fix: panic: close of closed channel

[ldez]

I used a global cancellable context because cancel is concurrency safe (and so can be called several times).

The background of the problem is the fact of having go routines inside go routines inside go routines.

I think I will rewrite this section at some point.

---

To reproduce, I replaced Next with a non-existent method Foo.

git clone https://github.com/NVIDIA/aistore.git
cd aistore
git checkout 7730e4290

diff --git i/cmd/cli/cli/lhotse.go w/cmd/cli/cli/lhotse.go
index 1a13987aa..c16e0b1a2 100644
--- i/cmd/cli/cli/lhotse.go
+++ w/cmd/cli/cli/lhotse.go
@@ -262,7 +262,7 @@ func lhotseMultiBatch(c *cli.Context, outCtx *mossReqParseCtx) error {
                // when batch is full
                if len(currentBatch) >= outCtx.batchSize {
                        // Get next filename from template
-                       shardName, hasNext := outCtx.pt.Next()
+                       shardName, hasNext := outCtx.pt.Foo()
                        if !hasNext {
                                return fmt.Errorf("template exhausted at batch %d (generated %d batches so far)", totalBatches+1, totalBatches)
                        }

---

Fixes #5938
Related to #5929, #5923

[alex-aizman]

for stopChan, maybe consider this type pattern:

• https://github.com/NVIDIA/aistore/blob/main/cmn/cos/sync.go#L153-L169

[ldez]

The usage of context in this situation seems better because of at least the parent-child relationship.

[bombsimon]

LGTM

[bombsimon]

It should be logically the same but when ctx.Done() is closed ctx.Err() will return a non nil error. Just a personal pref on what's easier to read I guess.

Suggested change

	select {
	case <-ctx.Done():
	return
	default:
	}
	if ctx.Err() != nil {
	return
	}

[ldez]

In this situation, for me, checking the error is confusing because we don't expect an error but a stop, and it will act like we are ignoring an error.

It's a personal pref too, but in this context, I prefer using Done.

[bombsimon]

Go for it, either works. ✅