CPU throttling observed after using get_enrichment_table_record function in VRL pipeline

[andrepbento]

We are experiencing CPU throttling issues after updating our VRL implementation to include the get_enrichment_table_record function call in our Vector pipeline. The throttling started immediately after this change.

Details:

• Vector version: v0.49.0
• Environment: GKE with n2-highcpu-32 machine types with one to three pods per node
• Traffic: 10 TiB ingestion per day, steady flow of 1.8 MiB/s
• Pipeline uses two lists for enrichment: one list with a few MB and another list with ~30 MB

We have checked other metrics such as Network RX/TX and Disk IO, but none correlate with the CPU throttling.
The issue only started after switching to the VRL implementation that calls get_enrichment_table_record.

Expected behavior:
No CPU throttling or significant CPU spikes when using the get_enrichment_table_record function.

Additional context:
Increasing the minimum replicas from 3 to 6 helped diluting the ingestion traffic and reduced throttling. Also, increasing the CPU limit per pod to 10 helped, but CPU usage remains very low (<10%) constantly.
Any guidance on how to optimize or debug this further would be appreciated.