// Note: caller must deallocate return value w/ delete []
char *
GridUniverseLogic::scratchFilePath(gman_node_t *gman_node)
{
MyString filename;
filename.formatstr("%s%p.%d",scratch_prefix,
gman_node,daemonCore->getpid());
char *prefix = temp_dir_path();
ASSERT(prefix);
// note: dircat allocates with new char[]
char *finalpath = dircat(prefix,filename.Value());
free(prefix);
return finalpath;
}
// create a temporary lock path
// return value must be freed via delete[] by caller.
char *
FileLock::GetTempPath()
{
const char *suffix = "";
char *result = NULL;
char *path = param("LOCAL_DISK_LOCK_DIR");
if (!path) {
path = temp_dir_path();
suffix = "condorLocks";
}
result = dirscat(path, suffix);
free(path);
return result;
}
GridUniverseLogic::gman_node_t *
GridUniverseLogic::StartOrFindGManager(const char* owner, const char* domain,
const char* attr_value, const char* attr_name, int cluster, int proc)
{
gman_node_t* gman_node;
int pid;
// If attr_value is an empty string, convert to NULL since code
// after this point expects that.
if ( attr_value && strlen(attr_value)==0 ) {
attr_value = NULL;
attr_name = NULL;
}
if ( (gman_node=lookupGmanByOwner(owner, attr_value, cluster, proc)) ) {
// found it
return gman_node;
}
// not found. fire one up! we want to run the GManager as the user.
// but first, make certain we are not shutting down...
if (!gman_pid_table) {
// destructor has already been called; we are probably
// closing down.
return NULL;
}
#ifndef WIN32
if (owner && strcasecmp(owner, "root") == 0 ) {
dprintf(D_ALWAYS, "Tried to start condor_gmanager as root.\n");
return NULL;
}
#endif
dprintf( D_FULLDEBUG, "Starting condor_gmanager for owner %s (%d.%d)\n",
owner, cluster, proc);
char *gman_binary;
gman_binary = param("GRIDMANAGER");
if ( !gman_binary ) {
dprintf(D_ALWAYS,"ERROR - GRIDMANAGER not defined in config file\n");
return NULL;
}
ArgList args;
MyString error_msg;
args.AppendArg("condor_gridmanager");
args.AppendArg("-f");
char *gman_args = param("GRIDMANAGER_ARGS");
if(!args.AppendArgsV1RawOrV2Quoted(gman_args,&error_msg)) {
dprintf( D_ALWAYS, "ERROR: failed to parse gridmanager args: %s\n",
error_msg.Value());
free(gman_binary);
free(gman_args);
return NULL;
}
free(gman_args);
// build a constraint
if ( !owner ) {
dprintf(D_ALWAYS,"ERROR - missing owner field\n");
free(gman_binary);
return NULL;
}
MyString constraint;
if ( !attr_name ) {
constraint.formatstr("(%s=?=\"%s\"&&%s==%d)",
ATTR_OWNER,owner,
ATTR_JOB_UNIVERSE,CONDOR_UNIVERSE_GRID);
} else {
constraint.formatstr("(%s=?=\"%s\"&&%s=?=\"%s\"&&%s==%d)",
ATTR_OWNER,owner,
attr_name,attr_value,
ATTR_JOB_UNIVERSE,CONDOR_UNIVERSE_GRID);
args.AppendArg("-A");
args.AppendArg(attr_value);
}
args.AppendArg("-C");
args.AppendArg(constraint.Value());
MyString full_owner_name(owner);
if ( domain && *domain ) {
full_owner_name.formatstr_cat( "@%s", domain );
}
args.AppendArg("-o");
args.AppendArg(full_owner_name.Value());
if (!init_user_ids(owner, domain)) {
dprintf(D_ALWAYS,"ERROR - init_user_ids() failed in GRIDMANAGER\n");
free(gman_binary);
return NULL;
}
static bool first_time_through = true;
if ( first_time_through ) {
// Note: Because first_time_through is static, this block runs only
// once per schedd invocation.
first_time_through = false;
// Clean up any old / abandoned scratch dirs.
dprintf(D_FULLDEBUG,"Checking for old gridmanager scratch dirs\n");
char *prefix = temp_dir_path();
ASSERT(prefix);
Directory tmp( prefix, PRIV_USER );
const char *f;
char const *dot;
int fname_pid;
int mypid = daemonCore->getpid();
int scratch_pre_len = strlen(scratch_prefix);
while ( (f=tmp.Next()) ) {
// skip regular files -- we only need to inspect subdirs
if ( !tmp.IsDirectory() ) {
continue;
}
// skip if it does not start with our prefix
if ( strncmp(scratch_prefix,f,scratch_pre_len) ) {
continue;
}
// skip if does not end w/ a pid
dot = strrchr(f,'.');
if ( !dot ) {
continue;
}
// skip if this pid is still alive and not ours
dot++; // skip over period
fname_pid = atoi(dot);
if ( fname_pid != mypid && daemonCore->Is_Pid_Alive(fname_pid) ) {
continue;
}
// if we made it here, blow away this subdir
if ( tmp.Remove_Current_File() ) {
dprintf(D_ALWAYS,"Removed old scratch dir %s\n",
tmp.GetFullPath());
}
} // end of while for cleanup of old scratch dirs
dprintf(D_FULLDEBUG,"Done checking for old scratch dirs\n");
if (prefix != NULL) {
free(prefix);
prefix = NULL;
}
} // end of once-per-schedd invocation block
// Create a temp dir for the gridmanager and append proper
// command-line arguments to tell where it is.
bool failed = false;
gman_node = new gman_node_t;
char *finalpath = scratchFilePath(gman_node);
priv_state saved_priv = set_user_priv();
if ( (mkdir(finalpath,0700)) < 0 ) {
// mkdir failed.
dprintf(D_ALWAYS,"ERROR - mkdir(%s,0700) failed in GRIDMANAGER, errno=%d (%s)\n",
finalpath, errno, strerror(errno));
failed = true;
}
set_priv(saved_priv);
uninit_user_ids();
args.AppendArg("-S"); // -S = "ScratchDir" argument
args.AppendArg(finalpath);
delete [] finalpath;
if ( failed ) {
// we already did dprintf reason to the log...
free(gman_binary);
delete gman_node;
return NULL;
}
if(IsFulldebug(D_FULLDEBUG)) {
MyString args_string;
args.GetArgsStringForDisplay(&args_string);
dprintf(D_FULLDEBUG,"Really Execing %s\n",args_string.Value());
}
pid = daemonCore->Create_Process(
gman_binary, // Program to exec
args, // Command-line args
PRIV_ROOT, // Run as root, so it can switch to
// PRIV_CONDOR
rid // Reaper ID
);
free(gman_binary);
if ( pid <= 0 ) {
dprintf ( D_ALWAYS, "StartOrFindGManager: Create_Process problems!\n" );
if (gman_node) delete gman_node;
return NULL;
}
// If we made it here, we happily started up a new gridmanager process
dprintf( D_ALWAYS, "Started condor_gmanager for owner %s pid=%d\n",
owner,pid);
// Make a new gman_node entry for our hashtable & insert it
if ( !gman_node ) {
gman_node = new gman_node_t;
}
gman_node->pid = pid;
gman_node->owner[0] = '\0';
gman_node->domain[0] = '\0';
if ( owner ) {
strcpy(gman_node->owner,owner);
}
if ( domain ) {
strcpy(gman_node->domain,domain);
}
MyString owner_key(owner);
if(attr_value){
owner_key += attr_value;
}
if (cluster) {
owner_key.formatstr_cat( "-%d.%d", cluster, proc );
}
ASSERT( gman_pid_table->insert(owner_key,gman_node) == 0 );
// start timer to signal gridmanager if we haven't already
if ( gman_node->add_timer_id == -1 ) { // == -1 means no timer set
gman_node->add_timer_id = daemonCore->Register_Timer(job_added_delay,
GridUniverseLogic::SendAddSignal,
"GridUniverseLogic::SendAddSignal");
daemonCore->Register_DataPtr(gman_node);
}
// All done
return gman_node;
}
