dotnet core - Server hangs on Production - nginx

Question

We are currently experiencing an issue when we run our dotnet core server setup on Production. We publish it in Bamboo and run it from an AWS linux server, and it sits behind an nginx reverse proxy.
Essentially, every few days our dotnet core server process will go mute. It silently accepts and hangs on web requests, and even silently ignores our (more polite) attempts to stop it. We have verified that it is actually the netcore process that hangs by sending curl requests directly to port 5000 from within the server. We've replicated our production deployment to the best of our ability to our test environment and have not been able to reproduce this failure mode.
We've monitored the server with NewRelic and have inspected it at times when it's gone into failure mode. We've not been able to correlate this behaviour with any significant level of traffic, RAM usage, CPU usage, or open file descriptor usage. Indeed, these measurements all seem to stay at very reasonable levels.
My team and I are a bit stuck as to what might be causing our hung server, or even what we can do next to diagnose it. What might be causing our server process to hang? What further steps can we take to diagnose the issue?
Extra Information
Our nginx conf template:
upstream wfe {
server 127.0.0.1:5000;
server 127.0.0.1:5001;
}
server {
listen 80 default_server;
location / {
proxy_set_header Host $http_host;
proxy_pass http://wfe;
proxy_read_timeout 20s;
# Attempting a fix suggested by:
# https://medium.com/#mshanak/soved-dotnet-core-too-many-open-files-in-system-when-using-postgress-with-entity-framework-c6e30eeff6d1
proxy_buffering off;
proxy_http_version 1.1;
proxy_set_header Upgrade $http_upgrade;
proxy_set_header Connection keep-alive;
proxy_cache_bypass $http_upgrade;
fastcgi_buffers 16 16k;
fastcgi_buffer_size 32k;
}
}
Our Program.cs:
using System.Diagnostics.CodeAnalysis;
using System.IO;
using System.Net;
using Microsoft.AspNetCore;
using Microsoft.AspNetCore.Hosting;
using Microsoft.Extensions.Logging;
using Serilog;
namespace MyApplication.Presentation
{
[ExcludeFromCodeCoverage]
public class Program
{
public static void Main(string[] args)
{
IWebHost host = WebHost.CreateDefaultBuilder(args)
#if DEBUG
.UseKestrel(options => options.Listen(IPAddress.Any, 5000))
#endif
.UseStartup<Startup>()
.UseSerilog()
.Build();
host.Run();
}
}
}
During our CD build process, we publish our application for deployment with:
dotnet publish --self-contained -c Release -r linux-x64
We then deploy the folder bin/Release/netcoreapp2.0/linux-x64 to our server, and run publish/<our-executable-name> from within.
EDIT: dotnet --version outputs 2.1.4, both on our CI platform and on the production server.
When the outage starts, nginx logs show that server responses to requests change from 200 to 502, with a single 504 being emitted at the time of the outage.
At the same time, logs from our server process just stop. And there are warnings there, but they're all explicit warnings that we've put into our application code. None of them indicate that any exceptions have been thrown.

Answer 1

After a few days of investigation I've found the reason of that issue. It is being caused by glibc >= 2.27, which lead to GC hang at some conditions, so there is almost nothing to do about it. However you have a few options:
Use Alpine Linux. It doesn't rely on glibc.
Use older distro like Debian 9, Ubuntu 16.04 or any other with glibc < 2.27
Try to patch glibc by yourself at your own risk: https://sourceware.org/bugzilla/show_bug.cgi?id=25847
Or wait for the glibc patch to be reviewed by community and included in your favorite distro.
More information can be found here: https://github.com/dotnet/runtime/issues/47700