From 4e5314393f3503221a1535fbdd05d8c0da9c8c1c Mon Sep 17 00:00:00 2001
From: Martin Necas <mnecas@redhat.com>
Date: Mon, 7 Oct 2024 16:19:05 +0200
Subject: [PATCH] MTV-1537 | Optimalise the plan scheduler

Issue: When we start the warm migration with VM which has a lot of disks
we wait for the whole VM to get migrated. We do not ignore the disks
that has been already migrated. This can cause that when we have 2 VMs
with 10 disks each and on each there is one larger disk the whole
scheduler will be halted untill they finish. So even if the left 9 disks
will be done no migration will be started.

Fix: Subtract the finished disks from the disk count.

Fixes: https://issues.redhat.com/browse/MTV-1537

Signed-off-by: Martin Necas <mnecas@redhat.com>
---
 .../plan/scheduler/vsphere/scheduler.go       | 26 +++++++++++++++----
 1 file changed, 21 insertions(+), 5 deletions(-)

diff --git a/pkg/controller/plan/scheduler/vsphere/scheduler.go b/pkg/controller/plan/scheduler/vsphere/scheduler.go
index 0b2abef98..8c7888ad1 100644
--- a/pkg/controller/plan/scheduler/vsphere/scheduler.go
+++ b/pkg/controller/plan/scheduler/vsphere/scheduler.go
@@ -107,7 +107,7 @@ func (r *Scheduler) buildInFlight() (err error) {
 			return
 		}
 		if vmStatus.Running() {
-			r.inFlight[vm.Host] += r.cost(vm)
+			r.inFlight[vm.Host] += r.cost(vm, vmStatus)
 		}
 	}
 
@@ -153,7 +153,7 @@ func (r *Scheduler) buildInFlight() (err error) {
 				}
 				return err
 			}
-			r.inFlight[vm.Host] += r.cost(vm)
+			r.inFlight[vm.Host] += r.cost(vm, vmStatus)
 		}
 	}
 
@@ -174,7 +174,7 @@ func (r *Scheduler) buildPending() (err error) {
 		if !vmStatus.MarkedStarted() && !vmStatus.MarkedCompleted() {
 			pending := &pendingVM{
 				status: vmStatus,
-				cost:   r.cost(vm),
+				cost:   r.cost(vm, vmStatus),
 			}
 			r.pending[vm.Host] = append(r.pending[vm.Host], pending)
 		}
@@ -182,16 +182,32 @@ func (r *Scheduler) buildPending() (err error) {
 	return
 }
 
-func (r *Scheduler) cost(vm *model.VM) int {
+func (r *Scheduler) cost(vm *model.VM, vmStatus *plan.VMStatus) int {
 	if coldLocal, _ := r.Plan.VSphereColdLocal(); coldLocal {
 		/// virt-v2v transfers one disk at a time
 		return 1
 	} else {
 		// CDI transfers the disks in parallel by different pods
-		return len(vm.Disks)
+		return len(vm.Disks) - r.finishedDisks(vmStatus)
 	}
 }
 
+// finishedDisks returns a number of the disks that have completed the disk transfer
+// This can reduce the migration time as VMs with one large disks and many small disks won't halt the scheduler
+func (r *Scheduler) finishedDisks(vmStatus *plan.VMStatus) int {
+	var resp = 0
+	for _, step := range vmStatus.Pipeline {
+		if step.Name == "DiskTransfer" {
+			for _, task := range step.Tasks {
+				if task.Phase == "Completed" {
+					resp += 1
+				}
+			}
+		}
+	}
+	return resp
+}
+
 // Return a map of all the VMs that could be scheduled
 // based on the available host capacities.
 func (r *Scheduler) schedulable() (schedulable map[string][]*pendingVM) {