Sometimes all the threads of the server fall and the server stops responding to requests

• GNAT version: 12.2.1 (alire)
• AWS version: 23 (alire)
• OS: Debian GNU/Linux 11 (bullseye) x86_64
• Host: KVM/QEMU (Standard PC (i440FX + PIIX, 1996) pc-i440fx-6.2)
• CPU: AMD EPYC 7282 (4) @ 2.794GHz

Nginx (HTTP2) is used as a reverse proxy server.

Process logs

Fatal error, slot 5 is dead now.
raised AWS.NET.SOCKET_ERROR : Send : [32] Broken pipe
0x6bb1a5 Aws.Net.Std.Raise_Exception at aws-net-std__gnat.adb:632
0x6bb615 Aws.Net.Std.Send at aws-net-std__gnat.adb:695
0x752463 Aws.Net.Send at aws-net.adb:480
0x66929e Aws.Net.Buffered.Flush at aws-net-buffered.adb:76
0x5dad81 Aws.Http2.Frame.Send at aws-http2-frame.adb:209
0x7da122 Aws.Server.Protocol_Handler at aws-server-protocol_handler.adb:295
0x7d5500 Aws.Server.LineT at aws-server.adb:395
0x9d1b39 System.Tasking.Stages.Task_Wrapper at s-tassta.adb:1211
[/lib/x86_64-linux-gnu/libpthread.so.0]
0x7fea3b8d7ea5
[/lib/x86_64-linux-gnu/libc.so.6]
0x7fea3b6b3a2d
0xfffffffffffffffe

Start with the stack trace entries in aws-http2-frame.adb, aws-server-protocol_handler.adb, and aws-server.adb, then review how the broken-pipe failure is handled when Nginx uses HTTP/2. Reproduce the server becoming unresponsive and identify the conditions that cause all server threads to stop; done means the failure is addressed and request handling remains available afterward.