Memory Spikes

[alterEgo123]

Hello, I'm running Overpass API in Kubernetes with an EFS volume using the official image wiktorn/overpass-api:latest.
I use these environment variables:

env:
- name: OVERPASS_META
  value: "yes"
- name: OVERPASS_MODE
  value: "clone"
- name: OVERPASS_PLANET_URL
  value: "https://planet.openstreetmap.org/planet/planet-latest.osm.bz2"
- name: OVERPASS_DIFF_URL
  value: "https://planet.openstreetmap.org/replication/minute/"
- name: OVERPASS_RULES_LOAD
  value: "10"
- name: OVERPASS_ALLOW_DUPLICATE_QUERIES
  value: "yes"

And I set a 25GB, 2.5 cpu limits, and 500GB of disk space.

I use this locustfile to test it, and even with a single user, the server reaches the memory limit quickly, restarting in ungraceful pod restart.

from locust import HttpUser, task, between
import random
import base64

USERNAME = "admin"
PASSWORD = "admin"


QUERIES = [
    # small bbox (most common)
    """
    [out:json][timeout:25];
    node["amenity"="restaurant"](36.85,10.15,36.86,10.16);
    out;
    """,

    # medium bbox
    """
    [out:json][timeout:25];
    node["amenity"="restaurant"](36.84,10.14,36.87,10.18);
    out;
    """,

    # heavier query (less frequent)
    """
    [out:json][timeout:25];
    node["shop"](36.8,10.1,36.9,10.2);
    out;
    """
]

class OverpassUser(HttpUser):
    wait_time = between(2, 10)  # realistic think time

    def on_start(self):
        credentials = f"{USERNAME}:{PASSWORD}"
        token = base64.b64encode(credentials.encode()).decode()
        self.headers = {
            "Authorization": f"Basic {token}",
            "Content-Type": "application/x-www-form-urlencoded"
        }

    @task
    def query_overpass(self):
        query = random.choice(QUERIES)
        self.client.post(
            "/api/interpreter",
            data={"data": query},
            headers=self.headers
        )

This is what the resource consumption looks like:

The pod restart causes this issue runtime error: open64: 2 No such file or directory /db/db//osm3s_osm_base Unix_Socket::7 but I solve it with chmod og+rx /db && exec /docker-entrypoint.sh /app/docker-entrypoint.sh in the entrypoint.

Is there a way to prevent the Overpass API server from reaching the limit and crashing?