Frontend query split timeout questions

[mengmengy]

Hello, I use cortex frontend only in my project, I only enabled query-fronend module with step alignment and splitting in yaml config and I also configured my own downstream prometeus url, below is my config file.

target: query-frontend

http_prefix: ""
limits:
  max_query_parallelism: 15

api:
  response_compression_enabled: true

server:
  http_listen_port: 9090
  http_server_read_timeout: 60s
  http_server_write_timeout: 60s

query_range:
  split_queries_by_interval: 24h
  align_queries_with_step: true
  max_retries: 2
  cache_results: false

frontend:
  query_stats_enabled: true 
  downstream_url: http://myDownstreamPrometheusUrl:8080 # downstream prometheus url

With my load test with k6 to my cortex frontend, I'm seeing most of my requests end-to-end duration is 60s, even though my downstream prometheus returned response in ~14s in P99. After adding tracing and debugging, I find out it's because this io.Copy(w, resp.body) taking time. I printed out error returned by this line, it returned this error: write tcp 10.200.1.15:9090->10.200.0.126:36774: i/o timeout. These requests didn't supply Accept-Encoding: gzip in request header.
After I supply Accept-Encoding: gzip to the request header, the i/o write timeout error gone and copied bytes count is 54,347,428. I can see the end-to-end latency reduced to ~20s in P99. I guess the i/o timeout I saw is because the response size too big.

Questions:
In order to get benefits from query split, I need to compress response?
Do you think there are other reasons causing this i/o issue I missed, like http connection setup etc...
Error returned by this io.Copy(w, resp.body) is swallowed and still return success response, why not return error response in case error returned?
May I get some thoughts here?

Environment
I'm running cortex frontend as a docker image in k8s. Nginx configured in k8s receive client requests and forward to cortex frontend. Nginx has timeout set to 60s.