/****** uti/prog/sge_get_alias_path() *****************************************
* NAME
* sge_get_alias_path() -- Return the path of the 'alias_file'
*
* SYNOPSIS
* const char* sge_get_alias_path(void)
*
* FUNCTION
* Return the path of the 'alias_file'
*
* NOTES
* MT-NOTE: sge_get_alias_path() is MT safe
*
******************************************************************************/
const char *sge_get_alias_path(void)
{
const char *sge_root, *sge_cell;
char *cp;
int len;
SGE_STRUCT_STAT sbuf;
DENTER_(TOP_LAYER, "sge_get_alias_path");
sge_root = sge_get_root_dir(1, NULL, 0, 1);
sge_cell = sge_get_default_cell();
if (SGE_STAT(sge_root, &sbuf)) {
CRITICAL((SGE_EVENT, MSG_SGETEXT_SGEROOTNOTFOUND_S , sge_root));
SGE_EXIT(NULL, 1);
}
len = strlen(sge_root) + strlen(sge_cell) + strlen(COMMON_DIR) + strlen(ALIAS_FILE) + 5;
if (!(cp = malloc(len))) {
CRITICAL((SGE_EVENT, MSG_MEMORY_MALLOCFAILEDFORPATHTOHOSTALIASFILE ));
SGE_EXIT(NULL, 1);
}
sprintf(cp, "%s/%s/%s/%s", sge_root, sge_cell, COMMON_DIR, ALIAS_FILE);
DRETURN_(cp);
}
/****** qmaster/setup_qmaster/sge_setup_qmaster() ******************************
* NAME
* sge_setup_qmaster() -- setup qmaster
*
* SYNOPSIS
* int sge_setup_qmaster(char* anArgv[])
*
* FUNCTION
* Process commandline arguments. Remove qmaster lock file. Write qmaster
* host to the 'act_qmaster' file. Initialize qmaster and reporting. Write
* qmaster PID file.
*
* NOTE: Before this function is invoked, qmaster must become admin user.
*
* INPUTS
* char* anArgv[] - commandline argument vector
*
* RESULT
* 0 - success
*
* NOTES
* MT-NOTE: sge_setup_qmaster() is NOT MT safe!
* MT-NOTE:
* MT-NOTE: This function must be called exclusively, with the qmaster main
* MT-NOTE: thread being the *only* active thread. In other words, do not
* MT-NOTE: invoke this function after any additional thread (directly or
* MT-NOTE: indirectly) has been created.
*
* Do *not* write the qmaster pid file, before 'qmaster_init()' did return
* successfully. Otherwise, if we do have a running qmaster and a second
* qmaster is started (illegally) on the same host, the second qmaster will
* overwrite the pid of the qmaster started first. The second qmaster will
* detect it's insubordinate doing and terminate itself, thus leaving behind
* a useless pid.
*
*******************************************************************************/
int sge_setup_qmaster(sge_gdi_ctx_class_t *ctx, char* anArgv[])
{
char err_str[1024];
const char *qualified_hostname = ctx->get_qualified_hostname(ctx);
const char *act_qmaster_file = ctx->get_act_qmaster_file(ctx);
DENTER(TOP_LAYER, "sge_setup_qmaster");
umask(022); /* this needs a better solution */
process_cmdline(anArgv);
INFO((SGE_EVENT, SFNMAX, MSG_STARTUP_BEGINWITHSTARTUP));
qmaster_unlock(QMASTER_LOCK_FILE);
if (write_qm_name(qualified_hostname, act_qmaster_file, err_str)) {
ERROR((SGE_EVENT, "%s\n", err_str));
SGE_EXIT(NULL, 1);
}
qmaster_init(ctx, anArgv);
sge_write_pid(QMASTER_PID_FILE);
DEXIT;
return 0;
} /* sge_setup_qmaster() */
/****** qmaster/setup_qmaster/sge_qmaster_thread_init() ************************
* NAME
* sge_qmaster_thread_init() -- Initialize a qmaster thread.
*
* SYNOPSIS
* int sge_qmaster_thread_init(bool switch_to_admin_user)
*
* FUNCTION
* Subsume functions which need to be called immediately after thread
* startup. This function does make sure that the thread local data
* structures do contain reasonable values.
*
* INPUTS
* bool switch_to_admin_user - become admin user if set to true
*
* RESULT
* 0 - success
*
* NOTES
* MT-NOTE: sge_qmaster_thread_init() is MT safe
* MT-NOTE:
* MT-NOTE: sge_qmaster_thread_init() should be invoked at the beginning
* MT-NOTE: of a thread function.
*
*******************************************************************************/
int
sge_qmaster_thread_init(sge_gdi_ctx_class_t **ctx_ref, u_long32 prog_id,
u_long32 thread_id, bool switch_to_admin_user)
{
const char *admin_user = NULL;
lList *alp = NULL;
sge_gdi_ctx_class_t *ctx = NULL;
DENTER(TOP_LAYER, "sge_qmaster_thread_init");
lInit(nmv);
if (sge_setup2(ctx_ref, prog_id, thread_id, &alp, true) != AE_OK) {
answer_list_output(&alp);
SGE_EXIT((void**)ctx_ref, 1);
}
ctx = *ctx_ref;
ctx->reresolve_qualified_hostname(ctx);
DEBUG((SGE_EVENT,"%s: qualified hostname \"%s\"\n", SGE_FUNC, ctx->get_qualified_hostname(ctx)));
admin_user = ctx->get_admin_user(ctx);
if (switch_to_admin_user == true) {
char str[MAX_STRING_SIZE];
if (sge_set_admin_username(admin_user, str) == -1) {
CRITICAL((SGE_EVENT, SFNMAX, str));
SGE_EXIT((void**)ctx_ref, 1);
}
if (sge_switch2admin_user()) {
CRITICAL((SGE_EVENT, SFNMAX, MSG_ERROR_CANTSWITCHTOADMINUSER));
SGE_EXIT((void**)ctx_ref, 1);
}
}
DEXIT;
return 0;
} /* sge_qmaster_thread_init() */
/****** uti/prog/sge_get_root_dir() *******************************************
* NAME
* sge_get_root_dir() -- SGE/SGEEE installation directory
*
* SYNOPSIS
* const char* sge_get_root_dir(int do_exit,
* char *buffer,
* size_t size,
* int do_error_log )
*
* FUNCTION
* This function returns the installation directory of SGE/SGEEE.
* This directory is defined by the SGE_ROOT environment variable
* of the calling process.
* If the environment variable does not exist or is not set then
* this function will handle this as error and return NULL
* (do_exit = 0). If 'do_exit' is 1 and an error occures, the
* function will terminate the calling application.
*
* INPUTS
* int do_exit - Terminate the application in case of an error
* char *buffer - buffer to be used for error message
* size_t size - size of buffer
* int do_error_log - enable/disable error logging
*
* RESULT
* const char* - Root directory of the SGE/SGEEE installation
*
* NOTES
* MT-NOTE: sge_get_arch() is MT safe
*******************************************************************************/
const char *sge_get_root_dir(int do_exit, char *buffer, size_t size, int do_error_log)
{
char *sge_root;
char *s;
DENTER_(TOP_LAYER, "sge_get_root_dir");
/*
* Read some env variables
*/
sge_root = getenv("SGE_ROOT");
/*
* Check the env variables
*/
if (sge_root) {
s = sge_root;
} else {
goto error;
}
if (!s || strlen(s)==0) {
goto error;
} else {
/*
* Get rid of trailing slash
*/
if (s[strlen(s)-1] == '/') {
s[strlen(s)-1] = '\0';
}
}
DRETURN_(s);
error:
if (do_error_log) {
if (buffer != NULL) {
sge_strlcpy(buffer, MSG_SGEROOTNOTSET, size);
} else {
CRITICAL((SGE_EVENT, MSG_SGEROOTNOTSET));
}
}
DEXIT_;
if (do_exit) {
SGE_EXIT(NULL, 1);
}
return NULL;
}
/*---------------------------------------------------------------------
* parse_cmdline_execd
*---------------------------------------------------------------------*/
static void parse_cmdline_execd(char **argv)
{
lList *ref_list = NULL, *alp = NULL, *pcmdline = NULL;
lListElem *aep;
u_long32 help = 0;
DENTER(TOP_LAYER, "parse_cmdline_execd");
alp = sge_parse_cmdline_execd(argv+1, &pcmdline);
if(alp) {
/*
** high level parsing error! show answer list
*/
for_each(aep, alp) {
fprintf(stderr, "%s", lGetString(aep, AN_text));
}
lFreeList(&alp);
lFreeList(&pcmdline);
/* TODO: replace with alpp and DRETURN */
SGE_EXIT(NULL, 1);
}
/****** qmaster/setup_qmaster/process_cmdline() ********************************
* NAME
* process_cmdline() -- Handle command line arguments
*
* SYNOPSIS
* static void process_cmdline(char **anArgv)
*
* FUNCTION
* Handle command line arguments. Parse argument vector and handle options.
*
* INPUTS
* char **anArgv - pointer to agrument vector
*
* RESULT
* void - none
*
* NOTES
* MT-NOTE: process_cmdline() is NOT MT safe.
*
*******************************************************************************/
static void process_cmdline(char **anArgv)
{
lList *alp, *pcmdline;
lListElem *aep;
u_long32 help = 0;
DENTER(TOP_LAYER, "process_cmdline");
alp = pcmdline = NULL;
alp = parse_cmdline_qmaster(&anArgv[1], &pcmdline);
if(alp) {
/*
** high level parsing error! show answer list
*/
for_each(aep, alp) {
fprintf(stderr, "%s", lGetString(aep, AN_text));
}
lFreeList(&alp);
lFreeList(&pcmdline);
SGE_EXIT(NULL, 1);
}
/****** uti/stdio/sge_peopen() ************************************************
* NAME
* sge_peopen_r() -- Advanced popen()
*
* SYNOPSIS
* pid_t sge_peopen_r(const char *shell, int login_shell,
* const char *command, const char *user,
* char **env, FILE **fp_in, FILE **fp_out,
* FILE **fp_err)
*
* FUNCTION
* Advanced popen() with additional parameters:
* - free shell usage
* - login shell if wanted
* - user under which to start (for root only)
* - stdin and stderr file pointers
* - wait for exactly the process we started
* File descriptors have to be closed with sge_peclose().
*
* This function is reentrant as long as env is not provided to
* this function. This means that the function can be used in
* multi thread processed as long as env is not used.
*
* INPUTS
* const char *shell - which shell to use
* int login_shell - make it a login shell?
* const char *command - name of the program
* const char *user - user under which to start (for root only)
* char **env - env variables to add to child
* FILE **fp_in - file input stream
* FILE **fp_out - file output stream
* FILE **fp_err - file error stream
*
* RESULT
* pid_t - process id
*
* NOTES
* MT-NOTE: sge_peopen() is MT safe
*
* DO NOT ADD ASYNC SIGNAL UNSAFE FUNCTIONS BETWEEN FORK AND EXEC
* DUE TO THE FACT THAT THIS FUNCTION WILL BE USED IN QMASTER
* (MULTITHREADED ENVIRONMENT) THIS MIGHT CAUSE A DEADLOCK
* IN A MASTER THREAD.
*
* SEE ALSO
* uti/stdio/sge_peclose()
******************************************************************************/
pid_t sge_peopen_r(const char *shell, int login_shell, const char *command,
const char *user, char **env, FILE **fp_in, FILE **fp_out,
FILE **fp_err, bool null_stderr)
{
pid_t pid;
int pipefds[3][2];
int i;
char arg0[256];
#if defined(SOLARIS)
char err_str[256];
#endif
struct passwd *pw = NULL;
uid_t myuid;
uid_t tuid;
DENTER(TOP_LAYER, "sge_peopen_r");
if (sge_has_admin_user()) {
sge_switch2start_user();
}
myuid = geteuid();
tuid = myuid;
/*
* open pipes - close on failure
*/
for (i = 0; i < 3; i++) {
if (pipe(pipefds[i]) != 0) {
while (--i >= 0) {
close(pipefds[i][0]);
close(pipefds[i][1]);
}
ERROR((SGE_EVENT, MSG_SYSTEM_FAILOPENPIPES_SS, command, strerror(errno)));
if (sge_has_admin_user()) {
sge_switch2admin_user();
}
DRETURN(-1);
}
}
/*
* set arg0 for exec call correctly to that
* either a normal shell or a login shell will be started
*/
if (login_shell) {
strcpy(arg0, "-");
} else {
strcpy(arg0, "");
}
strcat(arg0, shell);
DPRINTF(("arg0 = %s\n", arg0));
DPRINTF(("arg1 = -c\n"));
DPRINTF(("arg2 = %s\n", command));
/*
* prepare the change of the user which might be done after fork()
* if a user name is provided.
*
* this has to be done before the fork() afterwards it might cause
* a deadlock of the child because getpwnam() is not async-thread safe.
*/
if (user) {
struct passwd pw_struct;
int size = get_pw_buffer_size();
char *buffer = sge_malloc(size);
/*
* get information about the target user
*/
if (buffer != NULL) {
pw = sge_getpwnam_r(user, &pw_struct, buffer, size);
if (pw == NULL) {
ERROR((SGE_EVENT, MSG_SYSTEM_NOUSERFOUND_SS, user, strerror(errno)));
FREE(buffer);
if (sge_has_admin_user()) {
sge_switch2admin_user();
}
DRETURN(-1);
}
} else {
ERROR((SGE_EVENT, MSG_UTI_MEMPWNAM));
FREE(buffer);
if (sge_has_admin_user()) {
sge_switch2admin_user();
}
DRETURN(-1);
}
DPRINTF(("was able to resolve user\n"));
/*
* only prepare change of user if target user is different from current one
*/
if (myuid != pw->pw_uid) {
#if !(defined(WIN32) || defined(INTERIX)) /* var not needed */
int res;
#endif
if (myuid != SGE_SUPERUSER_UID) {
DPRINTF(("only root is allowed to switch to a different user\n"));
ERROR((SGE_EVENT, MSG_SYSTEM_NOROOTRIGHTSTOSWITCHUSER));
FREE(buffer);
DRETURN(-2);
}
DPRINTF(("Before initgroups\n"));
#if !(defined(WIN32) || defined(INTERIX)) /* initgroups not called */
res = initgroups(pw->pw_name, pw->pw_gid);
# if defined(SVR3) || defined(sun)
if (res < 0)
# else
if (res)
# endif
{
ERROR((SGE_EVENT, MSG_SYSTEM_INITGROUPSFORUSERFAILED_ISS, res, user, strerror(errno)));
FREE(buffer);
SGE_EXIT(NULL, 1);
}
DPRINTF(("Initgroups was successful\n"));
#endif /* WIN32 */
}
DPRINTF(("user = %s\n", user));
DPRINTF(("myuid = %d\n", (int)myuid));
if (pw != NULL) {
tuid = pw->pw_uid;
DPRINTF(("target uid = %d\n", (int)tuid));
}
FREE(buffer);
}
DPRINTF(("Now process will fork\n"));
#if defined(SOLARIS)
pid = sge_smf_contract_fork(err_str, 256);
#else
pid = fork();
#endif
/*
* in the child pid is 0
*/
if (pid == 0) {
/*
* close all fd's except that ones mentioned in keep_open
*/
int keep_open[6];
keep_open[0] = 0;
keep_open[1] = 1;
keep_open[2] = 2;
keep_open[3] = pipefds[0][0];
keep_open[4] = pipefds[1][1];
keep_open[5] = pipefds[2][1];
sge_close_all_fds(keep_open, 6);
/*
* shall we redirect stderr to /dev/null? Then
* - open "/dev/null"
* - set stderr to "dev/null"
* - close the stderr-pipe
* otherwise
* - redirect stderr to the pipe
*/
if (null_stderr) {
int fd = open("/dev/null", O_WRONLY);
if (fd != -1) {
close(2);
dup(fd);
close(pipefds[2][1]);
} else {
SGE_EXIT(NULL, 1);
}
} else {
close(2);
dup(pipefds[2][1]);
}
/*
* redirect stdin and stdout to the pipes
*/
close(0);
close(1);
dup(pipefds[0][0]);
dup(pipefds[1][1]);
if (pw != NULL) {
int lret = setuid(tuid);
if (lret) {
SGE_EXIT(NULL, 1);
}
}
/*
* set the environment if we got one as argument
*/
if (env != NULL) {
if (pw != NULL) {
addenv("HOME", pw->pw_dir);
addenv("SHELL", pw->pw_shell);
addenv("USER", pw->pw_name);
addenv("LOGNAME", pw->pw_name);
}
addenv("PATH", SGE_DEFAULT_PATH);
for(; *env; env++) {
putenv(*env);
}
}
execlp(shell, arg0, "-c", command, NULL);
}
if (pid < 0) {
for (i=0; i<3; i++) {
close(pipefds[i][0]);
close(pipefds[i][1]);
}
#if defined(SOLARIS)
if (pid < -1 && err_str) {
ERROR((SGE_EVENT, MSG_SMF_FORK_FAILED_SS, "sge_peopen()", err_str));
}
#endif
if (sge_has_admin_user()) {
sge_switch2admin_user();
}
DRETURN(-1);
}
/* close the childs ends of the pipes */
close(pipefds[0][0]);
close(pipefds[1][1]);
close(pipefds[2][1]);
/* return filehandles for stdin and stdout */
*fp_in = fdopen(pipefds[0][1], "a");
*fp_out = fdopen(pipefds[1][0], "r");
/* is stderr redirected to /dev/null? */
if (null_stderr) {
/* close the pipe and return NULL as filehandle */
close(pipefds[2][0]);
*fp_err = NULL;
} else {
/* return filehandle for stderr */
*fp_err = fdopen(pipefds[2][0], "r");
}
if (sge_has_admin_user()) {
sge_switch2admin_user();
}
DRETURN(pid);
}
bool sge_parse_qrsub(sge_gdi_ctx_class_t *ctx, lList *pcmdline, lList **alpp, lListElem **ar)
{
lListElem *ep = NULL, *next_ep = NULL;
lList *lp = NULL;
DENTER(TOP_LAYER, "sge_parse_qrsub");
/* -help print this help */
if ((ep = lGetElemStr(pcmdline, SPA_switch, "-help"))) {
lRemoveElem(pcmdline, &ep);
sge_usage(QRSUB, stdout);
DEXIT;
SGE_EXIT((void **)&ctx, 0);
}
/* -a date_time start time in [[CC]YY]MMDDhhmm[.SS] SGE_ULONG */
while ((ep = lGetElemStr(pcmdline, SPA_switch, "-a"))) {
lSetUlong(*ar, AR_start_time, lGetUlong(ep, SPA_argval_lUlongT));
lRemoveElem(pcmdline, &ep);
}
/* -e date_time end time in [[CC]YY]MMDDhhmm[.SS] SGE_ULONG*/
while ((ep = lGetElemStr(pcmdline, SPA_switch, "-e"))) {
lSetUlong(*ar, AR_end_time, lGetUlong(ep, SPA_argval_lUlongT));
lRemoveElem(pcmdline, &ep);
}
/* -d time duration in TIME format SGE_ULONG */
while ((ep = lGetElemStr(pcmdline, SPA_switch, "-d"))) {
lSetUlong(*ar, AR_duration, lGetUlong(ep, SPA_argval_lUlongT));
lRemoveElem(pcmdline, &ep);
}
/* -w e/v validate availability of AR request, default e SGE_ULONG */
while ((ep = lGetElemStr(pcmdline, SPA_switch, "-w"))) {
lSetUlong(*ar, AR_verify, lGetInt(ep, SPA_argval_lIntT));
lRemoveElem(pcmdline, &ep);
}
/* -N name AR name SGE_STRING */
while ((ep = lGetElemStr(pcmdline, SPA_switch, "-N"))) {
lSetString(*ar, AR_name, lGetString(ep, SPA_argval_lStringT));
lRemoveElem(pcmdline, &ep);
}
/* -A account_string AR name in accounting record SGE_STRING */
while ((ep = lGetElemStr(pcmdline, SPA_switch, "-A"))) {
lSetString(*ar, AR_account, lGetString(ep, SPA_argval_lStringT));
lRemoveElem(pcmdline, &ep);
}
/* -l resource_list request the given resources SGE_LIST */
parse_list_simple(pcmdline, "-l", *ar, AR_resource_list, 0, 0, FLG_LIST_APPEND);
centry_list_remove_duplicates(lGetList(*ar, AR_resource_list));
/* -u wc_user access list SGE_LIST */
/* -u ! wc_user TBD: Think about eval_expression support in compare allowed and excluded lists */
parse_list_simple(pcmdline, "-u", *ar, AR_acl_list, ARA_name, 0, FLG_LIST_MERGE);
/* -u ! list separation */
lp = lGetList(*ar, AR_acl_list);
next_ep = lFirst(lp);
while ((ep = next_ep)) {
bool is_xacl = false;
const char *name = lGetString(ep, ARA_name);
next_ep = lNext(ep);
if (name[0] == '!') { /* move this element to xacl_list */
is_xacl = true;
name++;
}
if (!is_hgroup_name(name)) {
struct passwd *pw;
struct passwd pw_struct;
char *buffer;
int size;
stringT group;
size = get_pw_buffer_size();
buffer = sge_malloc(size);
pw = sge_getpwnam_r(name, &pw_struct, buffer, size);
if (pw == NULL) {
answer_list_add_sprintf(alpp, STATUS_EUNKNOWN, ANSWER_QUALITY_ERROR, MSG_USER_XISNOKNOWNUSER_S, name);
FREE(buffer);
DRETURN(false);
}
sge_gid2group(pw->pw_gid, group, MAX_STRING_SIZE, MAX_NIS_RETRIES);
lSetString(ep, ARA_group, group);
FREE(buffer);
}
if (is_xacl) {
lListElem *new_ep = lAddSubStr(*ar, ARA_name, name, AR_xacl_list, ARA_Type);
lSetString(new_ep, ARA_group, lGetString(ep, ARA_group));
lRemoveElem(lp, &ep);
}
}
/* -q wc_queue_list reserve in queue(s) SGE_LIST */
parse_list_simple(pcmdline, "-q", *ar, AR_queue_list, 0, 0, FLG_LIST_APPEND);
/* -pe pe_name slot_range reserve slot range for parallel jobs */
while ((ep = lGetElemStr(pcmdline, SPA_switch, "-pe"))) {
lSetString(*ar, AR_pe, lGetString(ep, SPA_argval_lStringT)); /* SGE_STRING, */
lSwapList(*ar, AR_pe_range, ep, SPA_argval_lListT); /* SGE_LIST */
lRemoveElem(pcmdline, &ep);
}
/* AR_master_queue_list -masterq wc_queue_list, SGE_LIST bind master task to queue(s) */
parse_list_simple(pcmdline, "-masterq", *ar, AR_master_queue_list, 0, 0, FLG_LIST_APPEND);
/* -ckpt ckpt-name reserve in queue with ckpt method SGE_STRING */
while ((ep = lGetElemStr(pcmdline, SPA_switch, "-ckpt"))) {
lSetString(*ar, AR_checkpoint_name, lGetString(ep, SPA_argval_lStringT));
lRemoveElem(pcmdline, &ep);
}
/* -m b/e/a/n define mail notification events SGE_ULONG */
while ((ep = lGetElemStr(pcmdline, SPA_switch, "-m"))) {
u_long32 ul;
u_long32 old_mail_opts;
ul = lGetInt(ep, SPA_argval_lIntT);
if ((ul & NO_MAIL)) {
lSetUlong(*ar, AR_mail_options, 0);
} else {
old_mail_opts = lGetUlong(*ar, AR_mail_options);
lSetUlong(*ar, AR_mail_options, ul | old_mail_opts);
}
lRemoveElem(pcmdline, &ep);
}
/* -M user[@host],... notify these e-mail addresses SGE_LIST*/
parse_list_simple(pcmdline, "-M", *ar, AR_mail_list, MR_host, MR_user, FLG_LIST_MERGE);
/* -he yes/no hard error handling SGE_ULONG */
while ((ep = lGetElemStr(pcmdline, SPA_switch, "-he"))) {
lSetUlong(*ar, AR_error_handling, lGetUlong(ep, SPA_argval_lUlongT));
lRemoveElem(pcmdline, &ep);
}
/* -now reserve in queues with qtype interactive SGE_ULONG */
while ((ep = lGetElemStr(pcmdline, SPA_switch, "-now"))) {
u_long32 ar_now = lGetUlong(*ar, AR_type);
if(lGetInt(ep, SPA_argval_lIntT)) {
JOB_TYPE_SET_IMMEDIATE(ar_now);
} else {
JOB_TYPE_CLEAR_IMMEDIATE(ar_now);
}
lSetUlong(*ar, AR_type, ar_now);
lRemoveElem(pcmdline, &ep);
}
/* Remove the script elements. They are not stored in the ar structure */
if ((ep = lGetElemStr(pcmdline, SPA_switch, STR_PSEUDO_SCRIPT))) {
lRemoveElem(pcmdline, &ep);
}
if ((ep = lGetElemStr(pcmdline, SPA_switch, STR_PSEUDO_SCRIPTLEN))) {
lRemoveElem(pcmdline, &ep);
}
if ((ep = lGetElemStr(pcmdline, SPA_switch, STR_PSEUDO_SCRIPTPTR))) {
lRemoveElem(pcmdline, &ep);
}
ep = lFirst(pcmdline);
if(ep) {
const char *option = lGetString(ep,SPA_switch);
/* as jobarg are stored no switch values, need to be filtered */
if(sge_strnullcmp(option, "jobarg") != 0) {
answer_list_add_sprintf(alpp, STATUS_ESEMANTIC, ANSWER_QUALITY_ERROR,
MSG_PARSE_INVALIDOPTIONARGUMENTX_S,
lGetString(ep,SPA_switch));
} else {
answer_list_add_sprintf(alpp, STATUS_ESEMANTIC, ANSWER_QUALITY_ERROR,
MSG_PARSE_INVALIDOPTIONARGUMENT);
}
DRETURN(false);
}
if (lGetUlong(*ar, AR_start_time) == 0 && lGetUlong(*ar, AR_end_time) != 0 && lGetUlong(*ar, AR_duration) != 0) {
lSetUlong(*ar, AR_start_time, lGetUlong(*ar, AR_end_time) - lGetUlong(*ar, AR_duration));
} else if (lGetUlong(*ar, AR_start_time) != 0 && lGetUlong(*ar, AR_end_time) == 0 && lGetUlong(*ar, AR_duration) != 0) {
lSetUlong(*ar, AR_end_time, duration_add_offset(lGetUlong(*ar, AR_start_time), lGetUlong(*ar, AR_duration)));
lSetUlong(*ar, AR_duration, lGetUlong(*ar, AR_end_time) - lGetUlong(*ar, AR_start_time));
} else if (lGetUlong(*ar, AR_start_time) != 0 && lGetUlong(*ar, AR_end_time) != 0 && lGetUlong(*ar, AR_duration) == 0) {
lSetUlong(*ar, AR_duration, lGetUlong(*ar, AR_end_time) - lGetUlong(*ar, AR_start_time));
}
DRETURN(true);
}
static bool
sge_parse_qrstat(sge_gdi_ctx_class_t *ctx, lList **answer_list,
qrstat_env_t *qrstat_env, lList **cmdline)
{
bool ret = true;
DENTER(TOP_LAYER, "sge_parse_qrstat");
qrstat_env->is_summary = true;
while (lGetNumberOfElem(*cmdline)) {
u_long32 value;
/* -help */
if (opt_list_has_X(*cmdline, "-help")) {
sge_usage(QRSTAT, stdout);
DEXIT;
SGE_EXIT((void**)&ctx, 0);
}
/* -u */
while (parse_multi_stringlist(cmdline, "-u", answer_list,
&(qrstat_env->user_list), ST_Type, ST_name)) {
continue;
}
/* -explain */
while (parse_flag(cmdline, "-explain", answer_list, &value)) {
qrstat_filter_add_core_attributes(qrstat_env);
qrstat_filter_add_explain_attributes(qrstat_env);
qrstat_env->is_explain = (value > 0) ? true : false;
continue;
}
/* -xml */
while (parse_flag(cmdline, "-xml", answer_list, &value)) {
qrstat_filter_add_core_attributes(qrstat_env);
qrstat_filter_add_xml_attributes(qrstat_env);
qrstat_env->is_xml = (value > 0) ? true : false;
continue;
}
/* -ar */
while (parse_u_longlist(cmdline, "-ar", answer_list, &(qrstat_env->ar_id_list))) {
qrstat_filter_add_core_attributes(qrstat_env);
qrstat_filter_add_ar_attributes(qrstat_env);
qrstat_filter_add_ar_where(qrstat_env);
qrstat_env->is_summary = false;
continue;
}
if (lGetNumberOfElem(*cmdline)) {
sge_usage(QRSTAT, stdout);
answer_list_add(answer_list, MSG_PARSE_TOOMANYOPTIONS,
STATUS_ESEMANTIC, ANSWER_QUALITY_ERROR);
ret = false;
break;
}
}
if (qrstat_env->is_summary) {
char user[128] = "";
if (sge_uid2user(geteuid(), user, sizeof(user), MAX_NIS_RETRIES)) {
answer_list_add_sprintf(answer_list, STATUS_ESEMANTIC,
ANSWER_QUALITY_CRITICAL,
MSG_SYSTEM_RESOLVEUSER_U, (u_long32)geteuid());
ret = false;
} else {
str_list_transform_user_list(&(qrstat_env->user_list), answer_list, user);
qrstat_filter_add_core_attributes(qrstat_env);
qrstat_filter_add_u_where(qrstat_env);
}
}
DRETURN(ret);
}
int main(int argc, char **argv) {
int ret = 0;
lList *pcmdline = NULL;
lList *answer_list = NULL;
sge_gdi_ctx_class_t *ctx = NULL;
qrstat_env_t qrstat_env;
DENTER_MAIN(TOP_LAYER, "qrsub");
/* Set up the program information name */
sge_setup_sig_handlers(QRSTAT);
log_state_set_log_gui(1);
if (sge_gdi2_setup(&ctx, QRSTAT, MAIN_THREAD, &answer_list) != AE_OK) {
answer_list_output(&answer_list);
goto error_exit;
}
qrstat_filter_init(&qrstat_env);
qrstat_filter_set_ctx(&qrstat_env, ctx);
/*
* stage 1: commandline parsing
*/
{
dstring file = DSTRING_INIT;
const char *user = ctx->get_username(ctx);
const char *cell_root = ctx->get_cell_root(ctx);
/* arguments from SGE_ROOT/common/sge_qrstat file */
get_root_file_path(&file, cell_root, SGE_COMMON_DEF_QRSTAT_FILE);
if (sge_parse_from_file_qrstat(sge_dstring_get_string(&file), &pcmdline, &answer_list) == true) {
/* arguments from $HOME/.sge_qrstat file */
if (get_user_home_file_path(&file, SGE_HOME_DEF_QRSTAT_FILE, user, &answer_list)) {
sge_parse_from_file_qrstat(sge_dstring_get_string(&file), &pcmdline, &answer_list);
}
}
sge_dstring_free(&file);
if (answer_list) {
answer_list_output(&answer_list);
lFreeList(&pcmdline);
sge_prof_cleanup();
SGE_EXIT((void**)&ctx, 1);
}
}
answer_list = cull_parse_cmdline(QRSTAT, argv+1, environ, &pcmdline, FLG_USE_PSEUDOS);
if (answer_list != NULL) {
answer_list_output(&answer_list);
lFreeList(&pcmdline);
goto error_exit;
}
/*
* stage 2: evalutate switches and modify qrstat_env
*/
if (!sge_parse_qrstat(ctx, &answer_list, &qrstat_env, &pcmdline)) {
answer_list_output(&answer_list);
lFreeList(&pcmdline);
goto error_exit;
}
/*
* stage 3: fetch data from master
*/
{
answer_list = ctx->gdi(ctx, SGE_AR_LIST, SGE_GDI_GET, &qrstat_env.ar_list,
qrstat_env.where_AR_Type, qrstat_env.what_AR_Type, false);
if (answer_list_has_error(&answer_list)) {
answer_list_output(&answer_list);
goto error_exit;
}
}
/*
* stage 4: create output in correct format
*/
{
qrstat_report_handler_t *handler = NULL;
if (qrstat_env.is_xml) {
handler = qrstat_create_report_handler_xml(&qrstat_env, &answer_list);
} else {
handler = qrstat_create_report_handler_stdout(&qrstat_env, &answer_list);
}
if (!qrstat_print(&answer_list, handler, &qrstat_env)) {
ret = 1;
}
if (qrstat_env.is_xml) {
qrstat_destroy_report_handler_xml(&handler, &answer_list);
} else {
qrstat_destroy_report_handler_stdout(&handler, &answer_list);
}
}
sge_gdi2_shutdown((void**)&ctx);
sge_prof_cleanup();
DRETURN(ret);
error_exit:
sge_gdi2_shutdown((void**)&ctx);
sge_prof_cleanup();
SGE_EXIT((void**)&ctx, 1);
DRETURN(1);
}
int
main(int argc, char **argv)
{
lList *opts_cmdline = NULL;
lList *opts_defaults = NULL;
lList *opts_scriptfile = NULL;
lList *opts_all = NULL;
lListElem *job = NULL;
lList *alp = NULL;
lListElem *ep;
int exit_status = 0;
int just_verify;
int tmp_ret;
int wait_for_job = 0, is_immediate = 0;
dstring session_key_out = DSTRING_INIT;
dstring diag = DSTRING_INIT;
dstring jobid = DSTRING_INIT;
u_long32 start, end, step;
u_long32 num_tasks;
int count, stat;
char *jobid_string = NULL;
bool has_terse;
drmaa_attr_values_t *jobids = NULL;
u_long32 prog_number = 0;
u_long32 myuid = 0;
const char *sge_root = NULL;
const char *cell_root = NULL;
const char *username = NULL;
const char *qualified_hostname = NULL;
const char *unqualified_hostname = NULL;
const char *mastername = NULL;
DENTER_MAIN(TOP_LAYER, "qsub");
prof_mt_init();
/* Set up the program information name */
sge_setup_sig_handlers(QSUB);
DPRINTF(("Initializing JAPI\n"));
if (japi_init(NULL, NULL, NULL, QSUB, false, NULL, &diag)
!= DRMAA_ERRNO_SUCCESS) {
fprintf(stderr, "\n");
fprintf(stderr, MSG_QSUB_COULDNOTINITIALIZEENV_S,
sge_dstring_get_string(&diag));
fprintf(stderr, "\n");
DEXIT;
SGE_EXIT((void**)&ctx, 1);
}
prog_number = ctx->get_who(ctx);
myuid = ctx->get_uid(ctx);
sge_root = ctx->get_sge_root(ctx);
cell_root = ctx->get_cell_root(ctx);
username = ctx->get_username(ctx);
qualified_hostname = ctx->get_qualified_hostname(ctx);
unqualified_hostname = ctx->get_unqualified_hostname(ctx);
mastername = ctx->get_master(ctx, false);
/*
* read switches from the various defaults files
*/
opt_list_append_opts_from_default_files(prog_number, cell_root, username, &opts_defaults, &alp, environ);
tmp_ret = answer_list_print_err_warn(&alp, NULL, NULL, MSG_WARNING);
if (tmp_ret > 0) {
DEXIT;
SGE_EXIT((void**)&ctx, tmp_ret);
}
/*
* append the commandline switches to the list
*/
opt_list_append_opts_from_qsub_cmdline(prog_number, &opts_cmdline, &alp,
argv + 1, environ);
tmp_ret = answer_list_print_err_warn(&alp, NULL, "qsub: ", MSG_QSUB_WARNING_S);
if (tmp_ret > 0) {
DEXIT;
SGE_EXIT((void**)&ctx, tmp_ret);
}
/*
* show usage if -help was in commandline
*/
if (opt_list_has_X(opts_cmdline, "-help")) {
sge_usage(QSUB, stdout);
DEXIT;
SGE_EXIT((void**)&ctx, 0);
}
/*
* We will only read commandline options from scripfile if the script
* itself should not be handled as binary
*/
if (opt_list_is_X_true(opts_cmdline, "-b") ||
(!opt_list_has_X(opts_cmdline, "-b") &&
opt_list_is_X_true(opts_defaults, "-b"))) {
DPRINTF(("Skipping options from script due to -b option\n"));
} else {
opt_list_append_opts_from_script(prog_number,
&opts_scriptfile, &alp,
opts_cmdline, environ);
tmp_ret = answer_list_print_err_warn(&alp, NULL, MSG_QSUB_COULDNOTREADSCRIPT_S,
MSG_WARNING);
if (tmp_ret > 0) {
DEXIT;
SGE_EXIT((void**)&ctx, tmp_ret);
}
}
/*
* Merge all commandline options and interprete them
*/
opt_list_merge_command_lines(&opts_all, &opts_defaults,
&opts_scriptfile, &opts_cmdline);
/*
* Check if -terse is requested
*/
has_terse = opt_list_has_X(opts_all, "-terse");
/* If "-sync y" is set, wait for the job to end. */
/* Remove all -sync switches since cull_parse_job_parameter()
* doesn't know what to do with them. */
while ((ep = lGetElemStr(opts_all, SPA_switch, "-sync"))) {
if (lGetInt(ep, SPA_argval_lIntT) == TRUE) {
wait_for_job = 1;
}
lRemoveElem(opts_all, &ep);
}
if (wait_for_job) {
DPRINTF(("Wait for job end\n"));
}
alp = cull_parse_job_parameter(myuid, username, cell_root, unqualified_hostname,
qualified_hostname, opts_all, &job);
tmp_ret = answer_list_print_err_warn(&alp, NULL, "qsub: ", MSG_WARNING);
if (tmp_ret > 0) {
DEXIT;
SGE_EXIT((void**)&ctx, tmp_ret);
}
if (set_sec_cred(sge_root, mastername, job, &alp) != 0) {
answer_list_output(&alp);
DEXIT;
SGE_EXIT((void**)&ctx, 1);
}
/* Check if job is immediate */
is_immediate = (int)JOB_TYPE_IS_IMMEDIATE(lGetUlong(job, JB_type));
DPRINTF(("Job is%s immediate\n", is_immediate ? "" : " not"));
DPRINTF(("Everything ok\n"));
if (lGetUlong(job, JB_verify)) {
cull_show_job(job, 0, false);
DEXIT;
SGE_EXIT((void**)&ctx, 0);
}
if (is_immediate || wait_for_job) {
pthread_t sigt;
qsub_setup_sig_handlers();
if (pthread_create(&sigt, NULL, sig_thread, (void *)NULL) != 0) {
fprintf(stderr, "\n");
fprintf(stderr, MSG_QSUB_COULDNOTINITIALIZEENV_S,
" error preparing signal handling thread");
fprintf(stderr, "\n");
exit_status = 1;
goto Error;
}
if (japi_enable_job_wait(username, unqualified_hostname, NULL, &session_key_out, error_handler, &diag) ==
DRMAA_ERRNO_DRM_COMMUNICATION_FAILURE) {
const char *msg = sge_dstring_get_string(&diag);
fprintf(stderr, "\n");
fprintf(stderr, MSG_QSUB_COULDNOTINITIALIZEENV_S,
msg?msg:" error starting event client thread");
fprintf(stderr, "\n");
exit_status = 1;
goto Error;
}
}
job_get_submit_task_ids(job, &start, &end, &step);
num_tasks = (end - start) / step + 1;
if (num_tasks > 1) {
int error = japi_run_bulk_jobs(&jobids, &job, start, end, step, &diag);
if (error != DRMAA_ERRNO_SUCCESS) {
/* No active session here means that japi_enable_job_wait() was
* interrupted by the signal handler, in which case we just break out
* quietly. */
if (error != DRMAA_ERRNO_NO_ACTIVE_SESSION) {
fprintf(stderr, MSG_QSUB_COULDNOTRUNJOB_S,
sge_dstring_get_string(&diag));
fprintf(stderr, "\n");
}
/* BUGFIX: Issuezilla #1013
* To quickly fix this issue, I'm mapping the JAPI/DRMAA error code
* back into a GDI error code. This is the easy solution. The
* correct solution would be to address issue #859, presumably by
* having JAPI reuse the GDI error codes instead of the JAPI error
* codes. */
if (error == DRMAA_ERRNO_TRY_LATER) {
exit_status = STATUS_NOTOK_DOAGAIN;
}
else {
exit_status = 1;
}
goto Error;
}
DPRINTF(("job id is: %ld\n", jobids->it.ji.jobid));
jobid_string = get_bulk_jobid_string((long)jobids->it.ji.jobid, start, end, step);
}
else if (num_tasks == 1) {
int error = japi_run_job(&jobid, &job, &diag);
if (error != DRMAA_ERRNO_SUCCESS) {
if (error != DRMAA_ERRNO_NO_ACTIVE_SESSION) {
fprintf(stderr, MSG_QSUB_COULDNOTRUNJOB_S,
sge_dstring_get_string(&diag));
fprintf(stderr, "\n");
}
/* BUGFIX: Issuezilla #1013
* To quickly fix this issue, I'm mapping the JAPI/DRMAA error code
* back into a GDI error code. This is the easy solution. The
* correct solution would be to address issue #859, presumably by
* having JAPI reuse the GDI error codes instead of the DRMAA error
* codes. */
if (error == DRMAA_ERRNO_TRY_LATER) {
exit_status = STATUS_NOTOK_DOAGAIN;
}
else {
exit_status = 1;
}
goto Error;
}
jobid_string = strdup(sge_dstring_get_string(&jobid));
DPRINTF(("job id is: %s\n", jobid_string));
sge_dstring_free(&jobid);
}
else {
fprintf(stderr, MSG_QSUB_COULDNOTRUNJOB_S, "invalid task structure");
fprintf(stderr, "\n");
exit_status = 1;
goto Error;
}
/* only success message is printed to stdout */
just_verify = (lGetUlong(job, JB_verify_suitable_queues)==JUST_VERIFY ||
lGetUlong(job, JB_verify_suitable_queues)==POKE_VERIFY);
DPRINTF(("Just verifying job\n"));
if (!just_verify) {
const char *output = sge_dstring_get_string(&diag);
/* print the tersed output */
if (has_terse) {
printf("%s", jobid_string);
} else if (output != NULL) {
printf("%s", output);
} else {
printf(MSG_QSUB_YOURJOBHASBEENSUBMITTED_SS, jobid_string, lGetString(job, JB_job_name));
}
printf("\n");
} else {
printf(MSG_JOB_VERIFYFOUNDQ);
printf("\n");
}
if ((wait_for_job || is_immediate) && !just_verify) {
int event;
if (is_immediate) {
fprintf(stderr, "%s\n", MSG_QSUB_WAITINGFORIMMEDIATEJOBTOBESCHEDULED);
/* We only need to wait for the first task to be scheduled to be able
* to say that the job is running. */
tmp_ret = japi_wait(DRMAA_JOB_IDS_SESSION_ANY, &jobid, &stat,
DRMAA_TIMEOUT_WAIT_FOREVER, JAPI_JOB_START, &event,
NULL, &diag);
if ((tmp_ret == DRMAA_ERRNO_SUCCESS) && (event == JAPI_JOB_START)) {
fprintf(stderr, "\n");
fprintf(stderr, MSG_QSUB_YOURIMMEDIATEJOBXHASBEENSUCCESSFULLYSCHEDULED_S,
jobid_string);
fprintf(stderr, "\n");
}
/* A job finish event here means that the job was rejected. */
else if ((tmp_ret == DRMAA_ERRNO_SUCCESS) &&
(event == JAPI_JOB_FINISH)) {
fprintf(stderr, "\n%s\n", MSG_QSUB_YOURQSUBREQUESTCOULDNOTBESCHEDULEDDTRYLATER);
exit_status = 1;
goto Error;
}
else {
/* Since we told japi_wait to wait forever, we know that if it gets
* a timeout, it's because it's been interrupted to exit, in which
* case we don't complain. Same for no active session. */
if ((tmp_ret != DRMAA_ERRNO_EXIT_TIMEOUT) &&
(tmp_ret != DRMAA_ERRNO_NO_ACTIVE_SESSION)) {
fprintf(stderr, "\n");
fprintf(stderr, MSG_QSUB_COULDNOTWAITFORJOB_S,
sge_dstring_get_string(&diag));
fprintf(stderr, "\n");
}
exit_status = 1;
goto Error;
}
}
if (wait_for_job) {
/* Rather than using japi_synchronize on ALL for bulk jobs, we use
* japi_wait on ANY num_tasks times because with synchronize, we would
* have to wait for all the tasks to finish before we know if any
* finished. */
for (count = 0; count < num_tasks; count++) {
/* Since there's only one running job in the session, we can just
* wait for ANY. */
if ((tmp_ret = japi_wait(DRMAA_JOB_IDS_SESSION_ANY, &jobid, &stat,
DRMAA_TIMEOUT_WAIT_FOREVER, JAPI_JOB_FINISH, &event,
NULL, &diag)) != DRMAA_ERRNO_SUCCESS) {
if ((tmp_ret != DRMAA_ERRNO_EXIT_TIMEOUT) &&
(tmp_ret != DRMAA_ERRNO_NO_ACTIVE_SESSION)) {
fprintf(stderr, "\n");
fprintf(stderr, MSG_QSUB_COULDNOTWAITFORJOB_S, sge_dstring_get_string(&diag));
fprintf(stderr, "\n");
}
exit_status = 1;
goto Error;
}
/* report how job finished */
/* If the job is an array job, use the first non-zero exit code as
* the exit code for qsub. */
if (exit_status == 0) {
exit_status = report_exit_status(stat,
sge_dstring_get_string(&jobid));
}
/* If we've already found a non-zero exit code, just print the exit
* info for the task. */
else {
report_exit_status(stat, sge_dstring_get_string(&jobid));
}
}
}
}
Error:
FREE(jobid_string);
lFreeList(&alp);
lFreeList(&opts_all);
if ((tmp_ret = japi_exit(JAPI_EXIT_NO_FLAG, &diag)) != DRMAA_ERRNO_SUCCESS) {
if (tmp_ret != DRMAA_ERRNO_NO_ACTIVE_SESSION) {
fprintf(stderr, "\n");
fprintf(stderr, MSG_QSUB_COULDNOTFINALIZEENV_S, sge_dstring_get_string(&diag));
fprintf(stderr, "\n");
}
else {
struct timespec ts;
/* We know that if we get a DRMAA_ERRNO_NO_ACTIVE_SESSION here, it's
* because the signal handler thread called japi_exit(). We know this
* because if the call to japi_init() fails, we just exit directly.
* If the call to japi_init() succeeds, then we have an active session,
* so coming here because of an error would not result in the
* DRMAA_ERRNO_NO_ACTIVE_SESSION error. */
DPRINTF(("Sleeping for 15 seconds to wait for the exit to finish.\n"));
sge_relative_timespec(15, &ts);
sge_mutex_lock("qsub_exit_mutex", SGE_FUNC, __LINE__, &exit_mutex);
while (!exited) {
if (pthread_cond_timedwait(&exit_cv, &exit_mutex, &ts) == ETIMEDOUT) {
DPRINTF(("Exit has not finished after 15 seconds. Exiting.\n"));
break;
}
}
sge_mutex_unlock("qsub_exit_mutex", SGE_FUNC, __LINE__, &exit_mutex);
}
}
sge_prof_cleanup();
/* This is an exit() instead of an SGE_EXIT() because when the qmaster is
* supended, SGE_EXIT() hangs. */
exit(exit_status);
DEXIT;
return exit_status;
}
int main(int argc, char **argv) {
lList *pcmdline = NULL;
lList *alp = NULL;
sge_gdi_ctx_class_t *ctx = NULL;
lList *ar_lp = NULL;
lListElem *ar = NULL;
DENTER_MAIN(TOP_LAYER, "qrsub");
/* Set up the program information name */
sge_setup_sig_handlers(QRSUB);
log_state_set_log_gui(1);
if (sge_gdi2_setup(&ctx, QRSUB, MAIN_THREAD, &alp) != AE_OK) {
answer_list_output(&alp);
goto error_exit;
}
/*
** stage 1 of commandline parsing
*/
{
dstring file = DSTRING_INIT;
const char *user = ctx->get_username(ctx);
const char *cell_root = ctx->get_cell_root(ctx);
/* arguments from SGE_ROOT/common/sge_ar_request file */
get_root_file_path(&file, cell_root, SGE_COMMON_DEF_AR_REQ_FILE);
if ((alp = parse_script_file(QRSUB, sge_dstring_get_string(&file), "", &pcmdline, environ,
FLG_HIGHER_PRIOR | FLG_IGN_NO_FILE)) == NULL) {
/* arguments from $HOME/.sge_ar_request file */
if (get_user_home_file_path(&file, SGE_HOME_DEF_AR_REQ_FILE, user, &alp)) {
lFreeList(&alp);
alp = parse_script_file(QRSUB, sge_dstring_get_string(&file), "", &pcmdline, environ,
FLG_HIGHER_PRIOR | FLG_IGN_NO_FILE);
}
}
sge_dstring_free(&file);
if (alp) {
answer_list_output(&alp);
lFreeList(&pcmdline);
goto error_exit;
}
}
alp = cull_parse_cmdline(QRSUB, argv+1, environ, &pcmdline, FLG_USE_PSEUDOS);
if (answer_list_print_err_warn(&alp, NULL, "qrsub: ", MSG_WARNING) > 0) {
lFreeList(&pcmdline);
goto error_exit;
}
if (!pcmdline) {
/* no command line option is present: print help to stderr */
sge_usage(QRSUB, stderr);
fprintf(stderr, "%s\n", MSG_PARSE_NOOPTIONARGUMENT);
goto error_exit;
}
/*
** stage 2 of command line parsing
*/
ar = lCreateElem(AR_Type);
if (!sge_parse_qrsub(ctx, pcmdline, &alp, &ar)) {
answer_list_output(&alp);
lFreeList(&pcmdline);
goto error_exit;
}
ar_lp = lCreateList(NULL, AR_Type);
lAppendElem(ar_lp, ar);
alp = ctx->gdi(ctx, SGE_AR_LIST, SGE_GDI_ADD | SGE_GDI_RETURN_NEW_VERSION, &ar_lp, NULL, NULL);
lFreeList(&ar_lp);
answer_list_on_error_print_or_exit(&alp, stdout);
if (answer_list_has_error(&alp)) {
sge_gdi2_shutdown((void**)&ctx);
sge_prof_cleanup();
if (answer_list_has_status(&alp, STATUS_NOTOK_DOAGAIN)) {
DRETURN(25);
} else {
DRETURN(1);
}
}
sge_gdi2_shutdown((void**)&ctx);
sge_prof_cleanup();
DRETURN(0);
error_exit:
sge_gdi2_shutdown((void**)&ctx);
sge_prof_cleanup();
SGE_EXIT((void**)&ctx, 1);
DRETURN(1);
}
/*-------------------------------------------------------------------*/
void sge_setup_sge_execd(sge_gdi_ctx_class_t *ctx, const char* tmp_err_file_name)
{
char err_str[MAX_STRING_SIZE];
int allowed_get_conf_errors = 5;
char* spool_dir = NULL;
const char *unqualified_hostname = ctx->get_unqualified_hostname(ctx);
const char *admin_user = ctx->get_admin_user(ctx);
DENTER(TOP_LAYER, "sge_setup_sge_execd");
/* TODO: is this the right place to switch the user ?
ports below 1024 ok */
/*
** switch to admin user
*/
if (sge_set_admin_username(admin_user, err_str)) {
CRITICAL((SGE_EVENT, SFNMAX, err_str));
/* TODO: remove */
SGE_EXIT(NULL, 1);
}
if (sge_switch2admin_user()) {
CRITICAL((SGE_EVENT, SFNMAX, MSG_ERROR_CANTSWITCHTOADMINUSER));
/* TODO: remove */
SGE_EXIT(NULL, 1);
}
while (gdi2_wait_for_conf(ctx, &Execd_Config_List)) {
if (allowed_get_conf_errors-- <= 0) {
CRITICAL((SGE_EVENT, SFNMAX, MSG_EXECD_CANT_GET_CONFIGURATION_EXIT));
/* TODO: remove */
SGE_EXIT(NULL, 1);
}
sleep(1);
ctx->get_master(ctx, true);
}
sge_show_conf();
/* get aliased hostname */
/* TODO: is this call needed ? */
ctx->reresolve_qualified_hostname(ctx);
spool_dir = mconf_get_execd_spool_dir();
DPRINTF(("chdir(\"/\")----------------------------\n"));
sge_chdir_exit("/",1);
DPRINTF(("Making directories----------------------------\n"));
sge_mkdir(spool_dir, 0755, 1, 0);
DPRINTF(("chdir(\"%s\")----------------------------\n", spool_dir));
sge_chdir_exit(spool_dir,1);
sge_mkdir(unqualified_hostname, 0755, 1, 0);
DPRINTF(("chdir(\"%s\",me.unqualified_hostname)--------------------------\n",
unqualified_hostname));
sge_chdir_exit(unqualified_hostname, 1);
/* having passed the previous statement we may
log messages into the ERR_FILE */
if ( tmp_err_file_name != NULL) {
sge_copy_append((char*)tmp_err_file_name, ERR_FILE, SGE_MODE_APPEND);
}
sge_switch2start_user();
if ( tmp_err_file_name != NULL) {
unlink(tmp_err_file_name);
}
sge_switch2admin_user();
log_state_set_log_as_admin_user(1);
sprintf(execd_messages_file, "%s/%s/%s", spool_dir,
unqualified_hostname, ERR_FILE);
log_state_set_log_file(execd_messages_file);
sprintf(execd_spool_dir, "%s/%s", spool_dir,
unqualified_hostname);
DPRINTF(("Making directories----------------------------\n"));
sge_mkdir(EXEC_DIR, 0775, 1, 0);
sge_mkdir(JOB_DIR, 0775, 1, 0);
sge_mkdir(ACTIVE_DIR, 0775, 1, 0);
#if defined(PLPA_LINUX) || defined(SOLARIS86) || defined(SOLARISAMD64)
/* initialize processor topology */
if (initialize_topology() != true) {
DPRINTF(("Couldn't initizalize topology-----------------------\n"));
}
#endif
sge_free(&spool_dir);
DRETURN_VOID;
}
/****** qmaster/sge_qmaster_main/main() ****************************************
* NAME
* main() -- qmaster entry point
*
* SYNOPSIS
* int main(int argc, char* argv[])
*
* FUNCTION
* Qmaster entry point.
*
* NOTE: The main thread must block all signals before any additional thread
* is created. Failure to do so will ruin signal handling!
*
* INPUTS
* int argc - number of commandline arguments
* char* argv[] - commandline arguments
*
* RESULT
* 0 - success
*
* NOTES
* We check whether 'SGE_ROOT' is set before we daemonize. Once qmaster is
* a daemon, we are no longer connected to a terminal and hence can not
* output an error message to stdout or stderr.
*
* We need to inovke 'prepare_enroll()' *before* the user id is switched via
* 'become_admin_user()'. This is because qmaster must be able to bind a so
* called reserved port (requires root privileges) if configured to do so.
*
*******************************************************************************/
int main(int argc, char* argv[])
{
int max_enroll_tries;
int ret_val;
int file_descriptor_settings_result = 0;
bool has_daemonized = false;
sge_gdi_ctx_class_t *ctx = NULL;
u_long32 start_time = sge_get_gmt();
monitoring_t monitor;
DENTER_MAIN(TOP_LAYER, "qmaster");
sge_monitor_init(&monitor, "MAIN", NONE_EXT, MT_WARNING, MT_ERROR);
prof_mt_init();
sge_get_root_dir(true, NULL, 0, true);
#ifdef __SGE_COMPILE_WITH_GETTEXT__
sge_init_language_func((gettext_func_type)gettext, (setlocale_func_type)setlocale, (bindtextdomain_func_type)bindtextdomain, (textdomain_func_type)textdomain);
sge_init_language(NULL,NULL);
#endif
/*
* qmaster doesn't support any commandline anymore,
* but we should show version string and -help option
*/
if (argc != 1) {
sigset_t sig_set;
sigfillset(&sig_set);
pthread_sigmask(SIG_SETMASK, &sig_set, NULL);
sge_qmaster_thread_init(&ctx, QMASTER, MAIN_THREAD, true);
sge_process_qmaster_cmdline(argv);
SGE_EXIT((void**)&ctx, 1);
}
/*
* daemonize qmaster
* set file descriptor limits
* and initialize libraries to be used in multi threaded environment
* also take care that finished child processed of this process become
* zombie jobs
*/
has_daemonized = sge_daemonize_qmaster();
file_descriptor_settings_result = set_file_descriptor_limit();
#if !defined(INTERIX) && !defined(CYGWIN)
init_sig_action_and_mask();
#endif
/* init qmaster threads without becomming admin user */
sge_qmaster_thread_init(&ctx, QMASTER, MAIN_THREAD, false);
ctx->set_daemonized(ctx, has_daemonized);
/* this must be done as root user to be able to bind ports < 1024 */
max_enroll_tries = 30;
while (cl_com_get_handle(prognames[QMASTER],1) == NULL) {
ctx->prepare_enroll(ctx);
max_enroll_tries--;
if (max_enroll_tries <= 0) {
/* exit after 30 seconds */
CRITICAL((SGE_EVENT, MSG_QMASTER_COMMUNICATION_ERRORS ));
SGE_EXIT((void**)&ctx, 1);
}
if (cl_com_get_handle(prognames[QMASTER],1) == NULL) {
/* sleep when prepare_enroll() failed */
sleep(1);
}
}
/*
* now the commlib up and running. Set qmaster application status function
* (commlib callback function for qping status information response
* messages (SIRM))
*/
ret_val = cl_com_set_status_func(sge_qmaster_application_status);
if (ret_val != CL_RETVAL_OK) {
ERROR((SGE_EVENT, cl_get_error_text(ret_val)));
}
/*
* now we become admin user change into the correct root directory set the
* the target for logging messages
*/
sge_become_admin_user(ctx->get_admin_user(ctx));
sge_chdir_exit(ctx->get_qmaster_spool_dir(ctx), 1);
log_state_set_log_file(ERR_FILE);
ctx->set_exit_func(ctx, sge_exit_func);
#if defined(SOLARIS)
/* Init shared SMF libs if necessary */
if (sge_smf_used() == 1 && sge_smf_init_libs() != 0) {
SGE_EXIT((void**)&ctx, 1);
}
#endif
/*
* We do increment the heartbeat manually here. This is the 'startup heartbeat'.
* The first time the hearbeat will be incremented through the heartbeat event
* handler is after about HEARTBEAT_INTERVAL seconds. The hardbeat event handler
* is setup during the initialisazion of the timer thread.
*/
inc_qmaster_heartbeat(QMASTER_HEARTBEAT_FILE, HEARTBEAT_INTERVAL, NULL);
/*
* Event master module has to be initialized already here because
* sge_setup_qmaster() might already access it although event delivery
* thread is not running.
*
* Corresponding shutdown is done in sge_event_master_terminate();
*
* EB: In my opinion the init function should called in
* sge_event_master_initialize(). Is it possible to move that call?
*/
sge_event_master_init();
sge_setup_qmaster(ctx, argv);
#ifndef USE_POLL
if (file_descriptor_settings_result == 1) {
WARNING((SGE_EVENT, MSG_QMASTER_FD_SETSIZE_LARGER_THAN_LIMIT_U, sge_u32c(FD_SETSIZE)));
WARNING((SGE_EVENT, MSG_QMASTER_FD_SETSIZE_COMPILE_MESSAGE1_U, sge_u32c(FD_SETSIZE - 20)));
WARNING((SGE_EVENT, MSG_QMASTER_FD_SETSIZE_COMPILE_MESSAGE2));
WARNING((SGE_EVENT, MSG_QMASTER_FD_SETSIZE_COMPILE_MESSAGE3));
}
#endif
/*
* Setup all threads and initialize corresponding modules.
* Order is important!
*/
sge_signaler_initialize(ctx);
sge_event_master_initialize(ctx);
sge_timer_initialize(ctx, &monitor);
sge_worker_initialize(ctx);
#if 0
sge_test_initialize(ctx);
#endif
sge_listener_initialize(ctx);
sge_scheduler_initialize(ctx, NULL);
#ifndef NO_JNI
sge_jvm_initialize(ctx, NULL);
#endif
INFO((SGE_EVENT, "qmaster startup took "sge_u32" seconds", sge_get_gmt() - start_time));
/*
* Block till signal from signal thread arrives us
*/
sge_thread_wait_for_signal();
/*
* Shutdown all threads and shutdown corresponding modules.
* Order is important!
*/
#ifndef NO_JNI
sge_jvm_terminate(ctx, NULL);
#endif
sge_scheduler_terminate(ctx, NULL);
sge_listener_terminate();
#if 0
sge_test_terminate(ctx);
#endif
sge_worker_terminate(ctx);
sge_timer_terminate();
sge_event_master_terminate();
sge_signaler_terminate();
/*
* Remaining shutdown operations
*/
sge_clean_lists();
sge_monitor_free(&monitor);
sge_shutdown((void**)&ctx, sge_qmaster_get_exit_state());
sge_prof_cleanup();
DEXIT;
return 0;
} /* main() */
/*----------------------------------------------------------------------------*/
int
main(int argc, char **argv)
{
int heartbeat = 0;
int last_heartbeat = 0;
int latest_heartbeat = 0;
int ret = 0;
int delay = 0;
time_t now, last;
/* const char *cp; */
char err_str[MAX_STRING_SIZE];
char shadowd_pidfile[SGE_PATH_MAX];
dstring ds;
char buffer[256];
pid_t shadowd_pid;
#if 1
static int check_interval = CHECK_INTERVAL;
static int get_active_interval = GET_ACTIVE_INTERVAL;
static int delay_time = DELAY_TIME;
static int sge_test_heartbeat = 0;
char binpath[SGE_PATH_MAX];
char oldqmaster[SGE_PATH_MAX];
char shadow_err_file[SGE_PATH_MAX];
char qmaster_out_file[SGE_PATH_MAX];
#endif
lList *alp = NULL;
sge_gdi_ctx_class_t *ctx = NULL;
DENTER_MAIN(TOP_LAYER, "sge_shadowd");
sge_dstring_init(&ds, buffer, sizeof(buffer));
/* initialize recovery control variables */
{
char *s;
int val;
if ((s=getenv("SGE_CHECK_INTERVAL")) &&
sscanf(s, "%d", &val) == 1)
check_interval = val;
if ((s=getenv("SGE_GET_ACTIVE_INTERVAL")) &&
sscanf(s, "%d", &val) == 1)
get_active_interval = val;
if ((s=getenv("SGE_DELAY_TIME")) &&
sscanf(s, "%d", &val) == 1)
delay_time = val;
if ((s=getenv("SGE_TEST_HEARTBEAT_TIMEOUT")) &&
sscanf(s, "%d", &val) == 1)
sge_test_heartbeat = val;
}
/* This needs a better solution */
umask(022);
#ifdef __SGE_COMPILE_WITH_GETTEXT__
/* init language output for gettext() , it will use the right language */
sge_init_language_func((gettext_func_type) gettext,
(setlocale_func_type) setlocale,
(bindtextdomain_func_type) bindtextdomain,
(textdomain_func_type) textdomain);
sge_init_language(NULL,NULL);
#endif /* __SGE_COMPILE_WITH_GETTEXT__ */
log_state_set_log_file(TMP_ERR_FILE_SHADOWD);
if (sge_setup2(&ctx, SHADOWD, MAIN_THREAD, &alp, false) != AE_OK) {
answer_list_output(&alp);
SGE_EXIT((void**)&ctx, 1);
}
/* AA: TODO: change this */
ctx->set_exit_func(ctx, shadowd_exit_func);
sge_setup_sig_handlers(SHADOWD);
#if defined(SOLARIS)
/* Init shared SMF libs if necessary */
if (sge_smf_used() == 1 && sge_smf_init_libs() != 0) {
SGE_EXIT((void**)&ctx, 1);
}
#endif
if (ctx->get_qmaster_spool_dir(ctx) != NULL) {
char *shadowd_name = SGE_SHADOWD;
/* is there a running shadowd on this host (with unqualified name) */
sprintf(shadowd_pidfile, "%s/"SHADOWD_PID_FILE, ctx->get_qmaster_spool_dir(ctx),
ctx->get_unqualified_hostname(ctx));
DPRINTF(("pidfilename: %s\n", shadowd_pidfile));
if ((shadowd_pid = sge_readpid(shadowd_pidfile))) {
DPRINTF(("shadowd_pid: "sge_U32CFormat"\n", sge_u32c(shadowd_pid)));
if (!sge_checkprog(shadowd_pid, shadowd_name, PSCMD)) {
CRITICAL((SGE_EVENT, MSG_SHADOWD_FOUNDRUNNINGSHADOWDWITHPIDXNOTSTARTING_I, (int) shadowd_pid));
SGE_EXIT((void**)&ctx, 1);
}
}
ctx->prepare_enroll(ctx);
/* is there a running shadowd on this host (with aliased name) */
sprintf(shadowd_pidfile, "%s/"SHADOWD_PID_FILE, ctx->get_qmaster_spool_dir(ctx),
ctx->get_qualified_hostname(ctx));
DPRINTF(("pidfilename: %s\n", shadowd_pidfile));
if ((shadowd_pid = sge_readpid(shadowd_pidfile))) {
DPRINTF(("shadowd_pid: "sge_U32CFormat"\n", sge_u32c(shadowd_pid)));
if (!sge_checkprog(shadowd_pid, shadowd_name, PSCMD)) {
CRITICAL((SGE_EVENT, MSG_SHADOWD_FOUNDRUNNINGSHADOWDWITHPIDXNOTSTARTING_I, (int) shadowd_pid));
SGE_EXIT((void**)&ctx, 1);
}
}
} else {
ctx->prepare_enroll(ctx);
}
if (parse_cmdline_shadowd(argc, argv) == 1) {
SGE_EXIT((void**)&ctx, 0);
}
if (ctx->get_qmaster_spool_dir(ctx) == NULL) {
CRITICAL((SGE_EVENT, MSG_SHADOWD_CANTREADQMASTERSPOOLDIRFROMX_S, ctx->get_bootstrap_file(ctx)));
SGE_EXIT((void**)&ctx, 1);
}
if (chdir(ctx->get_qmaster_spool_dir(ctx))) {
CRITICAL((SGE_EVENT, MSG_SHADOWD_CANTCHANGETOQMASTERSPOOLDIRX_S, ctx->get_qmaster_spool_dir(ctx)));
SGE_EXIT((void**)&ctx, 1);
}
if (sge_set_admin_username(ctx->get_admin_user(ctx), err_str)) {
CRITICAL((SGE_EVENT, SFNMAX, err_str));
SGE_EXIT((void**)&ctx, 1);
}
if (sge_switch2admin_user()) {
CRITICAL((SGE_EVENT, SFNMAX, MSG_SHADOWD_CANTSWITCHTOADMIN_USER));
SGE_EXIT((void**)&ctx, 1);
}
sprintf(shadow_err_file, "messages_shadowd.%s", ctx->get_unqualified_hostname(ctx));
sprintf(qmaster_out_file, "messages_qmaster.%s", ctx->get_unqualified_hostname(ctx));
sge_copy_append(TMP_ERR_FILE_SHADOWD, shadow_err_file, SGE_MODE_APPEND);
unlink(TMP_ERR_FILE_SHADOWD);
log_state_set_log_as_admin_user(1);
log_state_set_log_file(shadow_err_file);
{
int* tmp_fd_array = NULL;
unsigned long tmp_fd_count = 0;
if (cl_com_set_handle_fds(cl_com_get_handle(prognames[SHADOWD] ,0), &tmp_fd_array, &tmp_fd_count) == CL_RETVAL_OK) {
sge_daemonize(tmp_fd_array, tmp_fd_count, ctx);
if (tmp_fd_array != NULL) {
sge_free(&tmp_fd_array);
}
} else {
sge_daemonize(NULL, 0, ctx);
}
}
/* shadowd pid file will contain aliased name */
sge_write_pid(shadowd_pidfile);
starting_up();
sge_setup_sig_handlers(SHADOWD);
last_heartbeat = get_qmaster_heartbeat(QMASTER_HEARTBEAT_FILE, 30);
last = (time_t) sge_get_gmt(); /* set time of last check time */
delay = 0;
while (!shut_me_down) {
sleep(check_interval);
/* get current heartbeat file content */
heartbeat = get_qmaster_heartbeat(QMASTER_HEARTBEAT_FILE, 30);
now = (time_t) sge_get_gmt();
/* Only check when we could read the heartbeat file at least two times
* (last_heartbeat and heartbeat) without error
*/
if (last_heartbeat > 0 && heartbeat > 0) {
/*
* OK we have to heartbeat entries to check. Check times ...
* now = current time
* last = last check time
*/
if ( (now - last) >= (get_active_interval + delay) ) {
delay = 0;
if (last_heartbeat == heartbeat) {
DPRINTF(("heartbeat not changed since seconds: "sge_U32CFormat"\n", sge_u32c(now - last)));
delay = delay_time; /* set delay time */
/*
* check if we are a possible new qmaster host (lock file of qmaster active, etc.)
*/
ret = check_if_valid_shadow(binpath, oldqmaster,
ctx->get_act_qmaster_file(ctx),
ctx->get_shadow_master_file(ctx),
ctx->get_qualified_hostname(ctx),
ctx->get_binary_path(ctx));
if (ret == 0) {
/* we can start a qmaster on this host */
if (qmaster_lock(QMASTER_LOCK_FILE)) {
ERROR((SGE_EVENT, SFNMAX, MSG_SHADOWD_FAILEDTOLOCKQMASTERSOMBODYWASFASTER));
} else {
int out, err;
/* still the old qmaster name in act_qmaster file and still the old heartbeat */
latest_heartbeat = get_qmaster_heartbeat( QMASTER_HEARTBEAT_FILE, 30);
/* TODO: what do we when there is a timeout ??? */
DPRINTF(("old qmaster name in act_qmaster and old heartbeat\n"));
if (!compare_qmaster_names(ctx->get_act_qmaster_file(ctx), oldqmaster) &&
!shadowd_is_old_master_enrolled(sge_test_heartbeat, sge_get_qmaster_port(NULL), oldqmaster) &&
(latest_heartbeat == heartbeat)) {
char qmaster_name[256];
strcpy(qmaster_name, SGE_PREFIX);
strcat(qmaster_name, prognames[QMASTER]);
DPRINTF(("qmaster_name: "SFN"\n", qmaster_name));
/*
* open logfile as admin user for initial qmaster/schedd
* startup messages
*/
out = SGE_OPEN3(qmaster_out_file, O_CREAT|O_WRONLY|O_APPEND,
0644);
err = out;
if (out == -1) {
/*
* First priority is the master restart
* => ignore this error
*/
out = 1;
err = 2;
}
sge_switch2start_user();
ret = startprog(out, err, NULL, binpath, qmaster_name, NULL);
sge_switch2admin_user();
if (ret) {
ERROR((SGE_EVENT, SFNMAX, MSG_SHADOWD_CANTSTARTQMASTER));
}
close(out);
} else {
qmaster_unlock(QMASTER_LOCK_FILE);
}
}
} else {
if (ret == -1) {
/* just log the more important failures */
WARNING((SGE_EVENT, MSG_SHADOWD_DELAYINGSHADOWFUNCFORXSECONDS_U, sge_u32c(delay) ));
}
}
}
/* Begin a new interval, set timers and hearbeat to current values */
last = now;
last_heartbeat = heartbeat;
}
} else {
if (last_heartbeat < 0 || heartbeat < 0) {
/* There was an error reading heartbeat or last_heartbeat */
DPRINTF(("can't read heartbeat file. last_heartbeat="sge_U32CFormat", heartbeat="sge_U32CFormat"\n",
sge_u32c(last_heartbeat), sge_u32c(heartbeat)));
} else {
DPRINTF(("have to read the heartbeat file twice to check time differences\n"));
}
}
}
sge_shutdown((void**)&ctx, 0);
DRETURN(EXIT_SUCCESS);
}
int main(int argc, char **argv) {
int ret = STATUS_OK;
lList *alp = NULL;
lList *request_list = NULL;
lList *cmdline = NULL;
lListElem *aep;
int all_jobs = 0;
int all_users = 0;
u_long32 gdi_cmd = SGE_GDI_MOD;
int tmp_ret;
int me_who;
sge_gdi_ctx_class_t *ctx = NULL;
DENTER_MAIN(TOP_LAYER, "qalter");
prof_mt_init();
/*
** get command name: qalter or qresub
*/
if (!strcmp(sge_basename(argv[0], '/'), "qresub")) {
DPRINTF(("QRESUB\n"));
me_who = QRESUB;
} else if (!strcmp(sge_basename(argv[0], '/'), "qhold")) {
DPRINTF(("QHOLD\n"));
me_who = QHOLD;
} else if (!strcmp(sge_basename(argv[0], '/'), "qrls")) {
DPRINTF(("QRLS\n"));
me_who = QRLS;
} else {
DPRINTF(("QALTER\n"));
me_who = QALTER;
}
log_state_set_log_gui(1);
sge_setup_sig_handlers(me_who);
if (sge_gdi2_setup(&ctx, me_who, MAIN_THREAD, &alp) != AE_OK) {
answer_list_output(&alp);
SGE_EXIT((void**)&ctx, 1);
}
/*
** begin to work
*/
opt_list_append_opts_from_qalter_cmdline(me_who, &cmdline, &alp, argv + 1, environ);
tmp_ret = answer_list_print_err_warn(&alp, MSG_QALTER, MSG_QALTER, MSG_QALTERWARNING);
if (tmp_ret > 0) {
SGE_EXIT((void**)&ctx, tmp_ret);
}
/* handling the case that no command line parameter was specified */
if ((me_who == QHOLD || me_who == QRLS) && lGetNumberOfElem(cmdline) == 1) {
/* -h option is set implicitly for QHOLD and QRLS */
sge_usage(me_who, stderr);
fprintf(stderr, "%s\n", MSG_PARSE_NOOPTIONARGUMENT);
SGE_EXIT((void**)&ctx, 1);
} else if ((me_who == QRESUB || me_who == QALTER) && lGetNumberOfElem(cmdline) == 0) {
/* qresub and qalter have nothing set */
sge_usage(me_who, stderr);
fprintf(stderr, "%s\n", MSG_PARSE_NOOPTIONARGUMENT);
SGE_EXIT((void**)&ctx, 1);
} else if (opt_list_has_X(cmdline, "-help")) {
/* -help was specified */
sge_usage(me_who, stdout);
SGE_EXIT((void**)&ctx, 0);
}
alp = qalter_parse_job_parameter(me_who, cmdline, &request_list, &all_jobs,
&all_users);
DPRINTF(("all_jobs = %d, all_user = %d\n", all_jobs, all_users));
if (request_list && verify) {
/*
got a request list containing one element
for each job to be modified
save jobid all fields contain the same fields
so we may use show_job() with the first job
in our list
The jobid's in our request list get printed before
show_job()
*/
cull_show_job(lFirst(request_list), FLG_QALTER, false);
sge_prof_cleanup();
SGE_EXIT((void**)&ctx, 0);
}
tmp_ret = answer_list_print_err_warn(&alp, NULL, NULL, MSG_WARNING);
if (tmp_ret > 0) {
SGE_EXIT((void**)&ctx, tmp_ret);
}
if ((me_who == QALTER) ||
(me_who == QHOLD) ||
(me_who == QRLS)
) {
DPRINTF(("QALTER\n"));
gdi_cmd = SGE_GDI_MOD;
} else if (me_who == QRESUB){
DPRINTF(("QRESUB\n"));
gdi_cmd = SGE_GDI_COPY;
} else {
printf("unknown binary name.\n");
SGE_EXIT((void**)&ctx, 1);
}
if (all_jobs)
gdi_cmd |= SGE_GDI_ALL_JOBS;
if (all_users)
gdi_cmd |= SGE_GDI_ALL_USERS;
alp = ctx->gdi(ctx, SGE_JB_LIST, gdi_cmd, &request_list, NULL, NULL);
for_each (aep, alp) {
printf("%s\n", lGetString(aep, AN_text));
if (ret == STATUS_OK) {
ret = lGetUlong(aep, AN_status);
}
}
/****** sge_hostname/sge_get_qmaster_port() ************************************
* NAME
* sge_get_qmaster_port() -- get qmaster port
*
* SYNOPSIS
* int sge_get_qmaster_port(bool *from_services)
*
* FUNCTION
* This function returns the TCP/IP port of the qmaster daemon. It returns
* a cached value from a previous run. The cached value will be refreshed
* every 10 Minutes. The port may come from environment variable
* SGE_QMASTER_PORT or the services files entry "sge_qmaster".
*
* INPUTS
* bool *from_services - Pointer to a boolean which is set to true
* when the port value comes from the services file.
*
* RESULT
* int - port of qmaster
*
*******************************************************************************/
#define SGE_PORT_CACHE_TIMEOUT 60*10 /* 10 Min. */
int sge_get_qmaster_port(bool *from_services) {
char* port = NULL;
int int_port = -1;
struct timeval now;
static long next_timeout = 0;
static int cached_port = -1;
static bool is_port_from_services_file = false;
DENTER(GDI_LAYER, "sge_get_qmaster_port");
sge_mutex_lock("get_qmaster_port_mutex", SGE_FUNC, __LINE__, &get_qmaster_port_mutex);
/* check for reresolve timeout */
gettimeofday(&now,NULL);
if (next_timeout > 0 ) {
DPRINTF(("reresolve port timeout in "sge_U32CFormat"\n", sge_u32c( next_timeout - now.tv_sec)));
}
/* get port from cache when next_timeout for re-resolving is not reached */
if (cached_port >= 0 && next_timeout > now.tv_sec) {
int_port = cached_port;
if (from_services != NULL) {
*from_services = is_port_from_services_file;
}
DPRINTF(("returning cached port value: "sge_U32CFormat"\n", sge_u32c(int_port)));
sge_mutex_unlock("get_qmaster_port_mutex", SGE_FUNC, __LINE__, &get_qmaster_port_mutex);
DEXIT;
return int_port;
}
/* get port from environment variable SGE_QMASTER_PORT */
port = getenv("SGE_QMASTER_PORT");
if (port != NULL) {
int_port = atoi(port);
is_port_from_services_file = false;
}
/* get port from services file */
if (int_port <= 0) {
char buffer[2048];
struct servent se_result;
struct servent* se_help = NULL;
se_help = sge_getservbyname_r(&se_result, "sge_qmaster", buffer, sizeof(buffer));
if (se_help != NULL) {
int_port = ntohs(se_help->s_port);
if (int_port > 0) {
/* we found a port in the services */
is_port_from_services_file = true;
if (from_services != NULL) {
*from_services = is_port_from_services_file;
}
}
}
}
if (int_port <= 0 ) {
ERROR((SGE_EVENT, MSG_UTI_CANT_GET_ENV_OR_PORT_SS, "SGE_QMASTER_PORT", "sge_qmaster"));
if ( cached_port > 0 ) {
WARNING((SGE_EVENT, MSG_UTI_USING_CACHED_PORT_SU, "sge_qmaster", sge_u32c(cached_port) ));
int_port = cached_port;
} else {
sge_mutex_unlock("get_qmaster_port_mutex", SGE_FUNC, __LINE__, &get_qmaster_port_mutex);
SGE_EXIT(NULL, 1);
}
} else {
DPRINTF(("returning port value: "sge_U32CFormat"\n", sge_u32c(int_port)));
/* set new timeout time */
gettimeofday(&now,NULL);
next_timeout = now.tv_sec + SGE_PORT_CACHE_TIMEOUT;
/* set new port value */
cached_port = int_port;
}
sge_mutex_unlock("get_qmaster_port_mutex", SGE_FUNC, __LINE__, &get_qmaster_port_mutex);
DEXIT;
return int_port;
}
/*
* TODO: CLEANUP
*
* This function is DEPRECATED and should be removed with the next
* major release.
*
* This function can't be used in multi threaded environments because it
* might cause a deadlock in the executing qmaster thread.
* Use sge_peopen_r() instead.
*/
pid_t sge_peopen(const char *shell, int login_shell, const char *command,
const char *user, char **env, FILE **fp_in, FILE **fp_out,
FILE **fp_err, bool null_stderr)
{
pid_t pid;
int pipefds[3][2];
const char *could_not = MSG_SYSTEM_EXECBINSHFAILED;
const char *not_root = MSG_SYSTEM_NOROOTRIGHTSTOSWITCHUSER;
int i;
char arg0[256];
char err_str[256];
#if !(defined(WIN32) || defined(INTERIX)) /* var not needed */
int res;
#endif /* WIN32 */
uid_t myuid;
DENTER(TOP_LAYER, "sge_peopen");
/* open pipes - close on failure */
for (i=0; i<3; i++) {
if (pipe(pipefds[i]) != 0) {
while (--i >= 0) {
close(pipefds[i][0]);
close(pipefds[i][1]);
}
ERROR((SGE_EVENT, MSG_SYSTEM_FAILOPENPIPES_SS,
command, strerror(errno)));
DEXIT;
return -1;
}
}
#if defined(SOLARIS)
pid = sge_smf_contract_fork(err_str, 256);
#else
pid = fork();
#endif
if (pid == 0) { /* child */
int keep_open[6];
keep_open[0] = 0;
keep_open[1] = 1;
keep_open[2] = 2;
keep_open[3] = pipefds[0][0];
keep_open[4] = pipefds[1][1];
keep_open[5] = pipefds[2][1];
sge_close_all_fds(keep_open, 6);
/* shall we redirect stderr to /dev/null? */
if (null_stderr) {
/* open /dev/null */
int fd = open("/dev/null", O_WRONLY);
if (fd == -1) {
sprintf(err_str, MSG_ERROROPENINGFILEFORWRITING_SS, "/dev/null",
strerror(errno));
sprintf(err_str, "\n");
write(2, err_str, strlen(err_str));
SGE_EXIT(NULL, 1);
}
/* set stderr to /dev/null */
close(2);
dup(fd);
/* we don't need the stderr the pipe - close it */
close(pipefds[2][1]);
} else {
/* redirect stderr to the pipe */
close(2);
dup(pipefds[2][1]);
}
/* redirect stdin and stdout to the pipes */
close(0);
close(1);
dup(pipefds[0][0]);
dup(pipefds[1][1]);
if (user) {
struct passwd *pw;
struct passwd pw_struct;
char *buffer;
int size;
size = get_pw_buffer_size();
buffer = sge_malloc(size);
if (!(pw=sge_getpwnam_r(user, &pw_struct, buffer, size))) {
sprintf(err_str, MSG_SYSTEM_NOUSERFOUND_SS , user, strerror(errno));
sprintf(err_str, "\n");
write(2, err_str, strlen(err_str));
FREE(buffer);
SGE_EXIT(NULL, 1);
}
myuid = geteuid();
if (myuid != pw->pw_uid) {
/* Only change user if we differ from the wanted user */
if(myuid != SGE_SUPERUSER_UID) {
write(2, not_root, sizeof(not_root));
FREE(buffer);
SGE_EXIT(NULL, 1);
}
sprintf(err_str, "%s %d\n", pw->pw_name, (int)pw->pw_gid);
write(2, err_str, strlen(err_str));
#if !(defined(WIN32) || defined(INTERIX)) /* initgroups not called */
res = initgroups(pw->pw_name,pw->pw_gid);
# if defined(SVR3) || defined(sun)
if (res < 0)
# else
if (res)
# endif
{
sprintf(err_str, MSG_SYSTEM_INITGROUPSFORUSERFAILED_ISS ,
res, user, strerror(errno));
sprintf(err_str, "\n");
write(2, err_str, strlen(err_str));
FREE(buffer);
SGE_EXIT(NULL, 1);
}
#endif /* WIN32 */
if (setuid(pw->pw_uid)) {
sprintf(err_str, MSG_SYSTEM_SWITCHTOUSERFAILED_SS , user,
strerror(errno));
sprintf(err_str, "\n");
write(2, err_str, strlen(err_str));
FREE(buffer);
SGE_EXIT(NULL, 1);
}
}
addenv("HOME", pw->pw_dir);
addenv("SHELL", pw->pw_shell);
addenv("USER", pw->pw_name);
addenv("LOGNAME", pw->pw_name);
addenv("PATH", SGE_DEFAULT_PATH);
FREE(buffer);
}
if (login_shell)
strcpy(arg0, "-");
else
strcpy(arg0, "");
strcat(arg0, shell);
if (env)
for(; *env; env++)
putenv(*env);
execlp(shell, arg0, "-c", command, NULL);
write(2, could_not, sizeof(could_not));
SGE_EXIT(NULL, 1);
}
if (pid < 0) {
for (i=0; i<3; i++) {
close(pipefds[i][0]);
close(pipefds[i][1]);
}
#if defined(SOLARIS)
if (pid < -1 && err_str) {
ERROR((SGE_EVENT, MSG_SMF_FORK_FAILED_SS, "sge_peopen()", err_str));
}
#endif
/* fork could have failed, report it */
ERROR((SGE_EVENT, MSG_SMF_FORK_FAILED_SS, "sge_peopen()", strerror(errno)));
DEXIT;
return -1;
}
/* close the childs ends of the pipes */
close(pipefds[0][0]);
close(pipefds[1][1]);
close(pipefds[2][1]);
/* return filehandles for stdin and stdout */
*fp_in = fdopen(pipefds[0][1], "a");
*fp_out = fdopen(pipefds[1][0], "r");
/* is stderr redirected to /dev/null? */
if (null_stderr) {
/* close the pipe and return NULL as filehandle */
close(pipefds[2][0]);
*fp_err = NULL;
} else {
/* return filehandle for stderr */
*fp_err = fdopen(pipefds[2][0], "r");
}
DEXIT;
return pid;
}
int main(int argc, char *argv[])
{
const char *url;
int i, threads;
pthread_t *t;
int *args;
lList *answer_list = NULL;
lListElem *spooling_context;
DENTER_MAIN(TOP_LAYER, "test_berkeleydb_mt");
/* parse commandline parameters */
if (argc < 3) {
ERROR((SGE_EVENT, "usage: test_berkeleydb_mt <url> <threads> [<delay>]\n"));
ERROR((SGE_EVENT, " <url> = path or host:database\n"));
ERROR((SGE_EVENT, " <threads> = number of threads\n"));
ERROR((SGE_EVENT, " <delay> = delay after writing [ms]\n"));
SGE_EXIT(NULL, 1);
}
url = argv[1];
threads = atoi(argv[2]);
if (argc > 3) {
delay = atoi(argv[3]);
}
/* allocate memory for pthreads and arguments */
t = (pthread_t *)malloc(threads * sizeof(pthread_t));
args = (int *)malloc(threads * sizeof(int));
DPRINTF(("writing to database %s from %d threads\n", url, threads));
/* initialize spooling */
spooling_context = spool_create_dynamic_context(&answer_list, NULL, url, NULL);
answer_list_output(&answer_list);
if (spooling_context == NULL) {
SGE_EXIT(NULL, EXIT_FAILURE);
}
spool_set_default_context(spooling_context);
if (!spool_startup_context(&answer_list, spooling_context, true)) {
answer_list_output(&answer_list);
SGE_EXIT(NULL, EXIT_FAILURE);
}
answer_list_output(&answer_list);
/* let n threads to parallel spooling */
for (i = 0; i < threads; i++) {
args[i] = i + 1;
pthread_create(&(t[i]), NULL, work, (void*)(&args[i]));
}
/* also work in current thread */
work((void *)0);
/* wait for termination of all threads */
for (i = 0; i < threads; i++) {
pthread_join(t[i], NULL);
}
/* shutdown spooling */
spool_shutdown_context(&answer_list, spooling_context);
answer_list_output(&answer_list);
sge_free(&t);
DEXIT;
return EXIT_SUCCESS;
}
int sge_get_execd_port(void) {
char* port = NULL;
int int_port = -1;
struct timeval now;
static long next_timeout = 0;
static int cached_port = -1;
DENTER(TOP_LAYER, "sge_get_execd_port");
sge_mutex_lock("get_execd_port_mutex", SGE_FUNC, __LINE__, &get_execd_port_mutex);
/* check for reresolve timeout */
gettimeofday(&now,NULL);
if ( next_timeout > 0 ) {
DPRINTF(("reresolve port timeout in "sge_U32CFormat"\n", sge_u32c( next_timeout - now.tv_sec)));
}
if ( cached_port >= 0 && next_timeout > now.tv_sec ) {
int_port = cached_port;
DPRINTF(("returning cached port value: "sge_U32CFormat"\n", sge_u32c(int_port)));
sge_mutex_unlock("get_execd_port_mutex", SGE_FUNC, __LINE__, &get_execd_port_mutex);
return int_port;
}
port = getenv("SGE_EXECD_PORT");
if (port != NULL) {
int_port = atoi(port);
}
/* get port from services file */
if (int_port <= 0) {
char buffer[2048];
struct servent se_result;
struct servent* se_help = NULL;
se_help = sge_getservbyname_r(&se_result, "sge_execd", buffer, sizeof(buffer));
if (se_help != NULL) {
int_port = ntohs(se_help->s_port);
}
}
if (int_port <= 0 ) {
ERROR((SGE_EVENT, MSG_UTI_CANT_GET_ENV_OR_PORT_SS, "SGE_EXECD_PORT" , "sge_execd"));
if ( cached_port > 0 ) {
WARNING((SGE_EVENT, MSG_UTI_USING_CACHED_PORT_SU, "sge_execd", sge_u32c(cached_port) ));
int_port = cached_port;
} else {
sge_mutex_unlock("get_execd_port_mutex", SGE_FUNC, __LINE__, &get_execd_port_mutex);
SGE_EXIT(NULL, 1);
}
} else {
DPRINTF(("returning port value: "sge_U32CFormat"\n", sge_u32c(int_port)));
/* set new timeout time */
gettimeofday(&now,NULL);
next_timeout = now.tv_sec + SGE_PORT_CACHE_TIMEOUT;
/* set new port value */
cached_port = int_port;
}
sge_mutex_unlock("get_execd_port_mutex", SGE_FUNC, __LINE__, &get_execd_port_mutex);
DEXIT;
return int_port;
}
int main(int argc, char *argv[])
{
qevent_options enabled_options;
dstring errors = DSTRING_INIT;
int i, gdi_setup;
lList *alp = NULL;
sge_gdi_ctx_class_t *ctx = NULL;
sge_evc_class_t *evc = NULL;
DENTER_MAIN(TOP_LAYER, "qevent");
/* sge_mt_init(); */
/* dump pid to file */
qevent_dump_pid_file();
/* parse command line */
enabled_options.error_message = &errors;
qevent_set_option_struct(&enabled_options);
qevent_parse_command_line(argc, argv, &enabled_options);
/* check if help option is set */
if (enabled_options.help_option) {
qevent_show_usage();
sge_dstring_free(enabled_options.error_message);
SGE_EXIT((void**)&ctx, 0);
}
/* are there command line parsing errors ? */
if (sge_dstring_get_string(enabled_options.error_message)) {
ERROR((SGE_EVENT, "%s", sge_dstring_get_string(enabled_options.error_message) ));
qevent_show_usage();
sge_dstring_free(enabled_options.error_message);
SGE_EXIT((void**)&ctx, 1);
}
log_state_set_log_gui(1);
sge_setup_sig_handlers(QEVENT);
/* setup event client */
gdi_setup = sge_gdi2_setup(&ctx, QEVENT, MAIN_THREAD, &alp);
if (gdi_setup != AE_OK) {
answer_list_output(&alp);
sge_dstring_free(enabled_options.error_message);
SGE_EXIT((void**)&ctx, 1);
}
/* TODO: how is the memory we allocate here released ???, SGE_EXIT doesn't */
if (false == sge_gdi2_evc_setup(&evc, ctx, EV_ID_ANY, &alp, NULL)) {
answer_list_output(&alp);
sge_dstring_free(enabled_options.error_message);
SGE_EXIT((void**)&ctx, 1);
}
/* ok, start over ... */
/* check for testsuite option */
if (enabled_options.testsuite_option) {
/* only for testsuite */
qevent_testsuite_mode(evc);
sge_dstring_free(enabled_options.error_message);
SGE_EXIT((void**)&ctx, 0);
}
/* check for subscribe option */
if (enabled_options.subscribe_option) {
/* only for testsuite */
qevent_subscribe_mode(evc);
sge_dstring_free(enabled_options.error_message);
SGE_EXIT((void**)&ctx, 0);
}
if (enabled_options.trigger_option_count > 0) {
lCondition *where =NULL;
lEnumeration *what = NULL;
sge_mirror_initialize(evc, EV_ID_ANY, "sge_mirror -trigger", true,
NULL, NULL, NULL, NULL, NULL);
evc->ec_set_busy_handling(evc, EV_BUSY_UNTIL_ACK);
/* put out information about -trigger option */
for (i=0;i<enabled_options.trigger_option_count;i++) {
INFO((SGE_EVENT, "trigger script for %s events: %s\n",
qevent_get_event_name((enabled_options.trigger_option_events)[i]),
(enabled_options.trigger_option_scripts)[i]));
switch((enabled_options.trigger_option_events)[i]) {
case QEVENT_JB_END:
/* build mask for the job structure to contain only the needed elements */
where = NULL;
what = lWhat("%T(%I %I %I %I %I %I %I %I)", JB_Type, JB_job_number, JB_ja_tasks,
JB_ja_structure, JB_ja_n_h_ids, JB_ja_u_h_ids,
JB_ja_s_h_ids,JB_ja_o_h_ids, JB_ja_template);
/* register for job events */
sge_mirror_subscribe(evc, SGE_TYPE_JOB, analyze_jatask_event, NULL, NULL, where, what);
evc->ec_set_flush(evc, sgeE_JOB_DEL,true, 1);
/* the mirror interface registers more events, than we need,
thus we free the ones, we do not need */
/* evc->ec_unsubscribe(evc, sgeE_JOB_LIST); */
evc->ec_unsubscribe(evc, sgeE_JOB_MOD);
evc->ec_unsubscribe(evc, sgeE_JOB_MOD_SCHED_PRIORITY);
evc->ec_unsubscribe(evc, sgeE_JOB_USAGE);
evc->ec_unsubscribe(evc, sgeE_JOB_FINAL_USAGE);
/* evc->ec_unsubscribe(evc, sgeE_JOB_ADD); */
/* free the what and where mask */
lFreeWhere(&where);
lFreeWhat(&what);
break;
case QEVENT_JB_TASK_END:
/* build mask for the job structure to contain only the needed elements */
where = NULL;
what = lWhat("%T(%I)", JAT_Type, JAT_status);
/* register for JAT events */
sge_mirror_subscribe(evc, SGE_TYPE_JATASK, analyze_jatask_event, NULL, NULL, where, what);
evc->ec_set_flush(evc, sgeE_JATASK_DEL,true, 1);
/* the mirror interface registers more events, than we need,
thus we free the ones, we do not need */
evc->ec_unsubscribe(evc, sgeE_JATASK_ADD);
evc->ec_unsubscribe(evc, sgeE_JATASK_MOD);
/* free the what and where mask */
lFreeWhere(&where);
lFreeWhat(&what);
break;
}
}
while(!shut_me_down) {
sge_mirror_error error = sge_mirror_process_events(evc);
if (error == SGE_EM_TIMEOUT && !shut_me_down ) {
sleep(10);
continue;
}
}
sge_mirror_shutdown(evc);
sge_dstring_free(enabled_options.error_message);
sge_prof_cleanup();
SGE_EXIT((void**)&ctx, 0);
return 0;
}
ERROR((SGE_EVENT, "no option selected\n" ));
qevent_show_usage();
sge_dstring_free(enabled_options.error_message);
sge_prof_cleanup();
SGE_EXIT((void**)&ctx, 1);
return 1;
}
/*-------------------------------------------------------------------------*/
int main(int argc, char **argv)
{
int ret;
int my_pid;
int ret_val;
int printed_points = 0;
int max_enroll_tries;
static char tmp_err_file_name[SGE_PATH_MAX];
time_t next_prof_output = 0;
int execd_exit_state = 0;
lList **master_job_list = NULL;
sge_gdi_ctx_class_t *ctx = NULL;
lList *alp = NULL;
DENTER_MAIN(TOP_LAYER, "execd");
#if defined(LINUX)
gen_procList ();
#endif
prof_mt_init();
set_thread_name(pthread_self(),"Execd Thread");
prof_set_level_name(SGE_PROF_CUSTOM1, "Execd Thread", NULL);
prof_set_level_name(SGE_PROF_CUSTOM2, "Execd Dispatch", NULL);
#ifdef __SGE_COMPILE_WITH_GETTEXT__
/* init language output for gettext() , it will use the right language */
sge_init_language_func((gettext_func_type) gettext,
(setlocale_func_type) setlocale,
(bindtextdomain_func_type) bindtextdomain,
(textdomain_func_type) textdomain);
sge_init_language(NULL,NULL);
#endif /* __SGE_COMPILE_WITH_GETTEXT__ */
/* This needs a better solution */
umask(022);
/* Initialize path for temporary logging until we chdir to spool */
my_pid = getpid();
sprintf(tmp_err_file_name,"%s."sge_U32CFormat"", TMP_ERR_FILE_EXECD, sge_u32c(my_pid));
log_state_set_log_file(tmp_err_file_name);
/* exit func for SGE_EXIT() */
sge_sig_handler_in_main_loop = 0;
sge_setup_sig_handlers(EXECD);
if (sge_setup2(&ctx, EXECD, MAIN_THREAD, &alp, false) != AE_OK) {
answer_list_output(&alp);
SGE_EXIT((void**)&ctx, 1);
}
ctx->set_exit_func(ctx, execd_exit_func);
#if defined(SOLARIS)
/* Init shared SMF libs if necessary */
if (sge_smf_used() == 1 && sge_smf_init_libs() != 0) {
SGE_EXIT((void**)&ctx, 1);
}
#endif
/* prepare daemonize */
if (!getenv("SGE_ND")) {
sge_daemonize_prepare(ctx);
}
if ((ret=sge_occupy_first_three())>=0) {
CRITICAL((SGE_EVENT, MSG_FILE_REDIRECTFD_I, ret));
SGE_EXIT((void**)&ctx, 1);
}
lInit(nmv);
/* unset XAUTHORITY if set */
if (getenv("XAUTHORITY") != NULL) {
sge_unsetenv("XAUTHORITY");
}
parse_cmdline_execd(argv);
/* exit if we can't get communication handle (bind port) */
max_enroll_tries = 30;
while (cl_com_get_handle(prognames[EXECD],1) == NULL) {
ctx->prepare_enroll(ctx);
max_enroll_tries--;
if (max_enroll_tries <= 0 || shut_me_down) {
/* exit after 30 seconds */
if (printed_points != 0) {
printf("\n");
}
CRITICAL((SGE_EVENT, MSG_COM_ERROR));
SGE_EXIT((void**)&ctx, 1);
}
if (cl_com_get_handle(prognames[EXECD],1) == NULL) {
/* sleep when prepare_enroll() failed */
sleep(1);
if (max_enroll_tries < 27) {
printf(".");
printed_points++;
fflush(stdout);
}
}
}
if (printed_points != 0) {
printf("\n");
}
/*
* now the commlib up and running. Set execd application status function
* ( commlib callback function for qping status information response
* messages (SIRM) )
*/
ret_val = cl_com_set_status_func(sge_execd_application_status);
if (ret_val != CL_RETVAL_OK) {
ERROR((SGE_EVENT, cl_get_error_text(ret_val)) );
}
/* test connection */
{
cl_com_SIRM_t* status = NULL;
ret_val = cl_commlib_get_endpoint_status(ctx->get_com_handle(ctx),
(char *)ctx->get_master(ctx, true),
(char*)prognames[QMASTER], 1, &status);
if (ret_val != CL_RETVAL_OK) {
ERROR((SGE_EVENT, cl_get_error_text(ret_val)));
ERROR((SGE_EVENT, MSG_CONF_NOCONFBG));
}
cl_com_free_sirm_message(&status);
}
/* finalize daeamonize */
if (!getenv("SGE_ND")) {
sge_daemonize_finalize(ctx);
}
/* daemonizes if qmaster is unreachable */
sge_setup_sge_execd(ctx, tmp_err_file_name);
/* are we using qidle or not */
sge_ls_qidle(mconf_get_use_qidle());
sge_ls_gnu_ls(1);
DPRINTF(("use_qidle: %d\n", mconf_get_use_qidle()));
/* test load sensor (internal or external) */
{
lList *report_list = sge_build_load_report(ctx->get_qualified_hostname(ctx), ctx->get_binary_path(ctx));
lFreeList(&report_list);
}
/* here we have to wait for qmaster registration */
while (sge_execd_register_at_qmaster(ctx, false) != 0) {
if (sge_get_com_error_flag(EXECD, SGE_COM_ACCESS_DENIED, true)) {
/* This is no error */
DPRINTF(("***** got SGE_COM_ACCESS_DENIED from qmaster *****\n"));
}
if (sge_get_com_error_flag(EXECD, SGE_COM_ENDPOINT_NOT_UNIQUE, false)) {
execd_exit_state = SGE_COM_ENDPOINT_NOT_UNIQUE;
break;
}
if (shut_me_down != 0) {
break;
}
sleep(30);
}
/*
* Terminate on SIGTERM or hard communication error
*/
if (execd_exit_state != 0 || shut_me_down != 0) {
sge_shutdown((void**)&ctx, execd_exit_state);
DRETURN(execd_exit_state);
}
/*
* We write pid file when we are connected to qmaster. Otherwise an old
* execd might overwrite our pidfile.
*/
sge_write_pid(EXECD_PID_FILE);
/*
* At this point we are sure we are the only sge_execd and we are connected
* to the current qmaster. First we have to report any reaped children
* that might exist.
*/
starting_up();
/*
* Log a warning message if execd hasn't been started by a superuser
*/
if (!sge_is_start_user_superuser()) {
WARNING((SGE_EVENT, MSG_SWITCH_USER_NOT_ROOT));
}
#ifdef COMPILE_DC
if (ptf_init()) {
CRITICAL((SGE_EVENT, MSG_EXECD_NOSTARTPTF));
SGE_EXIT((void**)&ctx, 1);
}
INFO((SGE_EVENT, MSG_EXECD_STARTPDCANDPTF));
#endif
master_job_list = object_type_get_master_list(SGE_TYPE_JOB);
*master_job_list = lCreateList("Master_Job_List", JB_Type);
job_list_read_from_disk(master_job_list, "Master_Job_List",
0, SPOOL_WITHIN_EXECD,
job_initialize_job);
/* clean up jobs hanging around (look in active_dir) */
clean_up_old_jobs(ctx, 1);
execd_trash_load_report();
sge_set_flush_lr_flag(true);
sge_sig_handler_in_main_loop = 1;
if (thread_prof_active_by_id(pthread_self())) {
prof_start(SGE_PROF_CUSTOM1, NULL);
prof_start(SGE_PROF_CUSTOM2, NULL);
prof_start(SGE_PROF_GDI_REQUEST, NULL);
} else {
prof_stop(SGE_PROF_CUSTOM1, NULL);
prof_stop(SGE_PROF_CUSTOM2, NULL);
prof_stop(SGE_PROF_GDI_REQUEST, NULL);
}
PROF_START_MEASUREMENT(SGE_PROF_CUSTOM1);
/* Start dispatching */
execd_exit_state = sge_execd_process_messages(ctx);
/*
* This code is only reached when dispatcher terminates and execd goes down.
*/
/* log if we received SIGPIPE signal */
if (sge_sig_handler_sigpipe_received) {
sge_sig_handler_sigpipe_received = 0;
INFO((SGE_EVENT, "SIGPIPE received\n"));
}
#if defined(LINUX)
free_procList();
#endif
lFreeList(master_job_list);
PROF_STOP_MEASUREMENT(SGE_PROF_CUSTOM1);
if (prof_is_active(SGE_PROF_ALL)) {
time_t now = (time_t)sge_get_gmt();
if (now > next_prof_output) {
prof_output_info(SGE_PROF_ALL, false, "profiling summary:\n");
prof_reset(SGE_PROF_ALL,NULL);
next_prof_output = now + 60;
}
}
sge_prof_cleanup();
sge_shutdown((void**)&ctx, execd_exit_state);
DRETURN(execd_exit_state);
}
