Closed ventaubain closed 1 year ago
There are some "destroy" connections from the backend. It seems to be the problem and not "a bug". How works the sidecar about this "destroy" comportment ?
The problem seems to be an intern problem of the service. Not sidecar or Consul. I close the issue. Sorry for the mistake.
Overview of the Issue
I have 2 services in a Mesh network. Services are in a Nomad cluster. The service (1) is connected to the service (2) by an upstream on the localhost:7000. (2) is a socks5 proxy and has 3 replicas (default load balancing policy and this one has a correct comportment) and the request from (1) is socks5-based. At the beginning, both work correctly together but, after few times (not constant and no reason detected yet), the upstream connection seems to fail.
After few times, if I do a Curl (
curl --socks5-hostname localhost:7000 <URL>
), I obtaincurl: (97) Unable to receive initial SOCKS5 response.
. If I restart the sidecar of service (1), the connection is retablished and the upstream correctly works. If there is only one container and not 3 under the upstream, the connection seems not stable too. The problem could be correlated to the load balancing.The service (1) has an another upstream to a Mysql container and has no (detected) problem.
When the active connections (envoy_cluster_upstream_cx_active) is higher than 1k02, the proxy stops and never re-works except if I relaunch it. I use default sidecar configuration so I have:
It can be the reason but why crash with 1k02 and not 512 ?
Reproduction Steps
Steps to reproduce this issue, eg:
Consul info for both Client and Server
Nomad job configuration
Operating system and Environment details
Log Fragments
I don't find explicit error in log.