-	e.wg.Wait()
+	waitTimeout(ctx, &e.wg, 2*time.Second)
 	if e.io != nil {
 		for _, c := range e.closers {
 			c.Close()
diff --git a/linux/proc/utils.go b/linux/proc/utils.go
@@ -19,10 +19,12 @@
 package proc
 
 import (
+	"context"
 	"encoding/json"
 	"io"
 	"os"
 	"strings"
+	"sync"
 	"time"
 
 	"github.com/containerd/containerd/errdefs"
@@ -104,3 +106,21 @@ func checkKillError(err error) error {
 func hasNoIO(r *CreateConfig) bool {
 	return r.Stdin == "" && r.Stdout == "" && r.Stderr == ""
 }
+
+// waitTimeout handles waiting on a waitgroup with a specified timeout.
+// this is commonly used for waiting on IO to finish after a process has exited
+func waitTimeout(ctx context.Context, wg *sync.WaitGroup, timeout time.Duration) error {
+	ctx, cancel := context.WithTimeout(ctx, timeout)
+	defer cancel()
+	done := make(chan struct{}, 1)
+	go func() {
+		wg.Wait()
+		close(done)
+	}()
+	select {
+	case <-done:
+		return nil
+	case <-ctx.Done():
+		return ctx.Err()
+	}
+}

Original file line number	Diff line number	Diff line change
`@@ -94,7 +94,7 @@ func (e *execProcess) setExited(status int) {`
`94`	`94`	`}`
`95`	`95`
`96`	`96`	`func (e *execProcess) delete(ctx context.Context) error {`
`97`		`- e.wg.Wait()`
	`97`	`+ waitTimeout(ctx, &e.wg, 2*time.Second)`
`98`	`98`	`if e.io != nil {`
`99`	`99`	`for _, c := range e.closers {`
`100`	`100`	`c.Close()`