nil pointer dereference

[mvrhov]

Observed behavior

We are getting the following tack trace on some of our servers when we try to unsubscribe. I'd guess that info.Config is nil, but I'm unable to confirm.

[signal SIGSEGV: segmentation violation code=0x1 addr=0xc8 pc=0xc8ccca]

goroutine 8038 [running]:
github.com/nats-io/nats%2ego.(*js).upsertConsumer(0xc0002a2000, {0xc02b6d4130, 0x9}, {0xc000883078, 0x8}, 0xc010325a40, {0x0, 0x0, 0x0})
    /root/go/pkg/mod/github.com/nats-io/nats.go@v1.31.0/jsm.go:439 +0x80a
github.com/nats-io/nats%2ego.(*js).UpdateConsumer(0xc0377869a0?, {0xc02b6d4130?, 0x0?}, 0x108086f?, {0x0?, 0x0?, 0x0?})
    /root/go/pkg/mod/github.com/nats-io/nats.go@v1.31.0/jsm.go:369 +0x85

The relevant code from our side is.

kv.updateConsumer("empty")

func (kv *KV[T]) updateConsumer(newKey string) error {
    subjects := make([]string, 1, len(kv.keys)+1)
    subjects[0] = kv.prefix + newKey

    kv.l.RLock()
    for id := range kv.keys {
        subjects = append(subjects, kv.prefix+id)
    }
    kv.l.RUnlock()

    // copy of config is needed, because we update the consumer from different go routines
    config := kv.consumerConfig
    config.FilterSubjects = subjects
    _, err := kv.js.UpdateConsumer(kvBucketNamePre+kv.bucketName, &config)
    if err != nil {
        return fmt.Errorf("update consumer: %w", err)
    }

    return nil
}

Expected behavior

no panic

Server and client version

The client is 1.31, however the code in 1.32 is the same. Server 2.10.9

Host environment

No response

Steps to reproduce

I have no idea on how to reproduce this as it's also sporadic on our end.

edit: changed the server version from 2.10.8 to 2.10.9 (it automatically updated with our release , as we set the version in docker-compose to 2.10) as this is the version we were running

[piotrpio]

Hello @mvrhov, thank you for reporting the issue.

That does look like we're getting nil info from the server - I'll investigate and let you know.

[mvrhov]

If it helps. There is a bunch of those in the NATS server log at the time of above crash. They start approximately 40s before crash. I did turn the feature flag of at the time those messages start in the NATS log, and turning feature off results in mass unsubscribe from multiple servers and multiple subjects.

[1] 2024/01/15 06:21:29.036589 [WRN] Internal subscription on "$JS.API.CONSUMER.CREATE.KV_STATUS.AV7oozAP3wLqzUtRQqMAUP" took too long: 2.105295582s
[1] 2024/01/15 06:21:29.036602 [WRN] X.Y.Z.W:37370 - cid:60 - Readloop processing time: 2.105367662s
[1] 2024/01/15 06:21:30.396007 [WRN] Internal subscription on "$JS.API.CONSUMER.CREATE.KV_STATUS.0jsgR82U" took too long: 3.156122306s
[1] 2024/01/15 06:21:30.396041 [WRN] Internal subscription on "$JS.API.CONSUMER.CREATE.KV_STATUS.0jsgR82U" took too long: 3.156166899s