1
1

Merge pull request #3603 from rhc54/topic/fds

Add some debug code for detecting leaking file descriptors.
Этот коммит содержится в:
Ralph Castain 2017-05-29 12:20:57 -07:00 коммит произвёл GitHub
родитель 8e583dcf76 f3ab326b4a
Коммит a8b34d975a
4 изменённых файлов: 132 добавлений и 3 удалений

Просмотреть файл

@ -13,6 +13,13 @@
#include "orte_config.h"
#include "orte/constants.h"
#if HAVE_UNISTD_H
#include <unistd.h>
#endif
#if HAVE_FCNTL_H
#include <fcntl.h>
#endif
#include "opal/class/opal_list.h"
#include "opal/mca/event/event.h"
#include "opal/mca/pmix/pmix.h"
@ -714,6 +721,10 @@ void orte_state_base_track_procs(int fd, short argc, void *cbdata)
/* track job status */
jdata->num_terminated++;
if (jdata->num_terminated == jdata->num_procs) {
/* if requested, check fd status for leaks */
if (orte_state_base_run_fdcheck) {
orte_state_base_check_fds(jdata);
}
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_TERMINATED);
/* if they requested notification upon completion, provide it */
if (orte_get_attribute(&jdata->attributes, ORTE_JOB_NOTIFY_COMPLETION, NULL, OPAL_BOOL)) {
@ -1016,3 +1027,99 @@ void orte_state_base_check_all_complete(int fd, short args, void *cbdata)
OBJ_RELEASE(caddy);
}
void orte_state_base_check_fds(orte_job_t *jdata)
{
int nfds, i, fdflags, flflags;
char path[1024], info[256], **list=NULL, *status, *result, *r2;
ssize_t rc;
struct flock fl;
int cnt = 0;
/* get the number of available file descriptors
* for this daemon */
nfds = getdtablesize();
result = NULL;
/* loop over them and get their info */
for (i=0; i < nfds; i++) {
fdflags = fcntl(i, F_GETFD);
if (-1 == fdflags) {
/* no open fd in that slot */
continue;
}
flflags = fcntl(i, F_GETFL);
if (-1 == flflags) {
/* no open fd in that slot */
continue;
}
snprintf(path, 1024, "/proc/self/fd/%d", i);
memset(info, 0, 256);
/* read the info about this fd */
rc = readlink(path, info, 256);
if (-1 == rc) {
/* this fd is unavailable */
continue;
}
/* get any file locking status */
fl.l_type = F_WRLCK;
fl.l_whence = 0;
fl.l_start = 0;
fl.l_len = 0;
fcntl(i, F_GETLK, &fl);
/* construct the list of capabilities */
if (fdflags & FD_CLOEXEC) {
opal_argv_append_nosize(&list, "cloexec");
}
if (flflags & O_APPEND) {
opal_argv_append_nosize(&list, "append");
}
if (flflags & O_NONBLOCK) {
opal_argv_append_nosize(&list, "nonblock");
}
if (flflags & O_RDONLY) {
opal_argv_append_nosize(&list, "rdonly");
}
if (flflags & O_RDWR) {
opal_argv_append_nosize(&list, "rdwr");
}
if (flflags & O_WRONLY) {
opal_argv_append_nosize(&list, "wronly");
}
if (flflags & O_DSYNC) {
opal_argv_append_nosize(&list, "dsync");
}
if (flflags & O_RSYNC) {
opal_argv_append_nosize(&list, "rsync");
}
if (flflags & O_SYNC) {
opal_argv_append_nosize(&list, "sync");
}
if (F_UNLCK != fl.l_type) {
if (F_WRLCK == fl.l_type) {
opal_argv_append_nosize(&list, "wrlock");
} else {
opal_argv_append_nosize(&list, "rdlock");
}
}
if (NULL != list) {
status = opal_argv_join(list, ' ');
opal_argv_free(list);
list = NULL;
if (NULL == result) {
asprintf(&result, " %d\t(%s)\t%s\n", i, info, status);
} else {
asprintf(&r2, "%s %d\t(%s)\t%s\n", result, i, info, status);
free(result);
result = r2;
}
free(status);
}
++cnt;
}
asprintf(&r2, "%s: %d open file descriptors after job %d completed\n%s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), cnt, ORTE_LOCAL_JOBID(jdata->jobid), result);
opal_output(0, "%s", r2);
free(result);
free(r2);
}

Просмотреть файл

@ -4,6 +4,7 @@
* All rights reserved.
* Copyright (c) 2015 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2017 Intel, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -41,6 +42,20 @@
* Globals
*/
orte_state_base_module_t orte_state = {0};
bool orte_state_base_run_fdcheck = false;
static int orte_state_base_register(mca_base_register_flag_t flags)
{
orte_state_base_run_fdcheck = false;
mca_base_var_register("orte", "state", "base", "check_fds",
"Daemons should check fds for leaks after each job completes",
MCA_BASE_VAR_TYPE_BOOL, NULL, 0, 0,
OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_READONLY,
&orte_state_base_run_fdcheck);
return ORTE_SUCCESS;
}
static int orte_state_base_close(void)
{
@ -62,7 +77,8 @@ static int orte_state_base_open(mca_base_open_flag_t flags)
return mca_base_framework_components_open(&orte_state_base_framework, flags);
}
MCA_BASE_FRAMEWORK_DECLARE(orte, state, "ORTE State Machine", NULL,
MCA_BASE_FRAMEWORK_DECLARE(orte, state, "ORTE State Machine",
orte_state_base_register,
orte_state_base_open, orte_state_base_close,
mca_state_base_static_components, 0);
@ -95,4 +111,3 @@ OBJ_CLASS_INSTANCE(orte_state_caddy_t,
opal_object_t,
orte_state_caddy_construct,
orte_state_caddy_destruct);

Просмотреть файл

@ -1,6 +1,7 @@
/*
* Copyright (c) 2011-2013 Los Alamos National Security, LLC.
* All rights reserved.
* Copyright (c) 2017 Intel, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -31,6 +32,7 @@
BEGIN_C_DECLS
extern bool orte_state_base_run_fdcheck;
/*
* Base functions
*/
@ -75,7 +77,7 @@ ORTE_DECLSPEC void orte_state_base_cleanup_job(int fd, short argc, void *cbdata)
ORTE_DECLSPEC void orte_state_base_report_progress(int fd, short argc, void *cbdata);
ORTE_DECLSPEC void orte_state_base_track_procs(int fd, short argc, void *cbdata);
ORTE_DECLSPEC void orte_state_base_check_all_complete(int fd, short args, void *cbdata);
ORTE_DECLSPEC void orte_state_base_check_fds(orte_job_t *jdata);
END_C_DECLS
#endif

Просмотреть файл

@ -484,6 +484,11 @@ static void track_procs(int fd, short argc, void *cbdata)
jdata->map = NULL;
}
/* if requested, check fd status for leaks */
if (orte_state_base_run_fdcheck) {
orte_state_base_check_fds(jdata);
}
/* cleanup the job info */
opal_hash_table_set_value_uint32(orte_job_data, jdata->jobid, NULL);
OBJ_RELEASE(jdata);