The Coralogix team faced challenges with Kafka consumer performance due to issues stemming from the Java Virtual Machine (JVM) garbage collection, particularly with the G1GC algorithm. As they began batching multiple logs into single records, the size of the records increased, leading to the creation of humongous objects that caused memory fragmentation and frequent garbage collection pauses. By conducting experiments with different batch sizes, they discovered that smaller batches reduced garbage collection pauses significantly. The team upgraded to JDK 11, which offered improved G1GC performance, but ultimately found success by switching to the ZGC garbage collector, which handled large objects more efficiently. This solution greatly improved the stability and performance of their data streaming operations, demonstrating the importance of understanding and optimizing garbage collection processes in JVM environments.