consumer: concurrent handlers not working as expected

[pajadah]

Hi, I have a problem on my project that are using NSQ conccurent handler.
So I am using nsq.AddConcurrentHandlers(&MyHandler{}, 16). My project runs on 2 servers, so it means it will spawn 16 concurrent handlers from each server, with total of 32 handlers.
When one the topic depth is getting high (around 3 million), I check in nsqadmin that only 4 handlers out of 32 handlers are processing the messages. I wonder if there's any possible reason for this? since if I can get all that 32 handlers to process the messages, then there's won't be any high depth at all.

Here's my client code:

config := nsq.NewConfig()
config.DefaultRequeueDelay = 0
config.MaxAttempts = 0
config.MaxInFlight = 5000

q, err := nsq.NewConsumer("this_is_topic", "this_is_channel", config)
if err != nil {
    log.Println(err.Error())
    return
}

h := &MyHandler{
    q: q
}

q.AddConcurrentHandlers(h, 16)
q.SetLogger(log.New(os.Stderr, "nsq:", log.Ltime), nsq.LogLevelError)

err = p.connectToLookupds(q)
if err != nil {
    log.Println(err.Error())
}

[ploxiln]

Can you show more of your handler function? How long might it take to process a message?

MaxInFlight=5000 looks too high, you want it to be between 16 and maybe 100 depending on how many nsqd you have and how long the handler typically takes to process a message.

Setting MaxAttempts and DefaultRequeueDelay to 0 are not good ideas - I think you want MaxAttempts to be 1, and Default RequeueDelay should be positive, but you can leave it at the default and requeue with backoff=false so the consumer does not backoff when the message fails.

[ploxiln]

(can you also share stats from nsqd? and how many nsqd have this topic?)

[ploxiln]

What do you set MaxInFlight to? This is the other key variable to adjust for concurrent handling.