From f503d3d3f0d15743b73c3347a1f42f578ee6a481 Mon Sep 17 00:00:00 2001
From: Benjamin Tovar <btovar@nd.edu>
Date: Wed, 4 Dec 2024 10:08:42 -0500
Subject: [PATCH] Fallback resources: allocated to measured (#3998)

* add merge default to rmsummary

* fallback to committe resources when monitoring not enabled
---
 dttools/src/rmsummary.c             | 15 +++++++++++++++
 dttools/src/rmsummary.h             |  1 +
 taskvine/src/manager/vine_manager.c | 17 ++++++++++++-----
 3 files changed, 28 insertions(+), 5 deletions(-)

diff --git a/dttools/src/rmsummary.c b/dttools/src/rmsummary.c
index 0ec0657cda..c7eca1dd01 100644
--- a/dttools/src/rmsummary.c
+++ b/dttools/src/rmsummary.c
@@ -742,6 +742,21 @@ void rmsummary_merge_override_basic(struct rmsummary *dest, const struct rmsumma
 	RM_BIN_OP_BASIC(dest, src, override_field);
 }
 
+/* Copy the value for all the fields in src to dest when dest < 0 */
+static inline double default_field(double d, double s)
+{
+	return (d > -1) ? d : s;
+}
+
+void rmsummary_merge_default(struct rmsummary *dest, const struct rmsummary *src)
+{
+	if (!src) {
+		return;
+	}
+
+	RM_BIN_OP(dest, src, default_field);
+}
+
 struct rmsummary *rmsummary_copy(const struct rmsummary *src, int deep_copy)
 {
 	struct rmsummary *dest = rmsummary_create(-1);
diff --git a/dttools/src/rmsummary.h b/dttools/src/rmsummary.h
index ca8089a3f8..4f7ea8486b 100644
--- a/dttools/src/rmsummary.h
+++ b/dttools/src/rmsummary.h
@@ -111,6 +111,7 @@ void rmsummary_merge_max_w_time(struct rmsummary *dest, const struct rmsummary *
 struct rmsummary *rmsummary_copy(const struct rmsummary *src, int deep_copy);
 void rmsummary_merge_override(struct rmsummary *dest, const struct rmsummary *src);
 void rmsummary_merge_override_basic(struct rmsummary *dest, const struct rmsummary *src);
+void rmsummary_merge_default(struct rmsummary *dest, const struct rmsummary *src);
 void rmsummary_merge_max(struct rmsummary *dest, const struct rmsummary *src);
 void rmsummary_merge_min(struct rmsummary *dest, const struct rmsummary *src);
 void rmsummary_add(struct rmsummary *dest, const struct rmsummary *src);
diff --git a/taskvine/src/manager/vine_manager.c b/taskvine/src/manager/vine_manager.c
index 4ecdc63b7d..168d822186 100644
--- a/taskvine/src/manager/vine_manager.c
+++ b/taskvine/src/manager/vine_manager.c
@@ -566,12 +566,17 @@ static vine_result_code_t get_completion_result(struct vine_manager *q, struct v
 		t->time_workers_execute_last = observed_execution_time > execution_time ? execution_time : observed_execution_time;
 		t->time_workers_execute_last_start = start_time;
 		t->time_workers_execute_last_end = end_time;
-		t->resources_measured->wall_time = t->time_workers_execute_last_end - t->time_workers_execute_last_start;
 		t->time_workers_execute_all += t->time_workers_execute_last;
 		t->output_length = output_length;
 		t->result = task_status;
 		t->exit_code = exit_status;
 
+		/* fill resources measured with whatever vine reported/committed, as a fallback when task ran without monitoring enabled */
+		t->resources_measured->start = ((double)start_time) / ONE_SECOND;
+		t->resources_measured->end = ((double)end_time) / ONE_SECOND;
+		t->resources_measured->wall_time = ((double)t->time_workers_execute_last) / ONE_SECOND;
+		rmsummary_merge_override_basic(t->resources_measured, t->resources_allocated);
+
 		/* If output is less than 1KB stdout is sent along with completion msg. retrieve it from the link. */
 		if (bytes_sent) {
 			get_stdout(q, w, t, bytes_sent);
@@ -1168,12 +1173,14 @@ static void read_measured_resources(struct vine_manager *q, struct vine_task *t)
 {
 	char *summary = monitor_file_name(q, t, ".summary", 0);
 
-	if (t->resources_measured) {
-		rmsummary_delete(t->resources_measured);
-	}
-
+	struct rmsummary *tmp = t->resources_measured;
 	t->resources_measured = rmsummary_parse_file_single(summary);
 
+	/* read the fallback values set by get_completion_result, if any */
+	/* if tmp is null that's ok, both delete and merge_default check for it */
+	rmsummary_merge_default(t->resources_measured, tmp);
+	rmsummary_delete(tmp);
+
 	if (t->resources_measured) {
 		t->exit_code = t->resources_measured->exit_status;