fix: switch to shell-based multi-process for proper multi-core scaling

Crystal's Process.fork doesn't work reliably in Docker containers.
Switch to LD_PRELOAD SO_REUSEPORT shim + shell wrapper spawning
N independent processes (same approach as other entries).

Author: BennyFranciscus

frameworks/kemal/Dockerfile

@@ -3,8 +3,12 @@ FROM crystallang/crystal:1.14-alpine AS builder
 
 WORKDIR /app
 
-# Install SQLite dev headers
-RUN apk add --no-cache sqlite-dev sqlite-static
+# Install SQLite dev headers and gcc for reuseport shim
+RUN apk add --no-cache sqlite-dev sqlite-static gcc musl-dev
+
+# Build SO_REUSEPORT shim
+COPY reuseport.c /tmp/reuseport.c
+RUN gcc -shared -fPIC -o /tmp/libreuseport.so /tmp/reuseport.c -ldl
 
 # Install dependencies
 COPY shard.yml shard.lock ./
@@ -14,14 +18,18 @@ RUN shards install --production
 COPY src/ src/
 RUN crystal build src/server.cr -o server --release --no-debug
 
-# Runtime stage — need same libc as builder (musl on Alpine)
+# Runtime stage
 FROM alpine:3.20
 
 RUN apk add --no-cache ca-certificates libgcc gc libevent sqlite-libs pcre2
 
 WORKDIR /app
 COPY --from=builder /app/server /app/server
+COPY --from=builder /tmp/libreuseport.so /usr/lib/libreuseport.so
+COPY run.sh /app/run.sh
+RUN chmod +x /app/run.sh
 
 EXPOSE 8080
 
-CMD ["/app/server"]
+ENV LD_PRELOAD=/usr/lib/libreuseport.so
+CMD ["/app/run.sh"]

frameworks/kemal/reuseport.c

@@ -0,0 +1,14 @@
+/* LD_PRELOAD shim: set SO_REUSEPORT on every SOCK_STREAM bind() */
+#define _GNU_SOURCE
+#include <dlfcn.h>
+#include <sys/socket.h>
+#include <sys/types.h>
+
+typedef int (*bind_fn)(int, const struct sockaddr *, socklen_t);
+
+int bind(int sockfd, const struct sockaddr *addr, socklen_t addrlen) {
+    int optval = 1;
+    setsockopt(sockfd, SOL_SOCKET, SO_REUSEPORT, &optval, sizeof(optval));
+    bind_fn real_bind = (bind_fn)dlsym(RTLD_NEXT, "bind");
+    return real_bind(sockfd, addr, addrlen);
+}

frameworks/kemal/run.sh

@@ -0,0 +1,26 @@
+#!/bin/sh
+# Multi-process launcher: one worker per CPU core with SO_REUSEPORT
+NPROC=$(nproc 2>/dev/null || grep -c ^processor /proc/cpuinfo 2>/dev/null || echo 1)
+
+if [ "$NPROC" -le 1 ]; then
+  exec /app/server
+fi
+
+PIDS=""
+cleanup() {
+  for pid in $PIDS; do
+    kill "$pid" 2>/dev/null
+  done
+  wait
+  exit 0
+}
+trap cleanup INT TERM
+
+i=0
+while [ "$i" -lt "$NPROC" ]; do
+  /app/server &
+  PIDS="$PIDS $!"
+  i=$((i + 1))
+done
+
+wait

frameworks/kemal/src/server.cr

@@ -187,47 +187,12 @@ post "/upload" do |env|
 end
 
 # ---------------------------------------------------------------------------
-# Server config — multi-core via SO_REUSEPORT + fork
+# Server config
 # ---------------------------------------------------------------------------
 
 Kemal.config.port = 8080
 Kemal.config.env = "production"
-
-# Disable Kemal's default signal handlers and logging for workers
 Kemal.config.shutdown_message = false
 Kemal.config.logging = false
 
-worker_count = System.cpu_count.to_i
-worker_count = 1 if worker_count < 1
-
-def start_server
-  # Build Kemal's handler chain manually
-  Kemal.config.setup
-
-  # Register 404 handler (normally done by Kemal.run)
-  unless Kemal.config.error_handlers.has_key?(404)
-    error 404 do
-      render_404
-    end
-  end
-
-  handlers = Kemal.config.handlers
-
-  # Create server with SO_REUSEPORT
-  server = HTTP::Server.new(handlers)
-  server.bind_tcp("0.0.0.0", 8080, reuse_port: true)
-  server.listen
-end
-
-if worker_count > 1
-  pids = [] of Int64
-  worker_count.times do
-    child = Process.fork { start_server }
-    pids << child.pid
-  end
-  Signal::INT.trap { pids.each { |p| Process.signal(Signal::TERM, p) rescue nil }; exit }
-  Signal::TERM.trap { pids.each { |p| Process.signal(Signal::TERM, p) rescue nil }; exit }
-  sleep
-else
-  start_server
-end
+Kemal.run