Producer connection auto-closes after unspecified timeout

Describe the bug I came across this using the node client, but it appears to be a server issue. When I create a new producer, I can send messages successfully, which rules out TLS handshake problems. Connecting to pulsar appears to be okay.

If, however, the producer connection remains idle for about 200 milliseconds, I get the following console output (hosts and remote IPs omitted):

2020-03-05 10:17:12.571 INFO  ConnectionPool:85 | Created connection for pulsar+ssl://<host omitted>
2020-03-05 10:17:12.700 INFO  ClientConnection:330 | [172.19.0.2:36756 -> <IP omitted>:6651] Connected to broker
2020-03-05 10:17:12.825 INFO  BatchMessageContainer:43 | { BatchContainer [size = 0] [batchSizeInBytes_ = 0] [maxAllowedMessageBatchSizeInBytes_ = 131072] [maxAllowedNumMessagesInBatch_ = 1000] [topicName = persistent://<omitted>/test-topic] [producerName_ = ] [batchSizeInBytes_ = 0] [numberOfBatchesSent = 0] [averageBatchSize = 0]} BatchMessageContainer constructed
2020-03-05 10:17:12.825 INFO  HandlerBase:53 | [persistent://<omitted>/test-topic, ] Getting connection from pool
2020-03-05 10:17:12.852 INFO  ConnectionPool:85 | Created connection for pulsar+ssl://192.168.125.165:6651
2020-03-05 10:17:12.878 INFO  ClientConnection:332 | [172.19.0.2:36758 -> <IP omitted>:6651] Connected to broker through proxy. Logical broker: pulsar+ssl://192.168.125.165:6651
2020-03-05 10:17:12.993 INFO  ProducerImpl:151 | [persistent://<omitted>/test-topic, ] Created producer on broker [172.19.0.2:36758 -> <IP omitted>:6651] 
2020-03-05 10:17:13.093 INFO  ClientConnection:1349 | [172.19.0.2:36758 -> <IP omitted>:6651] Connection closed
2020-03-05 10:17:13.093 INFO  ClientConnection:1349 | [172.19.0.2:36756 -> <IP omitted>4:6651] Connection closed
2020-03-05 10:17:13.093 INFO  ClientConnection:235 | [172.19.0.2:36758 -> <IP omitted>:6651] Destroyed connection

To Reproduce

• Create a producer from a client library
• Sleep and observe the console output

Expected behavior Connection should remain open, or if the timeout is intentional, it should be exposed as a config option

Desktop (please complete the following information):

• Debian GNU/Linux 9 (stretch)

Additional context Issue cannot be reproduced through the CLI client, because creating a producer connection is triggered by explicitly sending a message, which means the connection will never remain idle.

[alexyans]

I don’t have access to the release version at the moment, but the bug was patched in the C++ client that other language-specific clients wrap around. That was about 9 months ago.

Perhaps the python lib needs to bump the dependency?

[alexyans]

Oh, good! Glad you worked it out then 😃