From f3ba65ddf8642b941cb04ddbb69fdb1e8515a6d2 Mon Sep 17 00:00:00 2001
From: Tom Clegg <tom@curii.com>
Date: Mon, 6 Sep 2021 21:51:40 -0400
Subject: [PATCH] 18102: Fix flapping lock/unlock.

Arvados-DCO-1.1-Signed-off-by: Tom Clegg <tom@curii.com>
---
 lib/dispatchcloud/container/queue.go          |  4 +-
 lib/dispatchcloud/scheduler/run_queue.go      | 11 +++-
 lib/dispatchcloud/scheduler/run_queue_test.go | 59 +++++++++++++++++++
 lib/dispatchcloud/test/queue.go               |  1 +
 4 files changed, 73 insertions(+), 2 deletions(-)

diff --git a/lib/dispatchcloud/container/queue.go b/lib/dispatchcloud/container/queue.go
index 7a2727c1e9..938ef915f2 100644
--- a/lib/dispatchcloud/container/queue.go
+++ b/lib/dispatchcloud/container/queue.go
@@ -31,6 +31,7 @@ type QueueEnt struct {
 	// populated.
 	Container    arvados.Container    `json:"container"`
 	InstanceType arvados.InstanceType `json:"instance_type"`
+	FirstSeenAt  time.Time            `json:"first_seen_at"`
 }
 
 // String implements fmt.Stringer by returning the queued container's
@@ -229,6 +230,7 @@ func (cq *Queue) delEnt(uuid string, state arvados.ContainerState) {
 	delete(cq.current, uuid)
 }
 
+// Caller must have lock.
 func (cq *Queue) addEnt(uuid string, ctr arvados.Container) {
 	it, err := cq.chooseType(&ctr)
 	if err != nil && (ctr.State == arvados.ContainerStateQueued || ctr.State == arvados.ContainerStateLocked) {
@@ -284,7 +286,7 @@ func (cq *Queue) addEnt(uuid string, ctr arvados.Container) {
 		"Priority":      ctr.Priority,
 		"InstanceType":  it.Name,
 	}).Info("adding container to queue")
-	cq.current[uuid] = QueueEnt{Container: ctr, InstanceType: it}
+	cq.current[uuid] = QueueEnt{Container: ctr, InstanceType: it, FirstSeenAt: time.Now()}
 }
 
 // Lock acquires the dispatch lock for the given container.
diff --git a/lib/dispatchcloud/scheduler/run_queue.go b/lib/dispatchcloud/scheduler/run_queue.go
index d2f6d1c2cb..5bd4477ca5 100644
--- a/lib/dispatchcloud/scheduler/run_queue.go
+++ b/lib/dispatchcloud/scheduler/run_queue.go
@@ -20,7 +20,16 @@ func (sch *Scheduler) runQueue() {
 		sorted = append(sorted, ent)
 	}
 	sort.Slice(sorted, func(i, j int) bool {
-		return sorted[i].Container.Priority > sorted[j].Container.Priority
+		if pi, pj := sorted[i].Container.Priority, sorted[j].Container.Priority; pi != pj {
+			return pi > pj
+		} else {
+			// When containers have identical priority,
+			// start them in the order we first noticed
+			// them. This avoids extra lock/unlock cycles
+			// when we unlock the containers that don't
+			// fit in the available pool.
+			return sorted[i].FirstSeenAt.Before(sorted[j].FirstSeenAt)
+		}
 	})
 
 	running := sch.pool.Running()
diff --git a/lib/dispatchcloud/scheduler/run_queue_test.go b/lib/dispatchcloud/scheduler/run_queue_test.go
index c8d45cbd19..5b5fa960a1 100644
--- a/lib/dispatchcloud/scheduler/run_queue_test.go
+++ b/lib/dispatchcloud/scheduler/run_queue_test.go
@@ -263,6 +263,65 @@ func (*SchedulerSuite) TestShutdownAtQuota(c *check.C) {
 	}
 }
 
+// Don't flap lock/unlock when equal-priority containers compete for
+// limited workers.
+//
+// (Unless we use FirstSeenAt as a secondary sort key, each runQueue()
+// tends to choose a different one of the equal-priority containers as
+// the "first" one that should be locked, and unlock the one it chose
+// last time. This generates logging noise, and fails containers by
+// reaching MaxDispatchAttempts quickly.)
+func (*SchedulerSuite) TestEqualPriorityContainers(c *check.C) {
+	logger := ctxlog.TestLogger(c)
+	ctx := ctxlog.Context(context.Background(), logger)
+	queue := test.Queue{
+		ChooseType: chooseType,
+		Logger:     logger,
+	}
+	for i := 0; i < 8; i++ {
+		queue.Containers = append(queue.Containers, arvados.Container{
+			UUID:     test.ContainerUUID(i),
+			Priority: 333,
+			State:    arvados.ContainerStateQueued,
+			RuntimeConstraints: arvados.RuntimeConstraints{
+				VCPUs: 3,
+				RAM:   3 << 30,
+			},
+		})
+	}
+	queue.Update()
+	pool := stubPool{
+		quota: 2,
+		unalloc: map[arvados.InstanceType]int{
+			test.InstanceType(3): 1,
+		},
+		idle: map[arvados.InstanceType]int{
+			test.InstanceType(3): 1,
+		},
+		running:   map[string]time.Time{},
+		creates:   []arvados.InstanceType{},
+		starts:    []string{},
+		canCreate: 1,
+	}
+	sch := New(ctx, &queue, &pool, nil, time.Millisecond, time.Millisecond)
+	for i := 0; i < 30; i++ {
+		sch.runQueue()
+		sch.sync()
+		time.Sleep(time.Millisecond)
+	}
+	c.Check(pool.shutdowns, check.Equals, 0)
+	c.Check(pool.starts, check.HasLen, 1)
+	unlocked := map[string]int{}
+	for _, chg := range queue.StateChanges() {
+		if chg.To == arvados.ContainerStateQueued {
+			unlocked[chg.UUID]++
+		}
+	}
+	for uuid, count := range unlocked {
+		c.Check(count, check.Equals, 1, check.Commentf("%s", uuid))
+	}
+}
+
 // Start lower-priority containers while waiting for new/existing
 // workers to come up for higher-priority containers.
 func (*SchedulerSuite) TestStartWhileCreating(c *check.C) {
diff --git a/lib/dispatchcloud/test/queue.go b/lib/dispatchcloud/test/queue.go
index 5973d16390..fcb2cfb33b 100644
--- a/lib/dispatchcloud/test/queue.go
+++ b/lib/dispatchcloud/test/queue.go
@@ -160,6 +160,7 @@ func (q *Queue) Update() error {
 			upd[ctr.UUID] = container.QueueEnt{
 				Container:    ctr,
 				InstanceType: it,
+				FirstSeenAt:  time.Now(),
 			}
 		}
 	}
-- 
2.30.2