envoy sidecar TLS proxy and cubestore-router to cubestore-worker traffic

I am deploying cubejs/cubestore into a kubernetes cluster. My deployment is similar [to this example](https://github.com/cube-js/cube/blob/master/examples/kubernetes/cluster/cubestore-workers-service.yaml) (I do not use a stateful set for the cubestore-worker or a headless service - just a deployment and a regular service).

So far I have used envoy sidecar proxies to
1. encrypt other non-cube pod traffic to the cube-api.
2. encrypt traffic from cubestore-worker to exterior pods like ksqldb.
3. encrypt traffic from cube-api/cube-refresh-worker to cubestore-router.

However, using the same techniques does not work from traffic from cubestore-router pod to cubestore-worker pod. Instead the router logs show the following error multiple times:

`ERROR [cubestore::scheduler] Error processing event Insert(Tables, 6): Corrupted message received. Please check your worker and meta connection environment variables.`

The cubestore-worker sidecar logs show the following error:

`[debug][http] [source/common/http/conn_manager_impl.cc:403] [Tags: "ConnectionId":"6"] dispatch error: http/1.1 protocol error: HPE_INVALID_METHOD`

All this suggests that there is some unique customized behaviour in the way the cubestore-router communicates with cubestore-worker. Can anyone help me better understand this behaviour?