From 85fc77f3d410bec2b830997b42e4fb033dfb924a Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Fri, 21 Dec 2007 19:34:33 +0000
Subject: [PATCH] add some comments to the node scheduling logic clean up logic
to read node/job/partition state file's version
---
src/slurmctld/job_mgr.c | 33 ++++++-------------------------
src/slurmctld/node_mgr.c | 18 ++++-------------
src/slurmctld/node_scheduler.c | 36 ++++++++++++++++++++++++----------
src/slurmctld/partition_mgr.c | 17 +++-------------
4 files changed, 39 insertions(+), 65 deletions(-)
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index 89da29930c1..42d3c025431 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -409,30 +409,9 @@ extern int load_all_job_state(void)
buffer = create_buf(data, data_size);
- /*
- * The old header of the "job_state" file simply contained a
- * timestamp, while the new header contains a "VERXXX" at the
- * beginning (VER001, VER002, etc), a timestamp, and the last
- * job id. To determine if we're looking at an old header or
- * new header, we first check if the file begins with "VER".
- *
- * Each field is preceeded by two bytes which contains the field
- * size. Since we are bypassing the "pack" functions in order
- * see if the header contains a "VERXXX" string, we need to make
- * sure that there is enough data in the buffer to compare against.
- */
- if (size_buf(buffer) >= sizeof(uint32_t) + strlen(JOB_STATE_VERSION))
- {
- char *ptr = get_buf_data(buffer);
-
- if (memcmp(&ptr[sizeof(uint32_t)], JOB_STATE_VERSION, 3) == 0)
- {
- safe_unpackstr_xmalloc(&ver_str, &ver_str_len, buffer);
- debug3("Version string in job_state header is %s",
- ver_str);
- }
- }
- if ((!ver_str) || (strcmp(ver_str, JOB_STATE_VERSION) != 0)) {
+ safe_unpackstr_xmalloc(&ver_str, &ver_str_len, buffer);
+ debug3("Version string in job_state header is %s", ver_str);
+ if (strcmp(ver_str, JOB_STATE_VERSION) != 0) {
error("***********************************************");
error("Can not recover job state, incompatable version");
error("***********************************************");
@@ -2830,11 +2809,11 @@ static int _validate_job_desc(job_desc_msg_t * job_desc_msg, int allocate,
if (job_desc_msg->job_min_threads == (uint16_t) NO_VAL)
job_desc_msg->job_min_threads = 1; /* default 1 thread per core */
if (job_desc_msg->job_min_memory == NO_VAL)
- job_desc_msg->job_min_memory = 1; /* default 1MB mem per node */
+ job_desc_msg->job_min_memory = 0; /* default 0MB mem per node */
if (job_desc_msg->job_max_memory == NO_VAL)
- job_desc_msg->job_max_memory = 1; /* default 1MB mem per node */
+ job_desc_msg->job_max_memory = 0; /* default 0MB mem per node */
if (job_desc_msg->job_min_tmp_disk == NO_VAL)
- job_desc_msg->job_min_tmp_disk = 1;/* default 1MB disk per node */
+ job_desc_msg->job_min_tmp_disk = 0;/* default 0MB disk per node */
return SLURM_SUCCESS;
}
diff --git a/src/slurmctld/node_mgr.c b/src/slurmctld/node_mgr.c
index 97c4600f09a..3f347b7ce27 100644
--- a/src/slurmctld/node_mgr.c
+++ b/src/slurmctld/node_mgr.c
@@ -446,20 +446,9 @@ extern int load_all_node_state ( bool state_only )
buffer = create_buf (data, data_size);
- /*
- * Check the data version so that when the format changes, we
- * we don't try to unpack data using the wrong format routines
- */
- if (size_buf(buffer) >= sizeof(uint32_t) + strlen(NODE_STATE_VERSION)) {
- char *ptr = get_buf_data(buffer);
-
- if (memcmp( &ptr[sizeof(uint32_t)], NODE_STATE_VERSION, 3) == 0) {
- safe_unpackstr_xmalloc( &ver_str, &name_len, buffer);
- debug3("Version string in node_state header is %s",
- ver_str);
- }
- }
- if ((!ver_str) || (strcmp(ver_str, NODE_STATE_VERSION) != 0)) {
+ safe_unpackstr_xmalloc( &ver_str, &name_len, buffer);
+ debug3("Version string in node_state header is %s", ver_str);
+ if (strcmp(ver_str, NODE_STATE_VERSION) != 0) {
error("*****************************************************");
error("Can not recover node state, data version incompatable");
error("*****************************************************");
@@ -468,6 +457,7 @@ extern int load_all_node_state ( bool state_only )
return EFAULT;
}
xfree(ver_str);
+
safe_unpack_time (&time_stamp, buffer);
while (remaining_buf (buffer) > 0) {
diff --git a/src/slurmctld/node_scheduler.c b/src/slurmctld/node_scheduler.c
index 5569b70aa88..204a9f5e7a9 100644
--- a/src/slurmctld/node_scheduler.c
+++ b/src/slurmctld/node_scheduler.c
@@ -610,7 +610,7 @@ _pick_best_nodes(struct node_set *node_set_ptr, int node_set_size,
int max_feature, min_feature;
bool runable_ever = false; /* Job can ever run */
bool runable_avail = false; /* Job can run with available nodes */
- uint32_t cr_enabled = 0;
+ static uint32_t cr_enabled = NO_VAL;
int shared = 0;
select_type_plugin_info_t cr_type = SELECT_TYPE_INFO_NONE;
@@ -619,17 +619,24 @@ _pick_best_nodes(struct node_set *node_set_ptr, int node_set_size,
return ESLURM_REQUESTED_NODE_CONFIG_UNAVAILABLE;
}
- /* Is Consumable Resources enabled? */
- error_code = select_g_get_info_from_plugin (SELECT_CR_PLUGIN,
- &cr_enabled);
- if (error_code != SLURM_SUCCESS)
- return error_code;
+ /* Are Consumable Resources enabled? */
+ if (cr_enabled == NO_VAL) {
+ cr_enabled = 0; /* select/linear and bluegene are no-ops */
+ error_code = select_g_get_info_from_plugin (SELECT_CR_PLUGIN,
+ &cr_enabled);
+ if (error_code != SLURM_SUCCESS) {
+ cr_enabled = NO_VAL;
+ return error_code;
+ }
+ }
shared = _resolve_shared_status(job_ptr->details->shared,
part_ptr->max_share, cr_enabled);
job_ptr->details->shared = shared;
if (cr_enabled) {
+ /* Determine which nodes might be used by this job based upon
+ * its ability to share resources */
job_ptr->cr_enabled = cr_enabled; /* CR enabled for this job */
cr_type = (select_type_plugin_info_t) slurmctld_conf.
@@ -660,8 +667,9 @@ _pick_best_nodes(struct node_set *node_set_ptr, int node_set_size,
}
if (job_ptr->details->req_node_bitmap) { /* specific nodes required */
- /* we have already confirmed that all of these nodes have a
- * usable configuration and are in the proper partition */
+ /* We have already confirmed that all of these nodes have a
+ * usable configuration and are in the proper partition.
+ * Check that these nodes can be used by this job. */
if (min_nodes != 0) {
total_nodes = bit_set_count(
job_ptr->details->req_node_bitmap);
@@ -723,7 +731,9 @@ _pick_best_nodes(struct node_set *node_set_ptr, int node_set_size,
if ((j >= 0) && (j > max_feature))
max_feature = j;
}
-
+
+ /* Accumulate resources for this job based upon its required
+ * features (possibly with node counts). */
for (j = min_feature; j <= max_feature; j++) {
for (i = 0; i < node_set_size; i++) {
@@ -829,6 +839,8 @@ _pick_best_nodes(struct node_set *node_set_ptr, int node_set_size,
avail_bitmap = backup_bitmap;
}
} /* for (i = 0; i < node_set_size; i++) */
+
+
#ifndef HAVE_BG
pick_code = 1;
if (job_ptr->details->req_node_bitmap &&
@@ -1234,6 +1246,7 @@ extern int select_nodes(struct job_record *job_ptr, bool test_only,
if (max_nodes < min_nodes) {
error_code = ESLURM_REQUESTED_PART_CONFIG_UNAVAILABLE;
} else {
+ /* Select resources for the job here */
error_code = _get_req_features(node_set_ptr, node_set_size,
&select_bitmap, job_ptr,
part_ptr, min_nodes, max_nodes,
@@ -1471,6 +1484,9 @@ static int _build_feature_list(struct job_record *job_ptr)
/*
* _build_node_list - identify which nodes could be allocated to a job
+ * based upon node features, memory, processors, etc. Note that a
+ * bitmap is set to indicate which of the job's features that the
+ * nodes satisfy.
* IN job_ptr - pointer to node to be scheduled
* OUT node_set_pptr - list of node sets which could be used for the job
* OUT node_set_size - number of node_set entries
@@ -1836,7 +1852,7 @@ extern void build_node_details(struct job_record *job_ptr)
* _valid_features("[fs1|fs2|fs3|fs4]", "fs3") returns a bitmap with
* the third bit set. For another example
* _valid_features("[fs1|fs2|fs3|fs4]", "fs1,fs3") returns a bitmap
- * with the first and third bits set. This function returns a bitmap
+ * with the first and third bits set. The function returns a bitmap
* with the first bit set if requirements are satisfied without a
* mutually exclusive feature list.
*/
diff --git a/src/slurmctld/partition_mgr.c b/src/slurmctld/partition_mgr.c
index b85da32dd2a..189e3949e54 100644
--- a/src/slurmctld/partition_mgr.c
+++ b/src/slurmctld/partition_mgr.c
@@ -438,20 +438,9 @@ int load_all_part_state(void)
buffer = create_buf(data, data_size);
- /*
- * Check the data version so that when the format changes, we
- * we don't try to unpack data using the wrong format routines
- */
- if (size_buf(buffer) >= sizeof(uint32_t) + strlen(PART_STATE_VERSION)) {
- char *ptr = get_buf_data(buffer);
-
- if (memcmp( &ptr[sizeof(uint32_t)], PART_STATE_VERSION, 3) == 0) {
- safe_unpackstr_xmalloc( &ver_str, &name_len, buffer);
- debug3("Version string in part_state header is %s",
- ver_str);
- }
- }
- if ((!ver_str) || (strcmp(ver_str, PART_STATE_VERSION) != 0)) {
+ safe_unpackstr_xmalloc( &ver_str, &name_len, buffer);
+ debug3("Version string in part_state header is %s", ver_str);
+ if (strcmp(ver_str, PART_STATE_VERSION) != 0) {
error("**********************************************************");
error("Can not recover partition state, data version incompatable");
error("**********************************************************");
--
GitLab