~In case anybody else stumbles upon this, my intuition was on the right track. Not actually anything to do with CPU or memory limits, but boosting them did help. The liveness and readiness checks were just guesses but not correct at all. After a minute or so, because the pod was neither live nor ready, it would be restarted which was sending the SIGINT to the Typesense daemon The fix was adding httpGet commands to /health on the

livenessProbe

and `readinessProbe`~