#define _GNU_SOURCE

#include <errno.h>
#include <signal.h>
#include <stdio.h>
#include <string.h>
#include <unistd.h>
extern "C" {
	#include <sched.h>
	#include <signal.h>
	#include <sys/eventfd.h>
	#include <sys/socket.h>
	#include <netinet/in.h>
}

#include <fstream.hfa>
#include <kernel.hfa>
#include <iofwd.hfa>
#include <stats.hfa>
#include <time.hfa>
#include <thread.hfa>

#include "filecache.hfa"
#include "options.hfa"
#include "worker.hfa"

extern void register_fixed_files( cluster &, int *, unsigned count );

Duration default_preemption() {
	return 0;
}

//=============================================================================================
// Stats Printer
//============================================================================================='

thread StatsPrinter {
	Worker * workers;
	int worker_cnt;
};

void ?{}( StatsPrinter & this, cluster & cl ) {
	((thread&)this){ "Stats Printer Thread", cl };
	this.worker_cnt = 0;
}

void ^?{}( StatsPrinter & mutex this ) {}

#define eng3(X) (ws(3, 3, unit(eng( X ))))

void main(StatsPrinter & this) {
	LOOP: for() {
		waitfor( ^?{} : this) {
			break LOOP;
		}
		or else {}

		sleep(10`s);

		print_stats_now( *active_cluster(), CFA_STATS_READY_Q | CFA_STATS_IO );
		if(this.worker_cnt != 0) {
			uint64_t tries = 0;
			uint64_t calls = 0;
			uint64_t header = 0;
			uint64_t splcin = 0;
			uint64_t splcot = 0;
			struct {
				volatile uint64_t calls;
				volatile uint64_t bytes;
			} avgrd[zipf_cnts];
			memset(avgrd, 0, sizeof(avgrd));

			for(i; this.worker_cnt) {
				tries += this.workers[i].stats.sendfile.tries;
				calls += this.workers[i].stats.sendfile.calls;
				header += this.workers[i].stats.sendfile.header;
				splcin += this.workers[i].stats.sendfile.splcin;
				splcot += this.workers[i].stats.sendfile.splcot;
				for(j; zipf_cnts) {
					avgrd[j].calls += this.workers[i].stats.sendfile.avgrd[j].calls;
					avgrd[j].bytes += this.workers[i].stats.sendfile.avgrd[j].bytes;
				}
			}

			double ratio = ((double)tries) / calls;

			sout | "----- Worker Stats -----";
			sout | "sendfile  : " | calls | "calls," | tries | "tries (" | ratio | " try/call)";
			sout | "            " | header | "header," | splcin | "splice in," | splcot | "splice out";
			sout | " - zipf sizes:";
			for(i; zipf_cnts) {
				double written = avgrd[i].calls > 0 ? ((double)avgrd[i].bytes) / avgrd[i].calls : 0;
				sout | "        " | zipf_sizes[i] | "bytes," | avgrd[i].calls | "shorts," | written | "written";
			}
		}
		else {
			sout | "No Workers!";
		}
	}
}

//=============================================================================================
// Globals
//=============================================================================================
struct ServerCluster {
	cluster self;
	processor    * procs;
	// io_context   * ctxs;
	StatsPrinter * prnt;

};

void ?{}( ServerCluster & this ) {
	(this.self){ "Server Cluster", options.clopts.params };

	cpu_set_t fullset;
	CPU_ZERO(&fullset);
	int ret = sched_getaffinity(getpid(), sizeof(fullset), &fullset);
	if( ret != 0 ) abort | "sched_getaffinity failed with" | errno | strerror( errno );
	int cnt = CPU_COUNT(&fullset);

	this.procs = alloc(options.clopts.nprocs);
	for(i; options.clopts.nprocs) {
		(this.procs[i]){ "Benchmark Processor", this.self };

		int c = 0;
		int n = 1 + (i % cnt);
		for(int j = 0; j < CPU_SETSIZE; j++) {
			if(CPU_ISSET(j, &fullset)) n--;
			if(n == 0) {
				c = j;
				break;
			}
		}
		cpu_set_t localset;
		CPU_ZERO(&localset);
		CPU_SET(c, &localset);
		ret = pthread_setaffinity_np(this.procs[i].kernel_thread, sizeof(localset), &localset);
		if( ret != 0 ) abort | "sched_getaffinity failed with" | ret | strerror( ret );

		#if !defined(__CFA_NO_STATISTICS__)
			if( options.clopts.procstats ) {
				print_stats_at_exit( *this.procs, this.self.print_stats );
			}
			if( options.clopts.viewhalts ) {
				print_halts( *this.procs );
			}
		#endif
	}

	if(options.stats) {
		this.prnt = alloc();
		(*this.prnt){ this.self };
	} else {
		this.prnt = 0p;
	}

	#if !defined(__CFA_NO_STATISTICS__)
		print_stats_at_exit( this.self, CFA_STATS_READY_Q | CFA_STATS_IO );
	#endif

	options.clopts.instance[options.clopts.cltr_cnt] = &this.self;
	options.clopts.cltr_cnt++;
}

void ^?{}( ServerCluster & this ) {
	delete(this.prnt);

	for(i; options.clopts.nprocs) {
		^(this.procs[i]){};
	}
	free(this.procs);

	^(this.self){};
}

extern void init_protocol(void);
extern void deinit_protocol(void);

//=============================================================================================
// Termination
//=============================================================================================

int closefd;
void cleanstop(int) {
	eventfd_t buffer = 1;
	char * buffer_s = (char*)&buffer;
	int ret = write(closefd, buffer_s, sizeof(buffer));
	if(ret < 0) abort( "eventfd write error: (%d) %s\n", (int)errno, strerror(errno) );
	return;
}

//=============================================================================================
// Main
//============================================================================================='
int main( int argc, char * argv[] ) {
	__sighandler_t s = 1p;
	signal(SIGPIPE, s);

	//===================
	// Parse args
	parse_options(argc, argv);

	//===================
	// Setup non-interactive termination
	if(!options.interactive) {
		closefd = eventfd(0, 0);
		if(closefd < 0) abort( "eventfd error: (%d) %s\n", (int)errno, strerror(errno) );

		sighandler_t prev = signal(SIGTERM, cleanstop);
		intptr_t prev_workaround = (intptr_t) prev;
		// can't use SIG_ERR it crashes the compiler
		if(prev_workaround == -1) abort( "signal setup error: (%d) %s\n", (int)errno, strerror(errno) );

		sout | "Signal termination ready";
	}

	//===================
	// Open Files
	if( options.file_cache.path ) {
		sout | "Filling cache from" | options.file_cache.path;
		fill_cache( options.file_cache.path );
	}

	//===================
	// Open Socket
	sout | getpid() | ": Listening on port" | options.socket.port;
	int server_fd = socket(AF_INET, SOCK_STREAM, 0);
	if(server_fd < 0) {
		abort( "socket error: (%d) %s\n", (int)errno, strerror(errno) );
	}

	int ret = 0;
	struct sockaddr_in address;
	int addrlen = sizeof(address);
	memset( (char *)&address, '\0' );
	address.sin_family = AF_INET;
	address.sin_addr.s_addr = htonl(INADDR_ANY);
	address.sin_port = htons( options.socket.port );

	int waited = 0;
	for() {
		int sockfd = server_fd;
		__CONST_SOCKADDR_ARG addr;
		addr.__sockaddr__ = (struct sockaddr *)&address;
		socklen_t addrlen = sizeof(address);
		ret = bind( sockfd, addr, addrlen );
		if(ret < 0) {
			if(errno == EADDRINUSE) {
				if(waited == 0) {
					if(!options.interactive) abort | "Port already in use in non-interactive mode. Aborting";
					sout | "Waiting for port";
				} else {
					sout | "\r" | waited | nonl;
					flush( sout );
				}
				waited ++;
				sleep( 1`s );
				continue;
			}
			abort( "bind error: (%d) %s\n", (int)errno, strerror(errno) );
		}
		break;
	}

	ret = listen( server_fd, options.socket.backlog );
	if(ret < 0) {
		abort( "listen error: (%d) %s\n", (int)errno, strerror(errno) );
	}

	//===================
	// Run Server Cluster
	{
		int pipe_cnt = options.clopts.nworkers * 2;
		int pipe_off;
		int * fds;
		[fds, pipe_off] = filefds( pipe_cnt );
		for(i; 0 ~ pipe_cnt ~ 2) {
			int ret = pipe(&fds[pipe_off + i]);
			if( ret < 0 ) { abort( "pipe error: (%d) %s\n", (int)errno, strerror(errno) ); }
		}

		// if(options.file_cache.path && options.file_cache.fixed_fds) {
		// 	register_fixed_files(cl, fds, pipe_off);
		// }

		{
			// Stats printer makes a copy so this needs to persist longer than normal
			Worker * workers;
			ServerCluster cl[options.clopts.nclusters];

			init_protocol();
			{
				workers = anew(options.clopts.nworkers);
				cl[0].prnt->workers = workers;
				cl[0].prnt->worker_cnt = options.clopts.nworkers;
				for(i; options.clopts.nworkers) {
					// if( options.file_cache.fixed_fds ) {
					// 	workers[i].pipe[0] = pipe_off + (i * 2) + 0;
					// 	workers[i].pipe[1] = pipe_off + (i * 2) + 1;
					// }
					// else
					{
						workers[i].pipe[0] = fds[pipe_off + (i * 2) + 0];
						workers[i].pipe[1] = fds[pipe_off + (i * 2) + 1];
						workers[i].sockfd  = server_fd;
						workers[i].addr    = (struct sockaddr *)&address;
						workers[i].addrlen = (socklen_t*)&addrlen;
						workers[i].flags   = 0;
					}
					unpark( workers[i] );
				}
				sout | options.clopts.nworkers | "workers started on" | options.clopts.nprocs | "processors /" | options.clopts.nclusters | "clusters";
				for(i; options.clopts.nclusters) {
					sout | options.clopts.thrd_cnt[i] | nonl;
				}
				sout | nl;
				{
					if(options.interactive) {
						char buffer[128];
						for() {
							int ret = cfa_read(0, buffer, 128, 0);
							if(ret == 0) break;
							if(ret < 0) abort( "main read error: (%d) %s\n", (int)errno, strerror(errno) );
							sout | "User wrote '" | "" | nonl;
							write(sout, buffer, ret - 1);
							sout | "'";
						}
					}
					else {
						char buffer[sizeof(eventfd_t)];
						int ret = cfa_read(closefd, buffer, sizeof(eventfd_t), 0);
						if(ret < 0) abort( "main read error: (%d) %s\n", (int)errno, strerror(errno) );
					}

					sout | "Shutdown received";
				}

				sout | "Notifying connections..." | nonl; flush( sout );
				for(i; options.clopts.nworkers) {
					workers[i].done = true;
				}
				sout | "done";

				sout | "Shutting down socket..." | nonl; flush( sout );
				int ret = shutdown( server_fd, SHUT_RD );
				if( ret < 0 ) {
					abort( "shutdown error: (%d) %s\n", (int)errno, strerror(errno) );
				}
				sout | "done";

				//===================
				// Close Socket
				sout | "Closing Socket..." | nonl; flush( sout );
				ret = close( server_fd );
				if(ret < 0) {
					abort( "close socket error: (%d) %s\n", (int)errno, strerror(errno) );
				}
				sout | "done";

				sout | "Stopping connection threads..." | nonl; flush( sout );
				for(i; options.clopts.nworkers) {
					join(workers[i]);
				}
			}
			sout | "done";

			sout | "Stopping protocol threads..." | nonl; flush( sout );
			deinit_protocol();
			sout | "done";

			sout | "Stopping printer threads..." | nonl; flush( sout );
			for(i; options.clopts.nclusters) {
				StatsPrinter * p = cl[i].prnt;
				if(p) join(*p);
			}
			sout | "done";

			// Now that the stats printer is stopped, we can reclaim this
			adelete(workers);

			sout | "Stopping processors/clusters..." | nonl; flush( sout );
		}
		sout | "done";

		sout | "Closing splice fds..." | nonl; flush( sout );
		for(i; pipe_cnt) {
			ret = close( fds[pipe_off + i] );
			if(ret < 0) {
				abort( "close pipe error: (%d) %s\n", (int)errno, strerror(errno) );
			}
		}
		free(fds);
		sout | "done";

		sout | "Stopping processors..." | nonl; flush( sout );
	}
	sout | "done";

	//===================
	// Close Files
	if( options.file_cache.path ) {
		sout | "Closing open files..." | nonl; flush( sout );
		close_cache();
		sout | "done";
	}
}

const size_t zipf_sizes[] = { 102, 204, 307, 409, 512, 614, 716, 819, 921, 1024, 2048, 3072, 4096, 5120, 6144, 7168, 8192, 9216, 10240, 20480, 30720, 40960, 51200, 61440, 71680, 81920, 92160, 102400, 204800, 307200, 409600, 512000, 614400, 716800, 819200, 921600 };
static_assert(zipf_cnts == sizeof(zipf_sizes) / sizeof(zipf_sizes[0]));