the ability to check/kill/reload any of the processes being monitored added.

3 files changed, 57 insertions, 36 deletions

diff --git a/src/dudki.cc b/src/dudki.cc
index b4e95a7..e91ad5e 100644
--- a/src/dudki.cc
+++ b/src/dudki.cc
@@ -41,185 +41,202 @@ void check_herd(configuration& config) {
 void signal_self(const configuration& config,int signum) {
     ifstream pids(config.pidfile.c_str(),ios::in);
     if(!pids)
 	throw runtime_error("Can't detect running instance");
     pid_t pid = 0;
     pids >> pid;
     if(!pid)
 	throw runtime_error("Can't detect running instance");
     if(pid==getpid())
 	throw 0;
     if(kill(pid,signum))
 	throw runtime_error("Failed to signal running instance");
 }
 
 int main(int argc,char **argv) {
     try {
 	_argv = new char*[argc+1];
 	if(!_argv)
 	    throw runtime_error("memory allocation problem at the very start");
 	memmove(_argv,argv,sizeof(*_argv)*(argc+1));
 	string config_file = DEFAULT_CONF_FILE;
 	enum {
 	    op_default,
 	    op_work,
-	    op_hup,
-	    op_term,
-	    op_check,
+	    op_signal,
 	    op_ensure,
 	    op_test
 	} op = op_default;
+	int op_signum = 0;
 	while(true) {
 #define	SHORTOPTSTRING "f:hVLrkcet"
 #ifdef HAVE_GETOPT_LONG
 	    static struct option opts[] = {
 		{ "help", no_argument, 0, 'h' },
 		{ "usage", no_argument, 0, 'h' },
 		{ "version", no_argument, 0, 'V' },
 		{ "license", no_argument, 0, 'L' },
 		{ "config", required_argument, 0, 'f' },
 		{ "kill", no_argument, 0, 'k' },
 		{ "reload", no_argument, 0, 'r' },
 		{ "check", no_argument, 0, 'c' },
 		{ "ensure", no_argument, 0, 'e' },
 		{ "test", no_argument, 0, 't' },
 		{ NULL, 0, 0, 0 }
 	    };
 	    int c = getopt_long(argc,argv,SHORTOPTSTRING,opts,NULL);
 #else /* !HAVE_GETOPT_LONG */
 	    int c = getopt(argc,argv,SHORTOPTSTRING);
 #endif /* /HAVE_GETOPT_LONG */
 	    if(c==-1)
 		break;
 	    switch(c) {
 		case 'h':
 		    cerr << PHEADER << endl
-			<< PCOPY << endl << endl <<
+			<< PCOPY << endl << endl
+			<< " " << argv[0] << " [options] [processes]" << endl << endl <<
 #ifdef HAVE_GETOPT_LONG
 			" -h, --help\n"
 			" --usage         display this text\n"
 			" -V, --version   display version number\n"
 			" -L, --license   show license\n"
 			" -f filename, --config=filename\n"
 			"                 specify the configuration file to use\n"
 			"\n"
-			" -k, --kill      stop running instance\n"
+			" -k, --kill      stop running instance (send SIGTERM)\n"
 			" -r, --reload    reload running instance (send SIGHUP)\n"
-			" -c, --check     check if dudki is running\n"
+			" -c, --check     check if the process is running\n"
+			"     (the above commands operate on dudki itself if no\n"
+			"      process name has been specified)\n"
 			" -e, --ensure    ensure that dudki is running\n"
 			" -t, --test      test configuration file and exit"
 #else /* !HAVE_GETOPT_LONG */
 			" -h              display this text\n"
 			" -V              display version number\n"
 			" -L              show license\n"
 			" -f filename     specify the configuration file to use\n"
 			"\n"
-			" -k              stop running instance\n"
+			" -k              stop running instance (send SIGTERM)\n"
 			" -r              reload running instance (send SIGHUP)\n"
-			" -c              check if dudki is running\n"
+			" -c              check if the process is running\n"
+			"     (the above commands operate on dudki itself if no\n"
+			"      process name has been specified)\n"
 			" -e              ensure that dudki is running\n"
 			" -t              test configuration file and exit"
 #endif /* /HAVE_GETOPT_LONG */
 			<< endl;
 		    exit(0);
 		    break;
 		case 'V':
 		    cerr << VERSION << endl;
 		    exit(0);
 		    break;
 		case 'L':
 		    extern const char *COPYING;
 		    cerr << COPYING << endl;
 		    exit(0);
 		    break;
 		case 'f':
 		    config_file = optarg;
 		    break;
 		case 'k':
 		    if(op!=op_default) {
 			cerr << "Can't obey two or more orders at once" << endl;
 			exit(1);
 		    }
-		    op = op_term;
+		    op = op_signal; op_signum = SIGTERM;
 		    break;
 		case 'r':
 		    if(op!=op_default) {
 			cerr << "Can't obey two or more orders at once" << endl;
 			exit(1);
 		    }
-		    op = op_hup;
+		    op = op_signal; op_signum = SIGHUP;
 		    break;
 		case 'c':
 		    if(op!=op_default) {
 			cerr << "Can't obey two or more orders at once" << endl;
 			exit(1);
 		    }
-		    op = op_check;
+		    op = op_signal; op_signum = 0;
 		    break;
 		case 'e':
 		    if(op!=op_default) {
 			cerr << "Can't obey two or more orders at once" << endl;
 			exit(1);
 		    }
 		    op = op_ensure;
 		    break;
 		case 't':
 		    if(op!=op_default) {
 			cerr << "Can't obey two or more orders at once" << endl;
 			exit(1);
 		    }
 		    op = op_test;
 		    break;
 		default:
 		    cerr << "Huh??" << endl;
 		    exit(1);
 		    break;
 	    }
 	}
 	const char *sid = *argv;
 	const char *t;
 	while(t = index(sid,'/')) {
 	    sid = t; sid++;
 	}
 	openlog(sid,LOG_CONS|LOG_PERROR|LOG_PID,LOG_DAEMON);
 	configuration config;
 	config.parse(config_file);
 	switch(op) {
 	    case op_test:
 		cerr << "Configuration OK" << endl;
 		break;
-	    case op_hup:
-		signal_self(config,SIGHUP);
-		break;
-	    case op_term:
-		signal_self(config,SIGTERM);
-		break;
-	    case op_check:
-		try{
-		    signal_self(config,0);
-		    exit(0);
+	    case op_signal:
+		try {
+		    if(optind>=argc) {
+			signal_self(config,op_signum);
+		    }else{
+			int failures = 0;
+			for(int narg=optind;narg<argc;narg++) {
+			    try {
+				processes_t::const_iterator i = config.processes.find(argv[narg]);
+				if(i==config.processes.end())
+				    throw runtime_error("no such process configured");
+				i->second.signal(op_signum);
+			    }catch(exception& e) {
+				cerr << "dudki(" << argv[narg] << "): " << e.what() << endl;
+				failures++;
+			    }
+			}
+			if(failures)
+			    throw runtime_error("not all processes have been successfully signaled");
+		    }
+		    if(!op_signum)
+			exit(0);
 		}catch(exception& e) {
-		    exit(1);
+		    if(!op_signum)
+			exit(1);
 		}
 	    case op_ensure:
 		try {
 		    signal_self(config,0);
 		    break;
 		}catch(exception& e) {
 		    syslog(LOG_NOTICE,"The dudki process is down, taking its place");
 		    config.daemonize = true;
 		}catch(int zero) {
 		    // we throw zero in case we're ensuring that this very process is running.
 		    // we don't have to daemonize if we're daemonic.
 		    config.daemonize = false;
 		}
 	    case op_default:
 	    case op_work:
 		{
 		    if(config.daemonize) {
 			pid_t pf = fork();
 			if(pf<0)
 			    throw runtime_error(string(__PRETTY_FUNCTION__)+": failed to fork()");
 			if(pf) {
 			    _exit(0);
 			}
 		    }
diff --git a/src/process.cc b/src/process.cc
index bfab311..1ffac9f 100644
--- a/src/process.cc
+++ b/src/process.cc
@@ -1,56 +1,45 @@
 #include <stdio.h>
 #include <sys/types.h>
 #include <unistd.h>
 #include <signal.h>
 #include <pwd.h>
 #include <grp.h>
 #include <sys/wait.h>
 #include <syslog.h>
 #include <errno.h>
 #include <iostream>
 #include <fstream>
 #include <stdexcept>
 using namespace std;
 #include "process.h"
 #include "configuration.h"
 
 void process::check(const string& id,configuration& config) {
-    bool running = false;
-    ifstream pids(pidfile.c_str(),ios::in);
-    if(pids) {
-	pid_t pid = 0;
-	pids >> pid;
-	pids.close();
-	if(pid) {
-	    if(!kill(pid,0)) {
-		running = true;
-	    }
-	}
-    }
-    if(running){
+    try {
+	signal(0);
 	patience = 0;
-    }else{
+    }catch(exception& e) {
 	if(patience>60) { // TODO: configurable
 	    patience = 0;
 	}else{
 	    if(patience<10) { // TODO: configurable
 		syslog(LOG_NOTICE,"The process '%s' is down, trying to launch.",id.c_str());
 		do_notify(id,"Starting up",
 			"The named process seems to be down. Dudki will try\n"
 			"to revive it by running the specified command.\n",
 			config);
 		try {
 		    launch(id,config);
 		}catch(exception& e) {
 		    syslog(LOG_ERR,"Error trying to launch process '%s': %s",id.c_str(),e.what());
 		}
 	    }else if(patience==10){ // TODO: configurable like the above
 		syslog(LOG_NOTICE,"Giving up on process '%s' for a while",id.c_str());
 		do_notify(id,"Giving up",
 			"After a number of attempts to relaunch the named process\n"
 			"It still seems to be down. Dudki is giving up attempts\n"
 			"to revive the process for a while.\n",
 			config);
 	    }
 	    patience++;
 	}
@@ -164,24 +153,37 @@ void process::notify_mailto(const string& email,const string& id,const string& e
     }
     // parent
     close(files[0]);
     FILE *mta = fdopen(files[1],"w");
     for(headers_t::const_iterator i=mailto_headers.begin();i!=mailto_headers.end();++i) {
 	fprintf(mta,"%s: %s\n",i->first.c_str(),i->second.c_str());
     }
     for(headers_t::const_iterator i=config.mailto_headers.begin();i!=config.mailto_headers.end();++i) {
 	if(mailto_headers.find(i->first)!=mailto_headers.end())
 	    continue;
 	fprintf(mta,"%s: %s\n",i->first.c_str(),i->second.c_str());
     }
     fprintf(mta,
 	    "Subject: [%s] %s\n\n"
 	    "%s\n"
 	    "---\n"
 	    "This message was sent automatically by the 'dudki' daemon\n",
 	    id.c_str(), event.c_str(),
 	    description.c_str() );
     fclose(mta);
     int status;
     waitpid(pid,&status,0);
     // TODO: check the return code
 }
+
+void process::signal(int signum) const {
+    ifstream pids(pidfile.c_str(),ios::in);
+    if(!pids)
+	throw runtime_error("no pidfile found");
+    pid_t pid = 0;
+    pids >> pid;
+    pids.close();
+    if(!pid)
+	throw runtime_error("no pid in pidfile");
+    if(kill(pid,signum))
+	throw runtime_error("failed to signal process");
+}
diff --git a/src/process.h b/src/process.h
index b6d7091..27ee049 100644
--- a/src/process.h
+++ b/src/process.h
@@ -8,29 +8,31 @@ using namespace std;
 class configuration;
 
 typedef map<string,string> headers_t;
 
 class process {
     public:
 	string pidfile;
 	string restart_cmd;
 	string notify;
 	string user;
 	string group;
 	string chroot;
 	headers_t mailto_headers;
 
 	int patience;
 
 	process()
 	    : patience(0) { }
 
 	void check(const string& id,configuration& config);
 	void launch(const string& id,configuration& config);
 	void do_notify(const string& id,const string& event,const string& description,configuration& config);
 	void notify_mailto(const string& email,const string& id,const string& event,
 		const string& description,configuration& config);
+
+	void signal(int signum) const;
 };
 
 typedef map<string,process> processes_t;
 
 #endif /* __PROCESS_H */