Consumer not retrying connection when dropped

[p-karanthaker]

Describe the bug
I am running a consumer with a single broker instance. When the connection to the broker is dropped - the consumer will sometimes hang and sometimes the kafkajs code to retry and restart will kick in. It appears to be random in what happens.

To Reproduce
I've used the sample consumer code from the KafkaJS website. Uploaded what I have as a Gist below.

1. consumer code
2. Block port 9092 (or whatever port your broker is on) to simulate connection drop with broker:
   2.1. sudo iptables -A OUTPUT -p tcp --sport 9092 -j DROP Prevent the server from reaching the client
   2.2. sudo iptables -A INPUT -p tcp --dport 9092 -j DROP Prevent the client from reaching the server

Expected behavior
I would expect kafkajs to always go into a retry/restart loop.

Observed behavior
I observed two things:

1. The expected behaviour happens
2. The consumer hangs and never retries/restarts

{"level":"ERROR","timestamp":"2021-08-02T09:27:19.903Z","logger":"kafkajs","message":"[Connection] Connection error: read ETIMEDOUT","broker":"192.168.1.77:9092","clientId":"example-consumer","stack":"Error: read ETIMEDOUT\n    at TCP.onStreamRead (internal/stream_base_commons.js:209:20)"}

The two behaviours seem inconsistent and random

Environment:

• OS: Ubuntu 20.04.2 LTS
• KafkaJS version 1.15.0
• Kafka version 2.7.0
• NodeJS version v12.22.3

[fatum]

We're experiencing the same thing

[Igor-Kuzmin]

We also faced this issue.
As a workaround we enabled enforceRequestTimeout and restarting the consumer manually based on CRASH event.

[backbone87]

Can confirm this behavior.

[backbone87]

Is there any way to help to move this forward towards a resolution?

[mguay22]

I can no longer reproduce this on v2.1.0. Can anyone confirm otherwise?

[mguay22]

I believe this is fixed by #1337