diff --git a/META b/META
index d9c132c8d2f83e98ea93d9bb4457ca3c2ecdb5ad..0892d78a8459ac3af609f023299b04da4cba3e89 100644
--- a/META
+++ b/META
@@ -11,7 +11,7 @@
Minor: 1
Micro: 0
Version: 2.1.0
- Release: 0.pre2
+ Release: 0.pre3
##
# When changing API_CURRENT update src/common/slurm_protocol_common.h
# with a new SLURM_PROTOCOL_VERSION signifing the old one and the version
diff --git a/NEWS b/NEWS
index f56bd189b6dbb07e902856bca05991e44a2138ed..139dd49a77c981a37519b969ae7cfa7b2c4805d9 100644
--- a/NEWS
+++ b/NEWS
@@ -211,6 +211,11 @@ documents those changes that are of interest to users and admins.
-- Slurm.pm loads without warning now on AIX systems
-- modified pmi code to do strncpy's on the correct len
-- Fix for filling in a qos structure to return SLURM_SUCCESS on success.
+ -- BLUEGENE - Added SLURM_BG_NUM_NODES with cnode count of allocation,
+ SLURM_JOB_NUM_NODES represents midplane counts until 2.1.
+ -- BLUEGENE - Added fix for if a block is in error state and the midplane
+ containning the block is also set to drain/down. This previously
+ prevented dynamic creation of new blocks when this state was present.
* Changes in SLURM 2.0.4
========================
diff --git a/src/common/env.c b/src/common/env.c
index 98b6d9cf6637690743cf99de3e1d1cfa7b24b4f6..6bbd90efe86eb04c81547758dfeeae8b6f150376 100644
--- a/src/common/env.c
+++ b/src/common/env.c
@@ -883,7 +883,8 @@ extern char *uint32_compressed_to_str(uint32_t array_len,
* SLURM_JOB_NODELIST
* SLURM_JOB_CPUS_PER_NODE
* LOADLBATCH (AIX only)
- * MPIRUN_PARTITION, MPIRUN_NOFREE, and MPIRUN_NOALLOCATE (BGL only)
+ * SLURM_BG_NUM_NODES, MPIRUN_PARTITION, MPIRUN_NOFREE, and
+ * MPIRUN_NOALLOCATE (BGL only)
*
* Sets OBSOLETE variables (needed for MPI, do not remove):
* SLURM_JOBID
@@ -911,6 +912,7 @@ env_array_for_job(char ***dest, const resource_allocation_response_msg_t *alloc,
&node_cnt);
if(!node_cnt)
node_cnt = alloc->node_cnt;
+ env_array_overwrite_fmt(dest, "SLURM_BG_NUM_NODES", "%u", node_cnt);
#endif
env_array_overwrite_fmt(dest, "SLURM_JOB_ID", "%u", alloc->job_id);
@@ -1070,6 +1072,9 @@ env_array_for_batch_job(char ***dest, const batch_job_launch_msg_t *batch,
env_array_overwrite_fmt(dest, "SLURM_JOB_ID", "%u", batch->job_id);
env_array_overwrite_fmt(dest, "SLURM_JOB_NUM_NODES", "%u", num_nodes);
+#ifdef HAVE_BG
+ env_array_overwrite_fmt(dest, "SLURM_BG_NUM_NODES", "%u", num_nodes);
+#endif
env_array_overwrite_fmt(dest, "SLURM_JOB_NODELIST", "%s", batch->nodes);
tmp = uint32_compressed_to_str(batch->num_cpu_groups,
diff --git a/src/plugins/select/bluegene/block_allocator/block_allocator.c b/src/plugins/select/bluegene/block_allocator/block_allocator.c
index b262acbba968f00083520f00e061bbf83257200a..3d10468dc1ce82bc2b4e53ab53e882a80d44044a 100644
--- a/src/plugins/select/bluegene/block_allocator/block_allocator.c
+++ b/src/plugins/select/bluegene/block_allocator/block_allocator.c
@@ -1223,10 +1223,11 @@ extern void ba_update_node_state(ba_node_t *ba_node, uint16_t state)
/* basically set the node as used */
if((node_base_state == NODE_STATE_DOWN)
- || (ba_node->state & NODE_STATE_DRAIN))
+ || (state & NODE_STATE_DRAIN))
ba_node->used = true;
else
ba_node->used = false;
+
ba_node->state = state;
}
@@ -1486,14 +1487,26 @@ extern int check_and_set_node_list(List nodes)
grid[ba_node->coord[X]]
[ba_node->coord[Y]]
[ba_node->coord[Z]];
+
if(ba_node->used && curr_ba_node->used) {
- debug4("I have already been to "
- "this node %c%c%c",
- alpha_num[ba_node->coord[X]],
- alpha_num[ba_node->coord[Y]],
- alpha_num[ba_node->coord[Z]]);
- rc = SLURM_ERROR;
- goto end_it;
+ /* Only error if the midplane isn't already
+ * marked down or in a error state outside of
+ * the bluegene block.
+ */
+ uint16_t base_state, node_flags;
+ base_state = curr_ba_node->state & NODE_STATE_BASE;
+ node_flags = curr_ba_node->state & NODE_STATE_FLAGS;
+ if (!(node_flags & (NODE_STATE_DRAIN | NODE_STATE_FAIL))
+ && (base_state != NODE_STATE_DOWN)) {
+ debug4("I have already been to "
+ "this node %c%c%c %s",
+ alpha_num[ba_node->coord[X]],
+ alpha_num[ba_node->coord[Y]],
+ alpha_num[ba_node->coord[Z]],
+ node_state_string(curr_ba_node->state));
+ rc = SLURM_ERROR;
+ goto end_it;
+ }
}
if(ba_node->used)
@@ -3770,7 +3783,6 @@ requested_end:
static bool _node_used(ba_node_t* ba_node, int x_size)
{
ba_switch_t* ba_switch = NULL;
-
/* if we've used this node in another block already */
if (!ba_node || ba_node->used) {
debug4("node %c%c%c used",
diff --git a/src/plugins/select/bluegene/plugin/block_sys.c b/src/plugins/select/bluegene/plugin/block_sys.c
index f0898407c52d6255df89366ab00f3c350da2b488..720e55f5a092d624fe19899acefe975348029a12 100755
--- a/src/plugins/select/bluegene/plugin/block_sys.c
+++ b/src/plugins/select/bluegene/plugin/block_sys.c
@@ -1083,7 +1083,7 @@ extern int load_state_file(List curr_block_list, char *dir_name)
#endif
slurm_mutex_lock(&block_state_mutex);
- reset_ba_system(false);
+ reset_ba_system(true);
/* Locks are already in place to protect part_list here */
bitmap = bit_alloc(node_record_count);
diff --git a/src/plugins/select/bluegene/plugin/select_bluegene.c b/src/plugins/select/bluegene/plugin/select_bluegene.c
index f56a7e1d5ca809336de53ba4ce6d9bce361feced..434972322efc526dc98fe008273dd41fc12791d3 100644
--- a/src/plugins/select/bluegene/plugin/select_bluegene.c
+++ b/src/plugins/select/bluegene/plugin/select_bluegene.c
@@ -1047,7 +1047,6 @@ extern int select_p_update_node_config (int index)
extern int select_p_update_node_state (int index, uint16_t state)
{
int x, y, z;
-
for (y = DIM_SIZE[Y] - 1; y >= 0; y--) {
for (z = 0; z < DIM_SIZE[Z]; z++) {
for (x = 0; x < DIM_SIZE[X]; x++) {
diff --git a/src/slurmctld/node_mgr.c b/src/slurmctld/node_mgr.c
index 02060e27cd19938031ce6a04601deb6e57da8d72..4ee7a54f8099aa223ac13ebb5a2618a992b8a33c 100644
--- a/src/slurmctld/node_mgr.c
+++ b/src/slurmctld/node_mgr.c
@@ -642,6 +642,20 @@ extern int load_all_node_state ( bool state_only )
node_ptr->last_idle = now;
}
xfree (node_name);
+ if(node_ptr) {
+ /* If the state is UNKNOWN we will assume IDLE
+ until the nodes check in. This is needed
+ for bluegene to set up drained nodes
+ correctly. */
+ if((node_ptr->node_state & NODE_STATE_BASE)
+ == NODE_STATE_UNKNOWN)
+ node_ptr->node_state |= NODE_STATE_IDLE;
+
+ select_g_update_node_state(
+ (node_ptr - node_record_table_ptr),
+ node_ptr->node_state);
+ }
+
}
fini: info("Recovered state of %d nodes", node_cnt);