Google Cloud App Engine (GAE) Flexible: Load Balancer error: backend_connection_closed_before_data_sent_to_client

Question

I have GAE flexible this app.yaml:

# [START runtime]
runtime: python
env: flex
entrypoint: gunicorn -b :$PORT ******.wsgi

automatic_scaling:
  min_num_instances: 2
  max_num_instances: 2

beta_settings:
    cloud_sql_instances: ******:europe-west6:*******

runtime_config:
  python_version: 3

handlers:
- url: /.*
  script: auto
  secure: always
  redirect_http_response_code: 301
# [END runtime]

After standard instance weekly restarting, I've seen some internal errors in logs and 502 errors on POST requests:

{
  "insertId": "2",
  "jsonPayload": {...},
  "timestamp": "2022-05-14T08:34:42.512018934Z",
  "severity": "ERROR",
  "logName": "projects/****/logs/compute.googleapis.com%2Fshielded_vm_integrity",
  "receiveTimestamp": "2022-05-14T08:34:44.518408254Z"
}

and

{
  "insertId": "*******",
  "jsonPayload": {
    "remoteIp": "******",
    "@type": "type.googleapis.com/google.cloud.loadbalancing.type.LoadBalancerLogEntry",
    "statusDetails": "backend_connection_closed_before_data_sent_to_client"
  },
  "httpRequest": {
    "requestMethod": "POST",
    "requestUrl": "https://*********",
    "requestSize": "2208",
    "status": 502,
    "responseSize": "451",
    "userAgent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:100.0) Gecko/20100101 Firefox/100.0",
    "remoteIp": "*******",
    "referer": "********",
    "serverIp": "********"
  },
  "resource": {
    "type": "http_load_balancer",
    "labels": {
      "url_map_name": "",
      "target_proxy_name": "",
      "zone": "global",
      "project_id": "********",
      "backend_service_name": "********",
      "forwarding_rule_name": ""
    }
  },
  "timestamp": "2022-05-14T08:36:28.064500Z",
  "severity": "WARNING",
  "logName": "projects/********/logs/requests",
  "trace": "projects/********/traces/********",
  "receiveTimestamp": "2022-05-14T08:36:28.882230792Z",
  "spanId": "********"
}

Also I have some messages with

/readiness_check failReason:"null"

I've read Tuning NGINX behind Google Cloud Platform HTTP(S) Load Balancer But as I understand it is related to GCE(not GAE).

For the current app it is VERY important to handle every request, so I want to be confident with fixing it.