From 1b52ea8a94c9cc7af3c2ff82f74da930ff0a063d Mon Sep 17 00:00:00 2001
From: Danny Auble <da@schedmd.com>
Date: Tue, 23 Jul 2013 14:00:06 -0700
Subject: [PATCH] Remove duplicate job_container/none logic and made
job_container/cncu work without a real cray system (pretty much what
job_container/none used to do.
---
src/plugins/job_container/cncu/Makefile.am | 5 +-
src/plugins/job_container/cncu/Makefile.in | 7 +-
.../job_container/cncu/job_container_cncu.c | 24 +-
.../job_container/none/job_container_none.c | 229 +-----------------
src/plugins/proctrack/pgid/Makefile.in | 22 +-
5 files changed, 39 insertions(+), 248 deletions(-)
diff --git a/src/plugins/job_container/cncu/Makefile.am b/src/plugins/job_container/cncu/Makefile.am
index ea411c1aa2c..88c9edb650e 100644
--- a/src/plugins/job_container/cncu/Makefile.am
+++ b/src/plugins/job_container/cncu/Makefile.am
@@ -6,10 +6,7 @@ PLUGIN_FLAGS = -module -avoid-version --export-dynamic
AM_CPPFLAGS = -I$(top_srcdir) -I$(top_srcdir)/src/common $(CRAY_CPPFLAGS)
-if HAVE_REAL_CRAY
-CNCU_PLUGIN = job_container_cncu.la
-endif
-pkglib_LTLIBRARIES = $(CNCU_PLUGIN)
+pkglib_LTLIBRARIES = job_container_cncu.la
# job_container/cncu plugin.
job_container_cncu_la_SOURCES = job_container_cncu.c
diff --git a/src/plugins/job_container/cncu/Makefile.in b/src/plugins/job_container/cncu/Makefile.in
index 9bc5aa5d42a..76a2e19a5fa 100644
--- a/src/plugins/job_container/cncu/Makefile.in
+++ b/src/plugins/job_container/cncu/Makefile.in
@@ -139,8 +139,6 @@ job_container_cncu_la_OBJECTS = $(am_job_container_cncu_la_OBJECTS)
job_container_cncu_la_LINK = $(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) \
$(LIBTOOLFLAGS) --mode=link $(CCLD) $(AM_CFLAGS) $(CFLAGS) \
$(job_container_cncu_la_LDFLAGS) $(LDFLAGS) -o $@
-@HAVE_REAL_CRAY_TRUE@am_job_container_cncu_la_rpath = -rpath \
-@HAVE_REAL_CRAY_TRUE@ $(pkglibdir)
DEFAULT_INCLUDES = -I.@am__isrc@ -I$(top_builddir) -I$(top_builddir)/slurm
depcomp = $(SHELL) $(top_srcdir)/auxdir/depcomp
am__depfiles_maybe = depfiles
@@ -398,8 +396,7 @@ top_srcdir = @top_srcdir@
AUTOMAKE_OPTIONS = foreign
PLUGIN_FLAGS = -module -avoid-version --export-dynamic
AM_CPPFLAGS = -I$(top_srcdir) -I$(top_srcdir)/src/common $(CRAY_CPPFLAGS)
-@HAVE_REAL_CRAY_TRUE@CNCU_PLUGIN = job_container_cncu.la
-pkglib_LTLIBRARIES = $(CNCU_PLUGIN)
+pkglib_LTLIBRARIES = job_container_cncu.la
# job_container/cncu plugin.
job_container_cncu_la_SOURCES = job_container_cncu.c
@@ -471,7 +468,7 @@ clean-pkglibLTLIBRARIES:
rm -f "$${dir}/so_locations"; \
done
job_container_cncu.la: $(job_container_cncu_la_OBJECTS) $(job_container_cncu_la_DEPENDENCIES) $(EXTRA_job_container_cncu_la_DEPENDENCIES)
- $(job_container_cncu_la_LINK) $(am_job_container_cncu_la_rpath) $(job_container_cncu_la_OBJECTS) $(job_container_cncu_la_LIBADD) $(LIBS)
+ $(job_container_cncu_la_LINK) -rpath $(pkglibdir) $(job_container_cncu_la_OBJECTS) $(job_container_cncu_la_LIBADD) $(LIBS)
mostlyclean-compile:
-rm -f *.$(OBJEXT)
diff --git a/src/plugins/job_container/cncu/job_container_cncu.c b/src/plugins/job_container/cncu/job_container_cncu.c
index 64e4ae208b0..999adbbf4af 100644
--- a/src/plugins/job_container/cncu/job_container_cncu.c
+++ b/src/plugins/job_container/cncu/job_container_cncu.c
@@ -43,7 +43,9 @@
#include <sys/stat.h>
#include <fcntl.h>
+#ifdef HAVE_REAL_CRAY
#include <job.h> /* Cray's job module component */
+#endif
#include "slurm/slurm_errno.h"
#include "src/common/slurm_xlator.h"
@@ -183,6 +185,7 @@ static int _restore_state(char *dir_name)
return error_code;
}
+#ifdef HAVE_REAL_CRAY
static void _stat_reservation(char *type, rid_t resv_id)
{
struct job_resv_stat buf;
@@ -196,6 +199,7 @@ static void _stat_reservation(char *type, rid_t resv_id)
buf.num_files, buf.num_ipc_objs);
}
}
+#endif
static bool _get_debug_flag(void)
{
@@ -266,8 +270,10 @@ extern int container_p_restore(char *dir_name, bool recover)
extern int container_p_create(uint32_t job_id)
{
+#ifdef HAVE_REAL_CRAY
rid_t resv_id = job_id;
int rc;
+#endif
int i, empty = -1, found = -1;
bool job_id_change = false;
@@ -298,6 +304,7 @@ extern int container_p_create(uint32_t job_id)
_save_state(state_dir);
slurm_mutex_unlock(&context_lock);
+#ifdef HAVE_REAL_CRAY
rc = job_create_reservation(resv_id, CREATE_FLAGS);
if ((rc == 0) || (errno == EEXIST)) {
if ((rc != 0) && (errno == EEXIST)) {
@@ -308,6 +315,9 @@ extern int container_p_create(uint32_t job_id)
_stat_reservation("create", resv_id);
return SLURM_SUCCESS;
}
+#else
+ return SLURM_SUCCESS;
+#endif
error("%s: create(%u): %m", plugin_type, job_id);
return SLURM_ERROR;
}
@@ -315,15 +325,18 @@ extern int container_p_create(uint32_t job_id)
/* Add proctrack container (PAGG) to a job container */
extern int container_p_add_cont(uint32_t job_id, uint64_t cont_id)
{
+#ifdef HAVE_REAL_CRAY
jid_t cjob_id = cont_id;
rid_t resv_id = job_id;
int rc;
+#endif
if (enable_debug) {
info("%s: adding cont(%u.%"PRIu64")",
plugin_type, job_id, cont_id);
}
+#ifdef HAVE_REAL_CRAY
rc = job_attach_reservation(cjob_id, resv_id, ADD_FLAGS);
if ((rc != 0) && (errno == ENOENT)) { /* Log and retry */
error("%s: add(%u.%"PRIu64"): No reservation found",
@@ -331,11 +344,15 @@ extern int container_p_add_cont(uint32_t job_id, uint64_t cont_id)
rc = job_create_reservation(resv_id, CREATE_FLAGS);
rc = job_attach_reservation(cjob_id, resv_id, ADD_FLAGS);
}
+
if (rc == 0) {
if (enable_debug)
_stat_reservation("add", resv_id);
return SLURM_SUCCESS;
}
+#else
+ return SLURM_SUCCESS;
+#endif
error("%s: add(%u.%"PRIu64"): %m", plugin_type, job_id, cont_id);
return SLURM_ERROR;
}
@@ -364,8 +381,10 @@ extern int container_p_add_pid(uint32_t job_id, pid_t pid, uid_t uid)
extern int container_p_delete(uint32_t job_id)
{
+#ifdef HAVE_REAL_CRAY
rid_t resv_id = job_id;
- int rc;
+#endif
+ int rc = 0;
int i, found = -1;
bool job_id_change = false;
@@ -384,8 +403,9 @@ extern int container_p_delete(uint32_t job_id)
if (job_id_change)
_save_state(state_dir);
slurm_mutex_unlock(&context_lock);
-
+#ifdef HAVE_REAL_CRAY
rc = job_end_reservation(resv_id, DELETE_FLAGS);
+#endif
if (rc == 0)
return SLURM_SUCCESS;
diff --git a/src/plugins/job_container/none/job_container_none.c b/src/plugins/job_container/none/job_container_none.c
index d883b072720..94705767921 100644
--- a/src/plugins/job_container/none/job_container_none.c
+++ b/src/plugins/job_container/none/job_container_none.c
@@ -42,11 +42,8 @@
#include <sys/stat.h>
#include <sys/types.h>
-#include "slurm/slurm_errno.h"
#include "src/common/slurm_xlator.h"
-#include "src/slurmd/common/proctrack.h"
-
-#define JOB_BUF_SIZE 128
+#include "slurm/slurm_errno.h"
/*
* These variables are required by the generic plugin interface. If they
@@ -79,124 +76,9 @@ const char plugin_name[] = "job_container none plugin";
const char plugin_type[] = "job_container/none";
const uint32_t plugin_version = 101;
-char *state_dir = NULL; /* state save directory */
-static bool enable_debug = false;
-
-static uint32_t *job_id_array = NULL;
-static uint32_t job_id_count = 0;
-static pthread_mutex_t context_lock = PTHREAD_MUTEX_INITIALIZER;
-
-static int _save_state(char *dir_name)
-{
- char *file_name;
- int ret = SLURM_SUCCESS;
- int state_fd;
-
- if (!dir_name) {
- error("job_container state directory is NULL");
- return SLURM_ERROR;
- }
- file_name = xstrdup(dir_name);
- xstrcat(file_name, "/job_container_state");
- (void) unlink(file_name);
- state_fd = creat(file_name, 0600);
- if (state_fd < 0) {
- error("Can't save state, error creating file %s %m",
- file_name);
- ret = SLURM_ERROR;
- } else {
- char *buf = (char *) job_id_array;
- size_t len = job_id_count * sizeof(uint32_t);
- while (1) {
- int wrote = write(state_fd, buf, len);
- if ((wrote < 0) && (errno == EINTR))
- continue;
- if (wrote == 0)
- break;
- if (wrote < 0) {
- error("Can't save job_container state: %m");
- ret = SLURM_ERROR;
- break;
- }
- buf += wrote;
- len -= wrote;
- }
- close(state_fd);
- }
- xfree(file_name);
-
- return ret;
-}
-
-static int _restore_state(char *dir_name)
-{
- char *data = NULL, *file_name;
- int error_code = SLURM_SUCCESS;
- int state_fd, data_allocated = 0, data_read = 0, data_size = 0;
-
- if (!dir_name) {
- error("job_container state directory is NULL");
- return SLURM_ERROR;
- }
-
- file_name = xstrdup(dir_name);
- xstrcat(file_name, "/job_container_state");
- state_fd = open (file_name, O_RDONLY);
- if (state_fd >= 0) {
- data_allocated = JOB_BUF_SIZE;
- data = xmalloc(data_allocated);
- while (1) {
- data_read = read(state_fd, &data[data_size],
- JOB_BUF_SIZE);
- if ((data_read < 0) && (errno == EINTR))
- continue;
- if (data_read < 0) {
- error ("Read error on %s, %m", file_name);
- error_code = SLURM_ERROR;
- break;
- } else if (data_read == 0)
- break;
- data_size += data_read;
- data_allocated += data_read;
- xrealloc(data, data_allocated);
- }
- close(state_fd);
- xfree(file_name);
- } else {
- error("No %s file for %s state recovery",
- file_name, plugin_type);
- xfree(file_name);
- return SLURM_SUCCESS;
- }
-
- if (error_code == SLURM_SUCCESS) {
- job_id_array = (uint32_t *) data;
- job_id_count = data_size / sizeof(uint32_t);
- }
-
- return error_code;
-}
-
-static bool _get_debug_flag(void)
-{
- if (slurm_get_debug_flags() & DEBUG_FLAG_JOB_CONT)
- return true;
- return false;
-}
-
extern void container_p_reconfig(void)
{
- bool new_debug_flag = _get_debug_flag();
-
- if (!enable_debug && new_debug_flag) {
- error("%s: DebugFlag enabled by reconfiguration, this may "
- "result in errors due to missing job cache information",
- plugin_name);
- } else if (enable_debug != new_debug_flag) {
- debug("%s: JobContainer DebugFlag changed to %d",
- plugin_name, (int) new_debug_flag);
- }
- enable_debug = new_debug_flag;
+ return;
}
/*
@@ -205,11 +87,7 @@ extern void container_p_reconfig(void)
*/
extern int init(void)
{
- enable_debug = _get_debug_flag();
- if (enable_debug)
- info("%s loaded", plugin_name);
- else
- debug("%s loaded", plugin_name);
+ debug("%s loaded", plugin_name);
return SLURM_SUCCESS;
}
@@ -220,133 +98,32 @@ extern int init(void)
*/
extern int fini(void)
{
- xfree(state_dir);
return SLURM_SUCCESS;
}
extern int container_p_restore(char *dir_name, bool recover)
{
- if (enable_debug) {
- int i;
-
- slurm_mutex_lock(&context_lock);
- _restore_state(dir_name);
- slurm_mutex_unlock(&context_lock);
- for (i = 0; i < job_id_count; i++) {
- if (job_id_array[i] == 0)
- continue;
- if (recover) {
- info("%s: recovered job(%u)",
- plugin_type, job_id_array[i]);
- } else {
- info("%s: purging job(%u)",
- plugin_type, job_id_array[i]);
- job_id_array[i] = 0;
- }
- }
- }
-
- xfree(state_dir);
- state_dir = xstrdup(dir_name);
return SLURM_SUCCESS;
}
extern int container_p_create(uint32_t job_id)
{
- if (enable_debug) {
- int i, empty = -1, found = -1;
- bool job_id_change = false;
-
- info("%s: creating(%u)", plugin_type, job_id);
- slurm_mutex_lock(&context_lock);
- for (i = 0; i < job_id_count; i++) {
- if (job_id_array[i] == 0) {
- empty = i;
- } else if (job_id_array[i] == job_id) {
- found = i;
- break;
- }
- }
- if (found == -1) {
- if (empty == -1) {
- empty = job_id_count;
- job_id_count += 4;
- job_id_array = xrealloc(job_id_array,
- sizeof(uint32_t)*job_id_count);
- }
- job_id_array[empty] = job_id;
- job_id_change = true;
- } else {
- info("%s: duplicate create job(%u)", plugin_type, job_id);
- }
- if (job_id_change)
- _save_state(state_dir);
- slurm_mutex_unlock(&context_lock);
- }
-
return SLURM_SUCCESS;
}
/* Add proctrack container (PAGG) to a job container */
extern int container_p_add_cont(uint32_t job_id, uint64_t cont_id)
{
- if (enable_debug) {
- /* This is called from slurmstepd, so the job_id_array is NULL
- * here.The array is only set by slurmstepd */
- info("%s: adding cont(%u.%"PRIu64")", plugin_type, job_id,
- cont_id);
- }
-
return SLURM_SUCCESS;
}
/* Add a process to a job container, create the proctrack container to add */
extern int container_p_add_pid(uint32_t job_id, pid_t pid, uid_t uid)
{
- if (enable_debug) {
- stepd_step_rec_t job;
-
- info("%s: adding pid(%u.%u)", plugin_type, job_id,
- (uint32_t) pid);
-
- memset(&job, 0, sizeof(stepd_step_rec_t));
- job.jmgr_pid = pid;
- job.uid = uid;
- if (proctrack_g_create(&job) != SLURM_SUCCESS) {
- error("%s: proctrack_g_create job(%u)",
- plugin_type, job_id);
- return SLURM_ERROR;
- }
-
- proctrack_g_add(&job, pid);
-
- return container_p_add_cont(job_id, job.cont_id);
- }
-
return SLURM_SUCCESS;
}
extern int container_p_delete(uint32_t job_id)
{
- if (enable_debug) {
- int i, found = -1;
- bool job_id_change = false;
-
- info("%s: deleting(%u)", plugin_type, job_id);
- slurm_mutex_lock(&context_lock);
- for (i = 0; i < job_id_count; i++) {
- if (job_id_array[i] == job_id) {
- job_id_array[i] = 0;
- job_id_change = true;
- found = i;
- }
- }
- if (found == -1)
- info("%s: no job for delete(%u)", plugin_type, job_id);
- if (job_id_change)
- _save_state(state_dir);
- slurm_mutex_unlock(&context_lock);
- }
-
return SLURM_SUCCESS;
}
diff --git a/src/plugins/proctrack/pgid/Makefile.in b/src/plugins/proctrack/pgid/Makefile.in
index ca379238f6c..c4a0639800f 100644
--- a/src/plugins/proctrack/pgid/Makefile.in
+++ b/src/plugins/proctrack/pgid/Makefile.in
@@ -23,15 +23,15 @@ am__make_dryrun = \
am__dry=no; \
case $$MAKEFLAGS in \
*\\[\ \ ]*) \
- echo 'am--echo: ; @echo "AM" OK' | $(MAKE) -f - 2>/dev/null \
- | grep '^AM OK$$' >/dev/null || am__dry=yes;; \
+ echo 'am--echo: ; @echo "AM" OK' | $(MAKE) -f - 2>/dev/null \
+ | grep '^AM OK$$' >/dev/null || am__dry=yes;; \
*) \
- for am__flg in $$MAKEFLAGS; do \
- case $$am__flg in \
- *=*|--*) ;; \
- *n*) am__dry=yes; break;; \
- esac; \
- done;; \
+ for am__flg in $$MAKEFLAGS; do \
+ case $$am__flg in \
+ *=*|--*) ;; \
+ *n*) am__dry=yes; break;; \
+ esac; \
+ done;; \
esac; \
test $$am__dry = yes; \
}
@@ -129,7 +129,7 @@ am__uninstall_files_from_dir = { \
test -z "$$files" \
|| { test ! -d "$$dir" && test ! -f "$$dir" && test ! -r "$$dir"; } \
|| { echo " ( cd '$$dir' && rm -f" $$files ")"; \
- $(am__cd) "$$dir" && rm -f $$files; }; \
+ $(am__cd) "$$dir" && rm -f $$files; }; \
}
am__installdirs = "$(DESTDIR)$(pkglibdir)"
LTLIBRARIES = $(pkglib_LTLIBRARIES)
@@ -410,7 +410,7 @@ $(srcdir)/Makefile.in: @MAINTAINER_MODE_TRUE@ $(srcdir)/Makefile.am $(am__confi
case '$(am__configure_deps)' in \
*$$dep*) \
( cd $(top_builddir) && $(MAKE) $(AM_MAKEFLAGS) am--refresh ) \
- && { if test -f $@; then exit 0; else break; fi; }; \
+ && { if test -f $@; then exit 0; else break; fi; }; \
exit 1;; \
esac; \
done; \
@@ -467,7 +467,7 @@ clean-pkglibLTLIBRARIES:
echo "rm -f \"$${dir}/so_locations\""; \
rm -f "$${dir}/so_locations"; \
done
-proctrack_pgid.la: $(proctrack_pgid_la_OBJECTS) $(proctrack_pgid_la_DEPENDENCIES) $(EXTRA_proctrack_pgid_la_DEPENDENCIES)
+proctrack_pgid.la: $(proctrack_pgid_la_OBJECTS) $(proctrack_pgid_la_DEPENDENCIES) $(EXTRA_proctrack_pgid_la_DEPENDENCIES)
$(proctrack_pgid_la_LINK) -rpath $(pkglibdir) $(proctrack_pgid_la_OBJECTS) $(proctrack_pgid_la_LIBADD) $(LIBS)
mostlyclean-compile:
--
GitLab