We've also run into exactly this problem. So far we've tested a combination of NiFi and Java versions with mixed success: seems like the issue is a Java one and is present in a few different versions/distributions of Java.
Same symptoms: InvokeHTTP slowly locks up until all threads are locked and the queues back up.
Couldn't reproduce on a simple AWS deployment, involing one EC2 instance running an HTTP server serving a small binary file, and another running a single node 2-core nifi cluster. Haven't tested with a larger instance yet.
|NiFi 1.9.2||NiFi 1.11.4||NiFi 1.12.1|
|Oracle Java 1.8u144||works||works||not tested|
|OpenJDK 1.8u242||n/t||n/t||about to test|
|Amazon Corretto 11.0.8||n/t||n/t||fails|
To update - the combination of NiFi 1.12.1 and OpenJDK 1.8u242 works fine - left it running overnight with no hanging. I expect NiFi 1.11.4 to also work with this version, and avoid the bleeding edge!
I've had some success in stabilising InvokeHTTP by adding a "Connection" property and setting its value to "close". As I understand it, this will effectively disable HTTP keepalives in the underlying libraries.