[2.34.0 -> 2.36.2] request to graphql-engine failed, reason: socket hang up

### Version Information

* hasura / graphql-engine: 2.36.2
* input-output-hk / cardano-graphql: 8.0.0
* cardano-node: 8.7.3

### Environment

On premise K8S cluster, using custom docker image based on ubuntu:22.04

### What is the current behaviour?

The **graphql-server** makes request to **graphql-engine** and get a lot of **ECONRESET**

```
2024-01-23T08:27:37.767785380Z {
"name":"cardano-graphql",
"hostname":"cardano-ogmios-preview-test-873-graphql-server-df88db448-cws9l",
"pid":1,
"level":50,
"err":{
"message":"request to http://cardano-ogmios-preview-test-873-graphql-engine-headless:8888/v1/graphql failed, reason: socket hang up",
"type":"system",
"errno":"ECONNRESET",
"code":"ECONNRESET"
},
"msg":"request to http://cardano-ogmios-preview-test-873-graphql-engine-headless:8888/v1/graphql failed, reason: socket hang up",
"time":"2024-01-23T08:27:37.767Z",
"v":0
}
```

The **graphql-engine** is started with basic parameters

```
graphql-engine serve \
--server-port 8888 \
--database-url "${HASURA_GRAPHQL_DATABASE_URL}"
```

### Any possible solutions/workarounds you're aware of?

Rolling back to version 2.34.0 fixed the issue.

### Keywords

* ECONNRESET

Start with the graphql-engine serve invocation on port 8888 and reproduce the graphql-server request against version 2.36.2, comparing it with 2.34.0 in the stated Kubernetes setup. Use graphql-engine and graphql-server logs to isolate the ECONNRESET, then document the cause and verify the fix with a regression test or reproducible request.

Written by the indexing model from the issue text.