Turn on self-spawning: main loop should never crash but exists only to

respawn child when it does.  Add crash command to test this.  Add
ability to set logging level from ctrl port.

xwords4/relay/Makefile

@@ -28,7 +28,7 @@ SRC = xwrelay.cpp \
 
 OBJ = $(patsubst %.cpp,%.o,$(SRC))
 LDFLAGS += -pthread -g -lmcheck
-CPPFLAGS += -g -Wall -DSVN_REV=\"$(shell svnversion -n .)\"
+CPPFLAGS += -DSPAWN_SELF -g -Wall -DSVN_REV=\"$(shell svnversion -n .)\"
 # turn on semaphore debugging
 # CPPFLAGS += -DDEBUG_LOCKS
 

xwords4/relay/configs.cpp

@@ -41,7 +41,6 @@ RelayConfigs* RelayConfigs::instance = NULL;
 /* static */ RelayConfigs* 
 RelayConfigs::GetConfigs()
 {
-    assert( instance != NULL );
     return instance;
 }
 
@@ -107,6 +106,8 @@ RelayConfigs::parse( const char* fname )
                     m_serverName = value;
                 } else if ( 0 == strcmp( line, "IDFILE" ) ) {
                     m_idFileName = value;
+                } else if ( 0 == strcmp( line, "LOGLEVEL" ) ) {
+                    m_logLevel = atoi(value);
                 } else {
                     logf( XW_LOGERROR, "unknown key %s with value %s\n",
                           line, value );

xwords4/relay/configs.h

@@ -39,6 +39,8 @@ class RelayConfigs {
     time_t GetHeartbeatInterval() { return m_heartbeatInterval; }
     const char*  GetServerName() { return m_serverName.c_str(); }
     const char*  GetIdFileName() { return m_idFileName.c_str(); }
+    int GetLogLevel(void) { return m_logLevel; }
+    void SetLogLevel(int ll) { m_logLevel = ll; }
 
  private:
     RelayConfigs( const char* cfile );
@@ -48,6 +50,7 @@ class RelayConfigs {
     time_t m_heartbeatInterval;
     int m_ctrlport;
     int m_port;
+    int m_logLevel;
     int m_nWorkerThreads;
     std::string m_serverName;
     std::string m_idFileName;

xwords4/relay/ctrl.cpp

@@ -43,10 +43,12 @@
 #include "crefmgr.h"
 #include "mlock.h"
 #include "xwrelay_priv.h"
+#include "configs.h"
 
 /* this is *only* for testing.  Don't abuse!!!! */
 extern pthread_rwlock_t gCookieMapRWLock;
 
+/* Return of true means exit the ctrl thread */
 typedef bool (*CmdPtr)( int socket, const char** args );
 
 typedef struct FuncRec {
@@ -70,6 +72,7 @@ static bool cmd_set( int socket, const char** args );
 static bool cmd_shutdown( int socket, const char** args );
 static bool cmd_rev( int socket, const char** args );
 static bool cmd_uptime( int socket, const char** args );
+static bool cmd_crash( int socket, const char** args );
 
 static void
 print_to_sock( int sock, bool addCR, const char* what, ... )
@@ -102,6 +105,7 @@ static const FuncRec gFuncs[] = {
     { "set", cmd_set },
     { "rev", cmd_rev },
     { "uptime", cmd_uptime },
+    { "crash", cmd_crash },
 };
 
 static bool
@@ -110,9 +114,11 @@ cmd_quit( int socket, const char** args )
     if ( 0 == strcmp( "help", args[1] ) ) {
         print_to_sock( socket, true, "* %s (disconnect from ctrl port)", 
                        args[0] );
+        return false;
     } else {
+        print_to_sock( socket, true, "bye bye" );
+        return true;
     }
-    return 0;
 }
 
 static void
@@ -136,14 +142,14 @@ static bool
 cmd_start( int socket, const char** args )
 {
     print_to_sock( socket, true, "* %s (unimplemented)", args[0] );
-    return 1;
+    return false;
 }
 
 static bool
 cmd_stop( int socket, const char** args )
 {
     print_to_sock( socket, true, "* %s (unimplemented)", args[0] );
-    return 1;
+    return false;
 }
 
 static bool
@@ -185,24 +191,50 @@ cmd_kill_eject( int socket, const char** args )
             ;
         print_to_sock( socket, true, msg, args[0], expl, args[0], args[0] );
     }
-    return 1;
+    return false;
 } /* cmd_kill_eject */
 
 static bool
 cmd_get( int socket, const char** args )
 {
-    print_to_sock( socket, true,
-                   "* %s -- lists all attributes (unimplemented)\n"
-                   "* %s <attribute> (unimplemented)",
-                   args[0], args[0] );
-    return 1;
+    if ( 0 == strcmp( args[1], "help" ) ) {
+        print_to_sock( socket, true,
+                       "* %s -- lists all attributes (unimplemented)\n"
+                       "* %s loglevel",
+                       args[0], args[0] );
+    } else {
+        const char* attr = args[1];
+        if ( (NULL != attr) && (0 == strcmp( attr, "loglevel" )) ) {
+            RelayConfigs* rc = RelayConfigs::GetConfigs();
+            if ( NULL != rc ) {
+                print_to_sock( socket, true, "loglevel=%d\n", 
+                               rc->GetLogLevel() );
+            } else {
+                logf( XW_LOGERROR, "RelayConfigs::GetConfigs() => NULL" );
+            }
+        }
+    }
+    return false;
 }
 
 static bool
 cmd_set( int socket, const char** args )
 {
-    print_to_sock( socket, true, "* %s <attribute> (unimplemented)", args[0] );
-    return 1;
+    if ( 0 == strcmp( args[1], "help" ) ) {
+        print_to_sock( socket, true, "* %s loglevel <n>\n", args[0] );
+    } else {
+        const char* attr = args[1];
+        const char* val = args[2];
+        if ( (NULL != attr) 
+             && (0 == strcmp( attr, "loglevel" ))
+                 && (NULL != val) ) {
+            RelayConfigs* rc = RelayConfigs::GetConfigs();
+            if ( rc != NULL ) {
+                rc->SetLogLevel( atoi(val) );
+            }
+        }
+    }
+    return false;
 }
 
 static bool
@@ -215,7 +247,7 @@ cmd_rev( int socket, const char** args )
     } else {
         print_to_sock( socket, true, "svn rev: %s", SVN_REV );
     }
-    return 0;
+    return false;
 }
 
 static bool
@@ -241,7 +273,23 @@ cmd_uptime( int socket, const char** args )
                        "uptime: %d days, %d hours, %d minutes, %ld seconds",
                        days, hours, minutes, seconds );
     }
-    return 0;
+    return false;
+}
+
+static bool
+cmd_crash( int socket, const char** args )
+{
+    if ( 0 == strcmp( args[1], "help" ) ) {
+        print_to_sock( socket, true,
+                       "* %s -- fires an assert (debug case) or divides-by-zero",
+                       args[0] );
+    } else {
+        assert(0);
+        int i = 1;
+        while ( i > 0 ) --i;
+        return 6/i > 0;
+    }
+    return false;
 }
 
 static bool
@@ -250,7 +298,7 @@ cmd_shutdown( int socket, const char** args )
     print_to_sock( socket, true,
                    "* %s  -- shuts down relay (exiting main) (unimplemented)",
                    args[0] );
-    return 1;
+    return false;
 }
 
 static void

xwords4/relay/states.cpp

@@ -81,6 +81,10 @@ StateTable g_stateTable[] = {
 { XWS_CONNECTING,     XWE_DISCONNMSG,    XWA_DISCONNECT,   XWS_CONNECTING },
 { XWS_MISSING,        XWE_DISCONNMSG,    XWA_DISCONNECT,   XWS_MISSING },
 
+    /* I'm seeing this but not sure how to handle.  Might disconnect be
+       needed now */
+{ XWS_MISSING,        XWE_FORWARDMSG,    XWA_DISCONNECT,   XWS_MISSING },
+
 { XWS_ANY,            XWE_NOMORESOCKETS, XWA_NONE,         XWS_DEAD },
 { XWS_ANY,            XWE_SHUTDOWN,      XWA_SHUTDOWN,     XWS_DEAD },
 
@@ -115,33 +119,36 @@ StateTable g_stateTable[] = {
     /* This is our bread-n-butter */
 { XWS_ALLCONNECTED,   XWE_FORWARDMSG,    XWA_FWD,           XWS_ALLCONNECTED },
 
-{ XWS_DEAD,           XWE_REMOVESOCKET,  XWA_REMOVESOCKET,  XWS_DEAD },
+{ XWS_DEAD,           XWE_REMOVESOCKET,  XWA_REMOVESOCKET,  XWS_DEAD }
 
-    /* Marks end of table */
-{ XWS_NONE,           XWE_NONE,          XWA_NONE,          XWS_NONE }
 };
 
 
-int
+bool
 getFromTable( XW_RELAY_STATE curState, XW_RELAY_EVENT curEvent,
               XW_RELAY_ACTION* takeAction, XW_RELAY_STATE* nextState )
 {
+    bool found = false;
     StateTable* stp = g_stateTable;
-    while ( stp->stateStart != XWS_NONE ) {
+    const StateTable* end = stp + sizeof(g_stateTable)/sizeof(g_stateTable[0]);
+    while ( stp < end ) {
         if ( stp->stateStart == curState || stp->stateStart == XWS_ANY ) {
             if ( stp->stateEvent == curEvent || stp->stateEvent == XWE_ANY ) {
                 *takeAction = stp->stateAction;
                 *nextState = stp->stateEnd;
-                return 1;
+                found = true;
+                break;
             }
         }
         ++stp;
     }
 
-    logf( XW_LOGERROR, "==> ERROR :: unable to find transition from %s on event %s",
-          stateString(curState), eventString(curEvent) );
-
-    return 0;
+    if ( !found ) {
+        logf( XW_LOGERROR, "==> ERROR :: unable to find transition from %s "
+              "on event %s",
+              stateString(curState), eventString(curEvent) );
+    }
+    return found;
 } /* getFromTable */
 
 #define CASESTR(s) case s: return #s

xwords4/relay/states.h

@@ -149,8 +149,8 @@ typedef enum {
 
 } XW_RELAY_ACTION;
 
-int getFromTable( XW_RELAY_STATE curState, XW_RELAY_EVENT curEvent,
-                  XW_RELAY_ACTION* takeAction, XW_RELAY_STATE* nextState );
+bool getFromTable( XW_RELAY_STATE curState, XW_RELAY_EVENT curEvent,
+                   XW_RELAY_ACTION* takeAction, XW_RELAY_STATE* nextState );
 
 
 char* stateString( XW_RELAY_STATE state );

xwords4/relay/xwrelay.conf

@@ -26,4 +26,8 @@ SERVERNAME=eehouse.org
 
 # Where will the file live that stores the last ID used for a new
 # connName.
-#IDFILE=/home/ehouse/xwrelay_id.txt
+IDFILE=/home/eehouse/xwrelay_id.txt
+
+# Initial level of logging.  See xwrelay_priv.h for values.  Currently
+# 0 means errors only, 1 info, 2 verbose and 3 very verbose.
+LOGLEVEL=0

xwords4/relay/xwrelay.cpp

@@ -73,7 +73,8 @@
 void
 logf( XW_LogLevel level, const char* format, ... )
 {
-    if ( level <= XW_LOGINFO ) {
+    RelayConfigs* rc = RelayConfigs::GetConfigs();
+    if ( NULL == rc || level <= rc->GetLogLevel() ) {
         FILE* where = stderr;
         struct tm* timp;
         struct timeval tv;
@@ -376,6 +377,16 @@ static int
 make_socket( unsigned long addr, unsigned short port )
 {
     int sock = socket( AF_INET, SOCK_STREAM, 0 );
+    assert( sock );
+
+    /* We may be relaunching after crashing with sockets open.  SO_REUSEADDR
+       allows them to be immediately rebound. */
+    int t = true;
+    if ( 0 != setsockopt( sock, SOL_SOCKET, SO_REUSEADDR, &t, sizeof(t) ) ) {
+        logf( XW_LOGERROR, "setsockopt failed. errno = %s (%d)\n", 
+              strerror(errno), errno );
+        return -1;
+    }
 
     struct sockaddr_in sockAddr;
     sockAddr.sin_family = AF_INET;
@@ -543,6 +554,8 @@ int main( int argc, char** argv )
     PermID::SetServerName( serverName );
     PermID::SetIDFileName( idFileName );
 
+    /* add signal handling here */
+
 #ifdef SPAWN_SELF
     /* loop forever, relaunching children as they die. */
     for ( ; ; ) {
@@ -554,7 +567,6 @@ int main( int argc, char** argv )
             logf( XW_LOGINFO, "parent waiting on child pid=%d", pid );
             waitpid( pid, &status, 0 );
             printWhy( status );
-            sleep( 45 );        /* give time to close sockets? */
         } else {
             logf( XW_LOGERROR, "fork() => %s", strerror(errno) );
         }