diff --git a/NEWS b/NEWS
index 34a3050cf20fa91c4fe03d4c05de5fc2e2c6b6b5..2f1d12b78e46244b292e4c0cbe7790ed4fab9912 100644
--- a/NEWS
+++ b/NEWS
@@ -111,6 +111,13 @@ documents those changes that are of interest to users and administrators.
-- job_submit/lua - expose admin_comment field.
-- Allow AdminComment field to be set by the job_submit plugin.
-- Allow AdminComment field to be changed by any Administrator.
+ -- Fix key words in jobcomp select.
+ -- MYSQL - Streamline job flush sql when doing a clean start on the slurmctld.
+ -- Fix potential infinite loop when talking to the DBD when shutting down
+ the slurmctld.
+ -- Fix MCS filter.
+ -- Make it so pmix can be included in the plugin rpm without having to
+ specify --with-pmix.
* Changes in Slurm 17.02.1-2
============================
diff --git a/doc/html/mcs.shtml b/doc/html/mcs.shtml
index fb003b146706d534ee179ce05e8989ab2fa3d1ac..cc9da0391c8ae3d97435f5f76caa2fe9ecaefd06 100644
--- a/doc/html/mcs.shtml
+++ b/doc/html/mcs.shtml
@@ -2,60 +2,168 @@
<h1>Multi-Category Security (MCS) Guide</h1>
-<h2>MCS Overview</h2>
-<p>Slurm can be configured to associate a category label to jobs and optionally
-ensure that nodes can only be shared among jobs having the same category label.
-Job and node information can optionally be filtered based on their MCS labels
-in coordination with the PrivateData option:
-only users having access to the associated MCS label will have access
-to the information. The MCS plugin is responsible for these features.</p>
-
-<p>Users may either request a particular category label for a job,
-or use the default value generated by the MCS plugin implementation.
-The MCS plugin is responsible for checking that the user provided label
-is valid for the user.</p>
-
-<p>MCS labels can be either enforced or specified on demand on jobs.
-When set to ondemand, MCS label will only be set when users specify a valid one
-at submission time.
-It is the responsibility of the MCS plugin to validate the correctness
-of the requested labels.
-When enforced, the MCS plugin implementation will always associate
-the default MCS label of users to their jobs unless users specify another
-valid one.</p>
-
-<p>The selection of nodes can be filtered on MCS labels :
-on demand (ondemand) or always (select) or never (noselect).
-User can force the filter with --exclusive=mcs option (except if noselect mode).</p>
-
-<p>The MCS category label (also called MCS label) for a job is shown in
-squeue with the format option mcslabel.
-The node's inherited MCS label is shown with scontrol show nodes.
-The sview command can also be used to see those MCS labels.</p>
-
-<p>The following configuration parameters are available:</p>
+<h2>Overview</h2>
+<p>The MCS Plugin is meant to extend the current Slurm functionality related to
+job nodes <b>exclusivity</b> and job/node information display <b>privacy</b>.
+
+<p>Slurm <b>OverSubscribe</b> option controls the ability of a partition to
+execute more than one job at a time on each resource, no matter "what type"
+of job. Slurm job submission clients can also use the <b>--exclusive</b> and
+<b>--oversubscribe</b> parameters to request how the job can be
+<a href="cons_res_share.html">shared</a>. In Slurm 15.08, <b>ExclusiveUser</b>
+slurm.conf parameter and the --exclusive=<b>user</b> client parameter value
+were introduced, extending the original exclusivity functionality. With this
+parameter enabled, the "type of job" now matters when considering exclusivity,
+so that now jobs can share resources <b>based on</b> job users, meaning that
+only jobs whose user is the same can share resources. This indeed added a new
+dimension to how Slurm manages exclusivity. With the introduction of the MCS
+Plugin, Slurm can now be configured to associate a <b>MCS_label</b> to jobs
+and optionally ensure that nodes can only be shared among jobs having the same
+label. This adds even more degrees of freedom to how Slurm manages exclusivity,
+providing end users with way much more flexibility on this area.</p>
+
+<p>Slurm also has the <b>PrivateData</b> <a href="slurm.conf.html">slurm.conf</a>
+parameter, which is used to control what type of information is hidden from
+regular users. Similarly to the exclusivity property, the MCS Plugin also
+extends the <b>privacy</b> one by filtering jobs and/or nodes information
+based on the users access to their <b>MCS_label</b>. This means that privacy
+is now less restrictive, and information is not just hidden or not to regular
+users, but now it is filtered <b>depending</b> on these configurable/requestable
+labels in coordination with the PrivateData option.
+
+<h2>Configuration</h2>
+<p>Two parameters are currently available to configure MCS: <b>MCSPlugin</b>
+and <b>MCSParameters</b>.
+
+<ul>
+<li><b>MCSPlugin</b> Specifies which plugin should be used. Plugins are
+mutually exclusive, and the type of label to be associated depends
+on the loaded plugin.</li><br>
+<ul>
+<li><b>mcs/none</b> is the default and disables MCS labels and functionality.
+</li><br>
+<li><b>mcs/account</b> MCS labels can only take a value equal to the job's
+--account. NOTE: this option requires enabling of accounting.
+</li><br>
+<li><b>mcs/group</b> MCS labels can only take a value equal to the job's user
+group.
+</li><br>
+<li><b>mcs/user</b> MCS labels can only take a value equal to the username of
+the job's --uid.
+</li><br>
+</ul>
+</ul>
+
+<p>MCS_labels on jobs can be displayed through 'squeue' with the format option
+<b>mcslabel</b> or through 'scontrol show job'. Nodes also acquire a MCS_label,
+which is inherited from the allocated job's MCS_label. The nodes label can be
+displayed with 'scontrol show nodes'. The 'sview' command can also be used to
+see these MCS_labels.</p>
+
+<p>Users may either request a particular category label for a job (through the
+<b>--mcs-label</b> option), or use the default value generated by the specific
+MCS plugin implementation. So labels can be configured to be enforced or set
+on demand, and the specific MCS Plugin is responsible for checking the validity
+of these labels. When enforced, the MCS Plugin implementation will always
+associate a MCS label to a submited job, either the default value or the one
+requested by the user (if it's considered correct).</p>
+
+<p>The selection (exclusivity) of nodes can be filtered on MCS labels either
+on demand (ondemand) or always (select) or never (noselect). User can force the
+filter with <b>--exclusive=mcs</b> option (except if noselect mode is enabled).
+</p>
+
+<p>Label enforcement, node selection filtering policy, private data based on
+labels and a list of user groups allowed to be mapped to MCS labels can be
+configured through MCSParameters option.</p>
+
<ul>
-<li><b>MCSPlugin:</b>
-Specifies which plugin should be used.
+<li><b>MCSParameters</b> Specifies the options to pass to the specific MCS
+Plugin implementation. Options should satisfy the following expression:
+<br>
+"[ondemand|enforced][,noselect|select|ondemandselect][,privatedata]:[mcs_plugin_parameters]".
+The defaults are "ondemand,ondemandselect" and no privatedata.
</li>
-<li><b>MCSParameters:</b>
-Specifies options to pass to the MCS plugin implementation.
-The string is of the form:<br>
-"[ondemand|enforced][,noselect|,select|,ondemandselect][,privatedata]:[mcs_plugin_parameters]"<br>
-The defaults are "ondemand,ondemandselect" and no privatedata</li>
+<br>
<ul>
-<li>[ondemand|enforced]: set MCS label on jobs either on demand (using --mcs-label option) or always</li>
-<li>[,noselect|,select|,ondemandselect]: select nodes with filter on MCS label : never,
-always or on demand (using --exclusive=mcs option)</li>
-<li>[,privatedata]: accordingly with privatedata option:<br>
-if privatedata and privatedata=jobs: jobs informations are filtered based on their MCS labels<br>
-if privatedata and privatedata=nodes: nodes informations are filtered based on their MCS labels<br>
-Only mcs/group is currently supporting the mcs_plugin_parameters option.
-It can be used to specify the list of user groups (separated by |) that can be mapped
-to MCS labels by the mcs/group plugin.</li>
+<li><b>ondemand|enforced</b> set MCS label on jobs either on demand (using
+--mcs-label option) or always</li>
+<li><b>noselect|select|ondemandselect</b> select nodes with filter on MCS
+label: never, always or on demand (using --exclusive=mcs option)</li>
+<li><b>privatedata</b> accordingly with PrivateData specific option:<br>
+<ul>
+<li>
+if MCS' privatedata and PrivateData's privatedata=jobs, jobs information is
+filtered based on their MCS labels<br>
+</li>
+<li>
+if MCS' privatedata and PrivateData's privatedata=nodes, nodes informations is
+filtered based on their MCS labels<br>
+</li>
</ul>
+<li><b>mcs_plugin_parameters</b> Only mcs/group is currently supporting this
+option. It can be used to specify the list of user groups (separated by the
+'<b>|</b>' character) that are allowed to be mapped to MCS labels by the
+mcs/group plugin.</li>
</li>
</ul>
+</li>
+</ul>
+
+Different requests and configurations lead to different combinations of
+use-cases. The following table is intended to help understand the end user
+the expected behavior (related to exclusivity) for a subset of these use-cases:
+<br>
+<br>
+
+<table style="page-break-inside: avoid; font-family: Arial,Helvetica,sans-serif;" border="1" bordercolor="#000000" cellpadding="6" cellspacing="0" width="100%">
+ <tr>
+ <td bgcolor="#e0e0e0">
+ Node filtering:
+ </td>
+ <td bgcolor="#e0e0e0">
+ Label enforcement: <b>ondemand</b><br>
+ (MCS_label set only if requested.)
+ </td>
+ <td bgcolor="#e0e0e0">
+ Label enforcement: <b>enforced</b><br>
+ (MCS_label is mandatory.)
+ </td>
+ </tr>
+ <tr>
+ <td bgcolor="#e0e0e0">
+ <b>noselect</b>
+ </td>
+ <td>
+ No filter on nodes even if --excluisve=mcs requested.
+ </td>
+ <td>
+ No filter on nodes even if --excluisve=mcs requested.
+ </td>
+ </tr>
+ <tr>
+ <td bgcolor="#e0e0e0">
+ <b>select</b>
+ </td>
+ <td>
+ Filter on nodes <b>only</b> if job MCS_label is set.
+ </td>
+ <td>
+ Always filter on nodes.
+ </td>
+ </tr>
+ <tr>
+ <td bgcolor="#e0e0e0">
+ <b>ondemandselect</b>
+ </td>
+ <td>
+ Filter on nodes <b>only</b> if --exclusive=mcs.
+ </td>
+ <td>
+ Filter on nodes <b>only</b> if --exclusive=mcs.
+ </td>
+ </tr>
+</table>
<h2>Some examples</h2>
diff --git a/slurm.spec b/slurm.spec
index 6bb3014d7eca1162920c2b8794a2a7eec40525e7..41a4957bebf54e0844e13e7ddd3b67fd350c73d0 100644
--- a/slurm.spec
+++ b/slurm.spec
@@ -16,7 +16,6 @@
# --with cray_alps %_with_cray_alps 1 build for a Cray system with ALPS
# --with cray_network %_with_cray_network 1 build for a non-Cray system with a Cray network
# --without debug %_without_debug 1 don't compile with debugging symbols
-# --with pmix %_with_pmix 1 build pmix support
# --with lua %_with_lua 1 build Slurm lua bindings (proctrack only for now)
# --without munge %_without_munge path don't build auth-munge RPM
# --with mysql %_with_mysql 1 require mysql/mariadb support
@@ -46,7 +45,6 @@
%slurm_without_opt cray_network
%slurm_without_opt salloc_background
%slurm_without_opt multiple_slurmd
-%slurm_without_opt pmix
# These options are only here to force there to be these on the build.
# If they are not set they will still be compiled if the packages exist.
@@ -397,7 +395,7 @@ according to the Slurm
%{?slurm_with_salloc_background:--enable-salloc-background} \
%{!?slurm_with_readline:--without-readline} \
%{?slurm_with_multiple_slurmd:--enable-multiple-slurmd} \
- %{?slurm_with_pmix:--with-pmix=%{?with_pmix_dir}} \
+ %{?slurm_with_pmix:--with-pmix=%{?slurm_with_pmix}} \
%{?with_freeipmi:--with-freeipmi=%{?with_freeipmi}}\
%{?slurm_with_shared_libslurm:--with-shared-libslurm}\
%{?with_cflags} \
@@ -666,6 +664,12 @@ test -f $RPM_BUILD_ROOT/%{_libdir}/slurm/launch_aprun.so &&
echo %{_libdir}/slurm/launch_aprun.so >> $LIST
test -f $RPM_BUILD_ROOT/%{_libdir}/slurm/mpi_mvapich.so &&
echo %{_libdir}/slurm/mpi_mvapich.so >> $LIST
+test -f $RPM_BUILD_ROOT/%{_libdir}/slurm/mpi_pmix.so &&
+ echo %{_libdir}/slurm/mpi_pmix.so >> $LIST
+test -f $RPM_BUILD_ROOT/%{_libdir}/slurm/mpi_pmix_v1.so &&
+ echo %{_libdir}/slurm/mpi_pmix_v1.so >> $LIST
+test -f $RPM_BUILD_ROOT/%{_libdir}/slurm/mpi_pmix_v2.so &&
+ echo %{_libdir}/slurm/mpi_pmix_v2.so >> $LIST
test -f $RPM_BUILD_ROOT/%{_libdir}/slurm/node_features_knl_cray.so &&
echo %{_libdir}/slurm/node_features_knl_cray.so >> $LIST
test -f $RPM_BUILD_ROOT/%{_libdir}/slurm/node_features_knl_generic.so &&
@@ -881,9 +885,6 @@ rm -rf $RPM_BUILD_ROOT
%{_libdir}/slurm/mpi_pmi2.so
%endif
%{_libdir}/slurm/mpi_none.so
-%if %{slurm_with pmix}
-%{_libdir}/slurm/mpi_pmix.so
-%endif
%{_libdir}/slurm/power_none.so
%{_libdir}/slurm/preempt_job_prio.so
%{_libdir}/slurm/preempt_none.so
diff --git a/src/common/slurm_mcs.c b/src/common/slurm_mcs.c
index e1ae551ac7fcc7ce1620a155ba40f08e33896e40..cd4a692f3128cc469c51a924f6b97f83087aff6c 100644
--- a/src/common/slurm_mcs.c
+++ b/src/common/slurm_mcs.c
@@ -79,20 +79,20 @@ extern int slurm_mcs_init(void)
if (init_run && g_mcs_context)
return retval;
- slurm_mutex_lock(&g_mcs_context_lock);
+ slurm_mutex_lock(&g_mcs_context_lock);
if (g_mcs_context)
goto done;
xfree(mcs_params);
xfree(mcs_params_common);
xfree(mcs_params_specific);
-
type = slurm_get_mcs_plugin();
mcs_params = slurm_get_mcs_plugin_params();
- if (mcs_params == NULL) {
+
+ if (mcs_params == NULL)
info("No parameter for mcs plugin, default values set");
- } else {
+ else {
mcs_params_common = xstrdup(mcs_params);
sep = xstrchr(mcs_params_common, ':');
if (sep != NULL) {
@@ -101,6 +101,7 @@ extern int slurm_mcs_init(void)
*sep = '\0';
}
}
+
_slurm_mcs_check_and_load_privatedata(mcs_params_common);
_slurm_mcs_check_and_load_enforced(mcs_params_common);
_slurm_mcs_check_and_load_select(mcs_params_common);
@@ -113,8 +114,8 @@ extern int slurm_mcs_init(void)
retval = SLURM_ERROR;
goto done;
}
- init_run = true;
+ init_run = true;
done:
slurm_mutex_unlock(&g_mcs_context_lock);
xfree(type);
@@ -140,7 +141,7 @@ extern int slurm_mcs_fini(void)
extern int slurm_mcs_reconfig(void)
{
slurm_mcs_fini();
- return(slurm_mcs_init());
+ return slurm_mcs_init();
}
/* slurm_mcs_get_params_specific
@@ -157,28 +158,32 @@ extern char *slurm_mcs_get_params_specific(void)
static int _slurm_mcs_check_and_load_enforced(char *params)
{
label_strict_enforced = false;
+
if ((params != NULL) && xstrcasestr(params, "enforced"))
label_strict_enforced = true;
else
info("mcs: MCSParameters = %s. ondemand set.", params);
+
return SLURM_SUCCESS;
}
static int _slurm_mcs_check_and_load_select(char *params)
{
select_value = MCS_SELECT_ONDEMANDSELECT;
+
if (params == NULL) {
return SLURM_SUCCESS;
}
- if (xstrcasestr(params, "noselect")) {
+
+ if (xstrcasestr(params, "noselect"))
select_value = MCS_SELECT_NOSELECT;
- } else if (xstrcasestr(params, "ondemandselect")) {
+ else if (xstrcasestr(params, "ondemandselect"))
select_value = MCS_SELECT_ONDEMANDSELECT;
- } else if (xstrcasestr(params, "select")) {
+ else if (xstrcasestr(params, "select"))
select_value = MCS_SELECT_SELECT;
- } else {
+ else
info("mcs: MCSParameters = %s. ondemandselect set.", params);
- }
+
return SLURM_SUCCESS;
}
@@ -188,11 +193,12 @@ static int _slurm_mcs_check_and_load_privatedata(char *params)
private_data = false;
return SLURM_SUCCESS;
}
- if (xstrcasestr(params, "privatedata")) {
+
+ if (xstrcasestr(params, "privatedata"))
private_data = true;
- } else {
+ else
private_data = false;
- }
+
return SLURM_SUCCESS;
}
@@ -201,18 +207,19 @@ extern int slurm_mcs_reset_params(void)
label_strict_enforced = false;
select_value = MCS_SELECT_ONDEMANDSELECT;
private_data = false;
+
return SLURM_SUCCESS;
}
extern int slurm_mcs_get_enforced(void)
{
- return(label_strict_enforced);
+ return label_strict_enforced;
}
extern int slurm_mcs_get_select(struct job_record *job_ptr)
{
if ((select_value == MCS_SELECT_SELECT) ||
- ((select_value == MCS_SELECT_ONDEMANDSELECT) &&
+ ((select_value == MCS_SELECT_ONDEMANDSELECT) &&
job_ptr->details &&
(job_ptr->details->whole_node == WHOLE_NODE_MCS)))
return 1;
@@ -237,5 +244,6 @@ extern int mcs_g_check_mcs_label(uint32_t user_id, char *mcs_label)
{
if (slurm_mcs_init() < 0)
return 0;
- return (int) (*(ops.check))(user_id, mcs_label);
+
+ return (int)(*(ops.check))(user_id, mcs_label);
}
diff --git a/src/common/slurm_persist_conn.c b/src/common/slurm_persist_conn.c
index ee749c08890c404e30db7574c26ba0983e0cf984..0ddd859b8d74ae955f955ac180af5caf24f1dd73 100644
--- a/src/common/slurm_persist_conn.c
+++ b/src/common/slurm_persist_conn.c
@@ -914,11 +914,12 @@ extern Buf slurm_persist_recv_msg(slurm_persist_conn_t *persist_conn)
return buffer;
endit:
- /* Close it since we abondoned it. If the connection does still exist
+ /* Close it since we abandoned it. If the connection does still exist
* on the other end we can't rely on it after this point since we didn't
* listen long enough for this response.
*/
- if (persist_conn->flags & PERSIST_FLAG_RECONNECT)
+ if (!(*persist_conn->shutdown) &&
+ persist_conn->flags & PERSIST_FLAG_RECONNECT)
slurm_persist_conn_reopen(persist_conn, true);
return NULL;
diff --git a/src/plugins/accounting_storage/mysql/as_mysql_job.c b/src/plugins/accounting_storage/mysql/as_mysql_job.c
index 06858d18f5cd76f94aae86759ef34077ae14644c..a00a88567491175c069d8b6aaadbf3da9fa2edf1 100644
--- a/src/plugins/accounting_storage/mysql/as_mysql_job.c
+++ b/src/plugins/accounting_storage/mysql/as_mysql_job.c
@@ -1554,20 +1554,21 @@ extern int as_mysql_flush_jobs_on_cluster(
if (state == JOB_SUSPENDED) {
if (suspended_char)
xstrfmtcat(suspended_char,
- " || job_db_inx=%s", row[0]);
+ ", %s", row[0]);
else
- xstrfmtcat(suspended_char, "job_db_inx=%s",
+ xstrfmtcat(suspended_char, "job_db_inx in (%s",
row[0]);
}
if (id_char)
- xstrfmtcat(id_char, " || job_db_inx=%s", row[0]);
+ xstrfmtcat(id_char, ", %s", row[0]);
else
- xstrfmtcat(id_char, "job_db_inx=%s", row[0]);
+ xstrfmtcat(id_char, "job_db_inx in (%s", row[0]);
}
mysql_free_result(result);
if (suspended_char) {
+ xstrfmtcat(suspended_char, ")");
xstrfmtcat(query,
"update \"%s_%s\" set "
"time_suspended=%ld-time_suspended "
@@ -1588,6 +1589,7 @@ extern int as_mysql_flush_jobs_on_cluster(
xfree(suspended_char);
}
if (id_char) {
+ xstrfmtcat(id_char, ")");
xstrfmtcat(query,
"update \"%s_%s\" set state=%d, "
"time_end=%ld where %s;",
diff --git a/src/plugins/jobcomp/mysql/mysql_jobcomp_process.c b/src/plugins/jobcomp/mysql/mysql_jobcomp_process.c
index 6ebe3fad8e4c2c0949170a7effe36702063b9dd8..8f9c92aa98c6e7f903074f2a95142d4911a7541b 100644
--- a/src/plugins/jobcomp/mysql/mysql_jobcomp_process.c
+++ b/src/plugins/jobcomp/mysql/mysql_jobcomp_process.c
@@ -107,7 +107,7 @@ extern List mysql_jobcomp_process_get_jobs(slurmdb_job_cond_t *job_cond)
while(jobcomp_table_fields[i].name) {
if (i)
xstrcat(tmp, ", ");
- xstrcat(tmp, jobcomp_table_fields[i].name);
+ xstrfmtcat(tmp, "`%s`", jobcomp_table_fields[i].name);
i++;
}
diff --git a/src/plugins/mcs/account/mcs_account.c b/src/plugins/mcs/account/mcs_account.c
index 3496742c5361d060d01959671ff271707da45ebb..e8c78734523fdf476e7164a7e8dae16496861005 100644
--- a/src/plugins/mcs/account/mcs_account.c
+++ b/src/plugins/mcs/account/mcs_account.c
@@ -98,21 +98,21 @@ extern int mcs_p_set_mcs_label(struct job_record *job_ptr, char *label)
{
int rc = SLURM_SUCCESS;
xfree(job_ptr->mcs_label);
+
if (label != NULL) {
/* test label param */
- if (!xstrcmp(label, job_ptr->account)) {
+ if (!xstrcmp(label, job_ptr->account))
job_ptr->mcs_label = xstrdup(job_ptr->account);
- } else {
+ else
rc = SLURM_ERROR;
- }
} else {
if ((slurm_mcs_get_enforced() == 0) && job_ptr->details &&
- (job_ptr->details->whole_node != WHOLE_NODE_MCS)) {
+ (job_ptr->details->whole_node != WHOLE_NODE_MCS))
;
- } else {
+ else
job_ptr->mcs_label = xstrdup(job_ptr->account);
- }
}
+
return rc;
}
@@ -127,14 +127,16 @@ extern int mcs_p_check_mcs_label(uint32_t user_id, char *mcs_label)
memset(&assoc_rec, 0, sizeof(slurmdb_assoc_rec_t));
assoc_rec.acct = mcs_label;
assoc_rec.uid = user_id;
+
if (mcs_label != NULL) {
if (!assoc_mgr_fill_in_assoc(acct_db_conn, &assoc_rec,
- accounting_enforce,
- (slurmdb_assoc_rec_t **)
- NULL, false))
+ accounting_enforce,
+ (slurmdb_assoc_rec_t **) NULL,
+ false))
rc = SLURM_SUCCESS;
else
rc = SLURM_ERROR;
}
+
return rc;
}
diff --git a/src/plugins/mcs/group/mcs_group.c b/src/plugins/mcs/group/mcs_group.c
index bbf3e84c9ceba5f2e9fba63f36ba69fe77766af2..7dc77b2ff2e9dbfd824b4cb446b99237a8519ce8 100644
--- a/src/plugins/mcs/group/mcs_group.c
+++ b/src/plugins/mcs/group/mcs_group.c
@@ -92,12 +92,15 @@ extern int init(void)
{
debug("%s loaded", plugin_name);
mcs_params_specific = slurm_mcs_get_params_specific();
+
if (_check_and_load_params() != 0) {
- info("mcs: plugin warning : no group in %s", mcs_params_specific);
+ info("mcs: plugin warning : no group in %s",
+ mcs_params_specific);
xfree(mcs_params_specific);
/* no need to check others options : default values used */
return SLURM_SUCCESS;
}
+
xfree(mcs_params_specific);
return SLURM_SUCCESS;
}
@@ -125,6 +128,7 @@ static int _get_user_groups(uint32_t user_id, uint32_t group_id,
user_name = uid_to_string((uid_t) user_id);
*ngroups = max_groups;
rc = getgrouplist(user_name, (gid_t) group_id, groups, ngroups);
+
if (rc < 0) {
error("getgrouplist(%s): %m", user_name);
rc = SLURM_ERROR;
@@ -132,6 +136,7 @@ static int _get_user_groups(uint32_t user_id, uint32_t group_id,
*ngroups = rc;
rc = SLURM_SUCCESS;
}
+
xfree(user_name);
return rc;
}
@@ -154,15 +159,17 @@ static int _check_and_load_params(void)
slurm_mcs_reset_params();
return SLURM_ERROR;
}
+
n = strlen(mcs_params_specific);
for (i = 0 ; i < n ; i++) {
if (mcs_params_specific[i] == '|')
nb_mcs_groups = nb_mcs_groups + 1;
}
+
if (nb_mcs_groups == 0) {
/* no | in param : just one group */
if (mcs_params_specific != NULL) {
- if ( gid_from_string(mcs_params_specific, &gid ) != 0 ) {
+ if (gid_from_string(mcs_params_specific, &gid ) != 0 ) {
info("mcs: Only one invalid group : %s. "
"ondemand, ondemandselect set", groups_names);
nb_mcs_groups = 0;
@@ -190,10 +197,12 @@ static int _check_and_load_params(void)
}
return SLURM_SUCCESS;
}
+
nb_mcs_groups = nb_mcs_groups + 1;
array_mcs_parameter = xmalloc(nb_mcs_groups * sizeof(uint32_t));
tmp_params = xstrdup(mcs_params_specific);
groups_names = strtok_r(tmp_params, "|", &name_ptr);
+
i = 0;
while (groups_names) {
if (i == (nb_mcs_groups - 1)) {
@@ -213,6 +222,7 @@ static int _check_and_load_params(void)
i = i + 1;
groups_names = strtok_r(NULL, "|", &name_ptr);
}
+
/* if no valid group : deselect all params */
if (nb_valid_group == 0) {
slurm_mcs_reset_params();
@@ -220,6 +230,7 @@ static int _check_and_load_params(void)
xfree(tmp_params);
return SLURM_ERROR;
}
+
xfree(tmp_params);
return SLURM_SUCCESS;
}
@@ -236,20 +247,20 @@ static int _find_mcs_label(gid_t *groups, int ngroups, char **result)
struct group *gr;
if (ngroups == 0)
- rc = SLURM_ERROR;
- else {
- for( i = 0 ; i < nb_mcs_groups ; i++) {
- for ( j = 0 ; j < ngroups ; j++) {
- tmp_group = (uint32_t) groups[j];
- if (array_mcs_parameter[i] == tmp_group ) {
- gr = getgrgid(groups[j]);
- *result = gr->gr_name;
- return rc;
- }
+ return SLURM_ERROR;
+
+ for (i = 0; i < nb_mcs_groups; i++) {
+ for (j = 0; j < ngroups; j++) {
+ tmp_group = (uint32_t) groups[j];
+ if (array_mcs_parameter[i] == tmp_group) {
+ gr = getgrgid(groups[j]);
+ *result = gr->gr_name;
+ return rc;
}
}
- rc = SLURM_ERROR;
}
+ rc = SLURM_ERROR;
+
return rc;
}
@@ -266,7 +277,7 @@ static int _check_mcs_label (struct job_record *job_ptr, char *label)
int ngroups = -1;
/* test if real unix group */
- if ( gid_from_string(label, &gid ) != 0 )
+ if (gid_from_string(label, &gid ) != 0)
return rc;
/* test if this group is owned by the user */
@@ -274,6 +285,7 @@ static int _check_mcs_label (struct job_record *job_ptr, char *label)
groups, MAX_GROUPS, &ngroups);
if (rc) /* Failed to get groups */
return rc;
+
rc = SLURM_ERROR;
for (i = 0; i < ngroups; i++) {
tmp_group = (uint32_t) groups[i];
@@ -282,16 +294,19 @@ static int _check_mcs_label (struct job_record *job_ptr, char *label)
break;
}
}
+
if (rc == SLURM_ERROR)
return rc;
+
rc = SLURM_ERROR;
/* test if mcs_label is in list of possible mcs_label */
- for( i = 0 ; i < nb_mcs_groups ; i++) {
- if (array_mcs_parameter[i] == gid ) {
+ for (i = 0; i < nb_mcs_groups; i++) {
+ if (array_mcs_parameter[i] == gid) {
rc = SLURM_SUCCESS;
return rc;
}
}
+
return rc;
}
@@ -311,6 +326,7 @@ extern int mcs_p_set_mcs_label (struct job_record *job_ptr, char *label)
if ((slurm_mcs_get_enforced() == 0) && job_ptr->details &&
(job_ptr->details->whole_node != WHOLE_NODE_MCS))
return SLURM_SUCCESS;
+
rc = _get_user_groups(job_ptr->user_id,job_ptr->group_id,
groups, MAX_GROUPS, &ngroups);
if (rc) { /* Failed to get groups */
@@ -319,6 +335,7 @@ extern int mcs_p_set_mcs_label (struct job_record *job_ptr, char *label)
else
return SLURM_ERROR;
}
+
rc = _find_mcs_label(groups, ngroups, &result);
if (rc) {
return SLURM_ERROR;
@@ -351,9 +368,9 @@ extern int mcs_p_check_mcs_label (uint32_t user_id, char *mcs_label)
if (mcs_label != NULL) {
/* test if real unix group */
- if ( gid_from_string(mcs_label, &gid ) != 0 ) {
+ if (gid_from_string(mcs_label, &gid ) != 0)
return rc;
- }
+
/* test if this group is owned by the user */
slurm_user_gid = gid_from_uid(user_id);
group_id = (uint32_t) slurm_user_gid;
@@ -361,6 +378,7 @@ extern int mcs_p_check_mcs_label (uint32_t user_id, char *mcs_label)
&ngroups);
if (rc) /* Failed to get groups */
return rc;
+
rc = SLURM_ERROR;
for (i = 0; i < ngroups; i++) {
tmp_group = (uint32_t) groups[i];
@@ -369,8 +387,8 @@ extern int mcs_p_check_mcs_label (uint32_t user_id, char *mcs_label)
break;
}
}
- } else {
+ } else
rc = SLURM_SUCCESS;
- }
+
return rc;
}
diff --git a/src/plugins/mcs/user/mcs_user.c b/src/plugins/mcs/user/mcs_user.c
index 5d5a71991e5feb2d40634ee5c61d3cf5170e546d..2ff26736f70fb1d8ae3b2fed0f1879209ba3110c 100644
--- a/src/plugins/mcs/user/mcs_user.c
+++ b/src/plugins/mcs/user/mcs_user.c
@@ -95,23 +95,24 @@ extern int mcs_p_set_mcs_label (struct job_record *job_ptr, char *label)
{
char *user = NULL;
int rc = SLURM_SUCCESS;
+
user = uid_to_string((uid_t) job_ptr->user_id);
xfree(job_ptr->mcs_label);
+
if (label != NULL) {
/* test label param */
- if (xstrcmp(label, user) == 0) {
+ if (xstrcmp(label, user) == 0)
job_ptr->mcs_label = xstrdup(user);
- } else {
+ else
rc = SLURM_ERROR;
- }
} else {
if ((slurm_mcs_get_enforced() == 0) && job_ptr->details &&
- (job_ptr->details->whole_node != WHOLE_NODE_MCS)) {
+ (job_ptr->details->whole_node != WHOLE_NODE_MCS))
;
- } else {
+ else
job_ptr->mcs_label = xstrdup(user);
- }
}
+
xfree(user);
return rc;
}
@@ -123,16 +124,16 @@ extern int mcs_p_check_mcs_label (uint32_t user_id, char *mcs_label)
{
char *user = NULL;
int rc = SLURM_SUCCESS;
+
user = uid_to_string((uid_t) user_id);
if (mcs_label != NULL) {
- if (xstrcmp(mcs_label, user) == 0) {
+ if (xstrcmp(mcs_label, user) == 0)
rc = SLURM_SUCCESS;
- } else {
+ else
rc = SLURM_ERROR;
- }
- } else {
+ } else
rc = SLURM_SUCCESS;
- }
+
xfree(user);
return rc;
}
diff --git a/src/slurmctld/node_scheduler.c b/src/slurmctld/node_scheduler.c
index 848d127307cc5440ca0117a80275dd1676ca4e67..cf0fab0b2f8c901ac3205c9101139d485cfcf71f 100644
--- a/src/slurmctld/node_scheduler.c
+++ b/src/slurmctld/node_scheduler.c
@@ -941,11 +941,8 @@ extern void filter_by_node_mcs(struct job_record *job_ptr, int mcs_select,
struct node_record *node_ptr;
int i;
- if (mcs_select != 1)
- return;
-
/* Need to filter out any nodes allocated with other mcs */
- if (job_ptr->mcs_label != NULL) {
+ if (job_ptr->mcs_label && (mcs_select == 1)) {
for (i = 0, node_ptr = node_record_table_ptr;
i < node_record_count; i++, node_ptr++) {
/* if there is a mcs_label -> OK if it's the same */
diff --git a/testsuite/expect/test17.51 b/testsuite/expect/test17.51
index 3ea77d2b361e4aaa67a6372647a3a1ad463bf03d..693a68d21692fed44f2bde4aca48cdf0740af6af 100755
--- a/testsuite/expect/test17.51
+++ b/testsuite/expect/test17.51
@@ -116,10 +116,11 @@ if {[is_super_user] == 0} {
get_conf_path
copy_conf
-send_user "\n---Checking sbatch uses mcs-label only for some jobs (ondemand mode)---\n"
+send_user "\n---Checking sbatch uses mcs-label only for some jobs (ondemand,select mode)---\n"
#
# Change the slurm.conf MCSparameters and MCSPlugin
+# test with ondemand,select
#
exec $bin_sed -i /^\[\t\s\]*MCSPlugin\[\t\s\]*=/Id $config_path/slurm.conf
exec $bin_sed -i /^\[\t\s\]*MCSParameters\[\t\s\]*=/Id $config_path/slurm.conf
@@ -158,9 +159,9 @@ if {$found == 0} {
send_user "\n---Checking sbatch fails with a bad mcs-label ---\n"
set timeout $max_job_delay
-make_bash_script $tmp_job "sleep 10"
+make_bash_script $tmp_job "sleep 30"
-spawn $sbatch -N1 --mcs-label=foo -t1 $tmp_job
+spawn $sbatch -N1 --mcs-label=foo -t10 $tmp_job
expect {
-re "Batch job submission failed: Invalid mcs_label specified" {
send_user "\nThis error is expected, no worries\n"
@@ -179,9 +180,9 @@ expect {
###### Check that sbatch uses mcs-label=user ######
send_user "\n---Checking sbatch uses mcs-label=user---\n"
-make_bash_script $tmp_job "sleep 10"
+make_bash_script $tmp_job "sleep 30"
set job_id 0
-spawn $sbatch -N1 -o/dev/null --exclusive=mcs -t1 $tmp_job
+spawn $sbatch -N1 -o/dev/null --exclusive=mcs -t10 $tmp_job
expect {
-re "Submitted batch job ($number)" {
set job_id $expect_out(1,string)
@@ -331,9 +332,9 @@ cancel_job $job_id
#
# Change the slurm.conf MCSparameters and MCSPlugin
-# test with enforced
+# test with enforced,noselect
#
-send_user "\n---Checking sbatch uses mcs-label with all jobs (enforced mode)---\n"
+send_user "n---Checking sbatch uses mcs-label with all jobs (enforced,noselect mode)---\n"
exec $bin_sed -i /^\[\t\s\]*MCSPlugin\[\t\s\]*=/Id $config_path/slurm.conf
exec $bin_sed -i /^\[\t\s\]*MCSParameters\[\t\s\]*=/Id $config_path/slurm.conf
exec $bin_sed -i /^\[\t\s\]*PrivateData\[\t\s\]*=/Id $config_path/slurm.conf
@@ -346,7 +347,7 @@ update_conf
###### Check that sbatch uses mcs-label=user ######
send_user "\n---Checking sbatch uses mcs-label=user---\n"
-make_bash_script $tmp_job "sleep 10"
+make_bash_script $tmp_job "sleep 30"
spawn $sbatch -N1 -o/dev/null -t1 $tmp_job
expect {
@@ -443,6 +444,291 @@ expect {
cancel_job $job_id
+#
+# Change the slurm.conf MCSparameters and MCSPlugin
+# test with ondemand,noselect
+#
+send_user "\n---Checking sbatch doesn't use mcs-label on filter (ondemand,noselect mode)---\n"
+exec $bin_sed -i /MCSPlugin/d $config_path/slurm.conf
+exec $bin_sed -i /MCSParameters/d $config_path/slurm.conf
+exec $bin_sed -i /PrivateData/d $config_path/slurm.conf
+exec $bin_echo MCSPlugin=mcs/user >> $config_path/slurm.conf
+exec $bin_echo MCSParameters=ondemand,noselect,privatedata >> $config_path/slurm.conf
+exec $bin_echo PrivateData=jobs,nodes >> $config_path/slurm.conf
+update_conf
+
+###### Check that sbatch uses mcs-label=user ######
+send_user "\n---Checking sbatch uses --exclusive=mcs ---\n"
+
+make_bash_script $tmp_job "sleep 30"
+
+spawn $sbatch -N1 --exclusive=mcs -o/dev/null -t10 $tmp_job
+expect {
+ -re "Submitted batch job ($number)" {
+ set job_id $expect_out(1,string)
+ exp_continue
+ }
+ timeout {
+ send_user "\nFAILURE: sbatch is not responding\n"
+ set exit_code 1
+ }
+ eof {
+ wait
+ }
+}
+
+if {$job_id == 0} {
+ send_user "\nFAILURE: job was not submitted\n"
+ set exit_code 1
+}
+
+set found 0
+sleep 3
+spawn $squeue --jobs=$job_id --noheader -O "mcslabel"
+expect {
+ -re "$user_name" {
+ send_user "\nMCS-label OK for this job\n"
+ set found 1
+ exp_continue
+ }
+ -re "(null)" {
+ send_user "\nFAILURE: NO MCS-label for this job\n"
+ set exit_code 1
+ exp_continue
+ }
+ timeout {
+ send_user "\nFAILURE: squeue is not responding\n"
+ set exit_code 1
+ }
+ eof {
+ wait
+ }
+}
+
+if {$found == 0} {
+ send_user "\nFAILURE: job was submitted with a bad mcs-label\n"
+ set exit_code 1
+}
+
+set found 0
+set node 0
+spawn $squeue --jobs=$job_id --noheader -O "nodelist"
+expect {
+ -re "($alpha_numeric_under)" {
+ set node $expect_out(1,string)
+ send_user "\nNode for this job : $node\n"
+ set found 1
+ }
+ timeout {
+ send_user "\nFAILURE: squeue is not responding\n"
+ set exit_code 1
+ }
+ eof {
+ wait
+ }
+}
+
+if {$found == 0} {
+ send_user "\nFAILURE: no node found in squeue command\n"
+ set exit_code 1
+}
+
+#
+# verify MCS of nodes
+#
+set found 0
+spawn -noecho $bin_bash -c "exec $scontrol show node=$node | $bin_grep MCS"
+expect {
+ -re "MCS_label=$user_name" {
+ send_user "\nFAILURE: a mcs_label is found for this job. It was not expected\n"
+ set exit_code 1
+ }
+ -re "MCS_label=N/A" {
+ send_user "\nNo mcs_label for this node. It was expected\n"
+ }
+ timeout {
+ send_user "\nFAILURE: scontrol is not responding\n"
+ set exit_code 1
+ }
+ eof {
+ wait
+ }
+}
+
+cancel_job $job_id
+
+###### Check that sbatch doesn't use mcs-label ######
+send_user "\n---Checking sbatch uses --exclusive=mcs ---\n"
+
+make_bash_script $tmp_job "sleep 30"
+set job_id 0
+spawn $sbatch -N1 -o/dev/null -t10 $tmp_job
+expect {
+ -re "Submitted batch job ($number)" {
+ set job_id $expect_out(1,string)
+ exp_continue
+ }
+ timeout {
+ send_user "\nFAILURE: sbatch is not responding\n"
+ set exit_code 1
+ }
+ eof {
+ wait
+ }
+}
+
+if {$job_id == 0} {
+ send_user "\nFAILURE: job was not submitted\n"
+ set exit_code 1
+}
+set found 0
+
+set found 0
+spawn $squeue --jobs=$job_id --noheader -O "mcslabel"
+expect {
+ -re "(null)" {
+ send_user "\nNO MCS-label for this job : this is expected\n"
+ exp_continue
+ }
+ -re "$user_name" {
+ send_user "\na MCS-label for this job : this is not expected\n"
+ set found 1
+ exp_continue
+ }
+ timeout {
+ send_user "\nFAILURE: squeue is not responding\n"
+ set exit_code 1
+ }
+ eof {
+ wait
+ }
+}
+
+if {$found == 1} {
+ send_user "\nFAILURE: job was submitted with a bad mcs-label\n"
+ set exit_code 1
+}
+
+cancel_job $job_id
+sleep 5
+send_user "\n---Checking sbatch uses mcs-label for all jobs (enforced,select mode)---\n"
+
+#
+# Change the slurm.conf MCSparameters and MCSPlugin
+# test with enforced,select
+#
+exec $bin_sed -i /MCSPlugin/d $config_path/slurm.conf
+exec $bin_sed -i /MCSParameters/d $config_path/slurm.conf
+exec $bin_sed -i /PrivateData/d $config_path/slurm.conf
+exec $bin_echo MCSPlugin=mcs/user >> $config_path/slurm.conf
+exec $bin_echo MCSParameters=enforced,select,privatedata >> $config_path/slurm.conf
+exec $bin_echo PrivateData=jobs,nodes >> $config_path/slurm.conf
+update_conf
+
+###### Check that sbatch uses mcs-label=user ######
+send_user "\n---Checking sbatch with no --exclusive=mcs ---\n"
+
+make_bash_script $tmp_job "sleep 30"
+set job_id 0
+spawn $sbatch -N1 -o/dev/null -t20 $tmp_job
+expect {
+ -re "Submitted batch job ($number)" {
+ set job_id $expect_out(1,string)
+ exp_continue
+ }
+ timeout {
+ send_user "\nFAILURE: sbatch is not responding\n"
+ set exit_code 1
+ }
+ eof {
+ wait
+ }
+}
+
+if {$job_id == 0} {
+ send_user "\nFAILURE: job was not submitted\n"
+ set exit_code 1
+ exit 1
+}
+
+set found 0
+spawn $squeue --jobs=$job_id --noheader -O "mcslabel"
+expect {
+ -re "(null)" {
+ send_user "\nNO MCS-label for this job : this is not expected\n"
+ exp_continue
+ }
+ -re "$user_name" {
+ send_user "\nMCS-label OK for this job\n"
+ set found 1
+ exp_continue
+ }
+ timeout {
+ send_user "\nFAILURE: squeue is not responding\n"
+ set exit_code 1
+ }
+ eof {
+ wait
+ }
+}
+
+if {$found == 0} {
+ send_user "\nFAILURE: job was submitted with a bad mcs-label\n"
+ set exit_code 1
+}
+
+set found 0
+set node 0
+sleep 2
+spawn $squeue --jobs=$job_id --noheader -O "nodelist"
+expect {
+ -re "($alpha_numeric_under)" {
+ set node $expect_out(1,string)
+ send_user "\nNode for this job : $node\n"
+ set found 1
+ }
+ timeout {
+ send_user "\nFAILURE: squeue is not responding\n"
+ set exit_code 1
+ }
+ eof {
+ wait
+ }
+}
+
+if {$found == 0} {
+ send_user "\nFAILURE: no node found in squeue command\n"
+ set exit_code 1
+}
+
+
+#
+# verify MCS of nodes
+#
+set found 0
+spawn -noecho $bin_bash -c "exec $scontrol show node=$node | $bin_grep MCS"
+expect {
+ -re "MCS_label=$user_name" {
+ send_user "\n mcs_label OK for node $node\n"
+ set found 1
+ exp_continue
+ }
+ timeout {
+ send_user "\nFAILURE: scontrol is not responding\n"
+ set exit_code 1
+ }
+ eof {
+ wait
+ }
+}
+
+if {$found == 0} {
+ send_user "\nFAILURE: job was submitted with node with bad mcs-label\n"
+ set exit_code 1
+}
+
+cancel_job $job_id
+
# Clean up vestigial files and restore original slurm.conf file
send_user "\nChanging slurm.conf back\n"
exec $bin_cp -v $cwd/slurm.conf.orig $config_path/slurm.conf