From 81514455056c92ab5526f1283f0beafd331f11af Mon Sep 17 00:00:00 2001
From: David Anderson <davea@ssl.berkeley.edu>
Date: Fri, 13 Dec 2024 15:19:41 -0800
Subject: [PATCH] scheduler: for BUDA GPU jobs, put resource usage info in the
 <workunit> element.

---
 sched/sched_locality.cpp |  2 +-
 sched/sched_send.cpp     | 57 ++++++++++++++++++++++++++++++++++++++--
 sched/sched_types.h      |  6 ++---
 sched/sched_version.cpp  |  1 -
 4 files changed, 59 insertions(+), 7 deletions(-)

diff --git a/sched/sched_locality.cpp b/sched/sched_locality.cpp
index 5869a8d8a2..b8b93bd32d 100644
--- a/sched/sched_locality.cpp
+++ b/sched/sched_locality.cpp
@@ -336,7 +336,7 @@ static int possibly_send_result(SCHED_DB_RESULT& result) {
     HOST_USAGE hu;
     check_buda_plan_class(wu, hu, is_buda, is_ok);
     if (is_buda) {
-        if (!is_ok) ERR_NO_APP_VERSION;
+        if (!is_ok) return ERR_NO_APP_VERSION;
     } else {
         hu = bavp->host_usage;
     }
diff --git a/sched/sched_send.cpp b/sched/sched_send.cpp
index 5fd5f1b066..0ce0f9f88e 100644
--- a/sched/sched_send.cpp
+++ b/sched/sched_send.cpp
@@ -570,10 +570,58 @@ static int insert_wu_tags(WORKUNIT& wu, APP& app) {
     return insert_after(wu.xml_doc, "<workunit>\n", buf);
 }
 
+// add host usage into to WU's xml_doc (for BUDA jobs)
+//
+static int add_usage_to_wu(WORKUNIT &wu, HOST_USAGE &hu) {
+    char buf[2048], buf2[2048];
+    snprintf(buf, sizeof(buf),
+        "   <avg_ncpus>%f</avg_ncpus>\n"
+        "   <flops>%f</flops>\n",
+        hu.avg_ncpus,
+        hu.projected_flops
+    );
+    if (hu.proc_type != PROC_TYPE_CPU) {
+        snprintf(buf2, sizeof(buf2),
+            "   <coproc>\n"
+            "        <type>%s</type>\n"
+            "        <count>%f</count>\n"
+            "    </coproc>\n",
+            proc_type_name_xml(hu.proc_type),
+            hu.gpu_usage
+        );
+        strcat(buf, buf2);
+    }
+    if (strlen(hu.cmdline)) {
+        snprintf(buf2, sizeof(buf2),
+            "   <cmdline>%s</cmdline>\n",
+            hu.cmdline
+        );
+        strcat(buf, buf2);
+    }
+
+    char *p = wu.xml_doc;
+    if (strlen(p) + strlen(buf) + 10 > sizeof(wu.xml_doc)) {
+        log_messages.printf(MSG_CRITICAL,
+            "add_usage_to_wu(): field too small: %ld %ld %ld\n",
+            strlen(p), strlen(buf), sizeof(wu.xml_doc)
+        );
+        return -1;
+    }
+    p = strstr(p, "</workunit>");
+    if (!p) {
+        log_messages.printf(MSG_CRITICAL, "add_usage_to_wu(): no end tag\n");
+        return -1;
+    }
+    strcpy(p, buf);
+    strcat(p, "</workunit>");
+    return 0;
+}
+
 // Add the given workunit, app, and app version to a reply.
 //
 static int add_wu_to_reply(
-    WORKUNIT& wu, SCHEDULER_REPLY&, APP* app, BEST_APP_VERSION* bavp
+    WORKUNIT& wu, SCHEDULER_REPLY&, APP* app, BEST_APP_VERSION* bavp,
+    bool is_buda, HOST_USAGE &hu
 ) {
     int retval;
     WORKUNIT wu2, wu3;
@@ -626,6 +674,11 @@ static int add_wu_to_reply(
         );
         return retval;
     }
+
+    if (is_buda) {
+        retval = add_usage_to_wu(wu2, hu);
+        if (retval) return retval;
+    }
     wu3 = wu2;
     if (strlen(config.replace_download_url_by_timezone)) {
         process_wu_timezone(wu2, wu3);
@@ -1015,7 +1068,7 @@ int add_result_to_reply(
     // done with DB updates.
     //
 
-    retval = add_wu_to_reply(wu, *g_reply, app, bavp);
+    retval = add_wu_to_reply(wu, *g_reply, app, bavp, is_buda, host_usage);
     if (retval) return retval;
 
     // Adjust available disk space.
diff --git a/sched/sched_types.h b/sched/sched_types.h
index 919aac69bc..03522db506 100644
--- a/sched/sched_types.h
+++ b/sched/sched_types.h
@@ -63,13 +63,13 @@ struct USER_MESSAGE {
 
 // The resource usage (CPU, GPU, RAM) of a job,
 // and estimates of its speed
-// Populated by plan-class functions if have plan class,
+// Populated by plan-class functions if there's a plan class,
 // else by HOST_USAGE::sequential_app()
 //
 struct HOST_USAGE {
     int proc_type;
     double gpu_usage;
-    double gpu_ram;
+    double gpu_ram;     // not currently used by client
     double avg_ncpus;
     double mem_usage;
         // mem usage if specified by the plan class
@@ -439,7 +439,7 @@ struct WORK_REQ_BASE {
         req_instances[proc_type] = 0;
     }
 
-    // older clients send send a single number, the requested duration of jobs
+    // older clients send a single number, the requested duration of jobs
     //
     double seconds_to_fill;
 
diff --git a/sched/sched_version.cpp b/sched/sched_version.cpp
index 59e0043d8f..5396f185d4 100644
--- a/sched/sched_version.cpp
+++ b/sched/sched_version.cpp
@@ -982,4 +982,3 @@ BEST_APP_VERSION* get_app_version(
     }
     return bavp;
 }
-