asim: correctly update queue.next in pushReplicateChange

Previously, we did not assign the next field of the queues correctly because we
passed the base queue by value into pushReplicateChange, which caused updates to
the next field to affect only the copy, not the original queue. This commit
fixes the issue by returning the correct next value and updating the queue
directly.

Epic: none
Release note: none

Author: wenyihu6

pkg/kv/kvserver/asim/queue/lease_queue.go

@@ -158,7 +158,7 @@ func (lq *leaseQueue) Tick(ctx context.Context, tick time.Time, s state.State) {
 			continue
 		}
 
-		pushReplicateChange(
+		lq.next = pushReplicateChange(
 			ctx, change, rng, tick, lq.settings.ReplicaChangeDelayFn(), lq.baseQueue)
 	}
 

pkg/kv/kvserver/asim/queue/replicate_queue.go

@@ -152,7 +152,7 @@ func (rq *replicateQueue) Tick(ctx context.Context, tick time.Time, s state.Stat
 			continue
 		}
 
-		pushReplicateChange(
+		rq.next = pushReplicateChange(
 			ctx, change, rng, tick, rq.settings.ReplicaChangeDelayFn(), rq.baseQueue)
 	}
 
@@ -166,12 +166,13 @@ func pushReplicateChange(
 	tick time.Time,
 	delayFn func(int64, bool) time.Duration,
 	queue baseQueue,
-) {
+) time.Time {
 	var stateChange state.Change
+	next := tick
 	switch op := change.Op.(type) {
 	case plan.AllocationNoop:
 		// Nothing to do.
-		return
+		return next
 	case plan.AllocationFinalizeAtomicReplicationOp:
 		panic("unimplemented finalize atomic replication op")
 	case plan.AllocationTransferLeaseOp:
@@ -194,9 +195,10 @@ func pushReplicateChange(
 	}
 
 	if completeAt, ok := queue.stateChanger.Push(tick, stateChange); ok {
-		queue.next = completeAt
 		log.VEventf(ctx, 1, "pushing state change succeeded, complete at %s (cur %s)", completeAt, tick)
+		next = completeAt
 	} else {
 		log.VEventf(ctx, 1, "pushing state change failed")
 	}
+	return next
 }

pkg/kv/kvserver/asim/tests/testdata/non_rand/example_add_node.txt

@@ -28,31 +28,38 @@ set_span_config
 
 eval duration=20m samples=1 seed=42
 ----
-OK
+failed assertion sample 1
+  balance stat=replicas threshold=(<1.05) ticks=6
+	max/mean=1.08 tick=0
+	max/mean=1.08 tick=1
+	max/mean=1.08 tick=2
+	max/mean=1.08 tick=3
+	max/mean=1.08 tick=4
+	max/mean=1.08 tick=5
 
 # Plot the replica count from the evaluation. Since there are 300 replicas on
 # s1 and the default RF=3, we expect the other stores to be up-replicated to
 # 300 replicas as well.
 plot stat=replicas sample=1
 ----
- 301 ┼──────────────────────────────────────╭────────────────────────────────────────
- 281 ┤                                   ╭──╯
- 261 ┤                                 ╭─╯╯
- 241 ┤                              ╭──╯
- 221 ┤                           ╭╭─╯
- 201 ┤                        ╭╭──╯
- 181 ┤                      ╭╭─╯
- 161 ┤                   ╭╭──╯
- 140 ┤                 ╭──╯
- 120 ┤              ╭╭─╯
- 100 ┤           ╭╭──╯
-  80 ┤         ╭──╯
-  60 ┤      ╭──╯
-  40 ┤    ╭─╯╯
-  20 ┤ ╭──╯
-   0 ┼─╯
+ 301 ┼───────────────────────────────────────────────────────────────────────────────
+ 281 ┤                                                                              ╭
+ 261 ┤                                                                     ╭─╭───────
+ 241 ┤                                                                ╭──────╯
+ 221 ┤                                                           ╭────╯
+ 201 ┤                                                     ╭╭────╯
+ 181 ┤                                               ╭──────╯
+ 161 ┤                                          ╭────╯╯
+ 140 ┤                                      ╭───╯
+ 120 ┤                                   ╭──╯╯
+ 100 ┤                              ╭────╯
+  80 ┤                           ╭──╯╯
+  60 ┤                     ╭╭────╯
+  40 ┤             ╭────────╯
+  20 ┤    ╭╭───────╯╯
+   0 ┼─────╯
                                           replicas
 initial store values: [s1=301, s2=0, s3=0] (stddev=141.89, mean=100.33, sum=301)
-last store values: [s1=300, s2=300, s3=301] (stddev=0.47, mean=300.33, sum=901)
+last store values: [s1=301, s2=272, s3=265] (stddev=15.58, mean=279.33, sum=838)
 
 # vim:ft=sh

pkg/kv/kvserver/asim/tests/testdata/non_rand/example_fulldisk.txt

@@ -23,46 +23,46 @@ OK
 # store 5. This is shown below as it sheds replicas.
 plot stat=replicas
 ----
- 347 ┤                                       ╭╮            ╭╮╭╭╮ ╭╮╭──╭╮─╭╮─╭╮─╭──╮╮╭
- 334 ┤     ╭╮ ╭╮    ╭╮  ╭─╮╮╭╮──╭╮╮╭╮─╭───╮─╭───────╮╭──╮╭─╯╰╮╭──╯╰╮╭─╯╰─╯╰─╯╰─╯─╯│╰╭
- 321 ┤ ╭───╭──╮╭╭───────╯╭│╭────╯╰─╯╰─╯╮╭─╰─╯╰╯╮╭╰─╯╰╯─╯╰╯╯╰─╰╯╰╮╭╯╰╯  ╰╯╰─╯      ╰─╯
- 307 ┼╮────╯╯╯╰─╯╰╯╯╰╯╯ ╰╯╰╯╯ ╰╯ ╰╯    ╰╯      ╰╯               ╰╯
- 294 ┤╰╮
- 281 ┤ │
- 268 ┤ ╰╮╭╮ ╭╮
- 255 ┤  ╰╯╰─╯│╭─╮ ╭╮
- 241 ┤       ╰╯ ╰─╯│╭─╮╭─╮ ╭╮
- 228 ┤             ╰╯ ╰╯ │╭╯╰╮╭╮ ╭╮ ╭╮
- 215 ┤                   ╰╯  ╰╯╰─╯╰─╯╰╮╭─╮    ╭╮
- 202 ┤                                ╰╯ │╭─╮ │╰╮╭╮ ╭─╮ ╭╮     ╭╮
- 189 ┤                                   ╰╯ ╰─╯ ╰╯╰╮│ │ │╰╮╭─╮ │╰╮ ╭╮  ╭╮
- 175 ┤                                             ╰╯ ╰─╯ ╰╯ ╰─╯ ╰─╯╰╮ │╰╮╭─╮ ╭─╮ ╭─╮
- 162 ┤                                                               ╰─╯ ╰╯ │ │ │ │ ╰
- 149 ┤                                                                      ╰─╯ ╰─╯
+ 344 ┤                                                       ╭╮   ╭╮╭╭╮╮╭─╮╮╭╮─╭─╮╭──
+ 331 ┤                       ╭╮ ╭╮╭─╮╭╮──╭──╮╭───────────────────────╯╰─╯─╰─╯╰────╯╯─
+ 318 ┤   ╭───╭─╮╮╭────────────────╯╯╰╯╰──╯╰╰╰╯╯ ╰╯
+ 305 ┼╮──────╯╯╰─╯╯
+ 293 ┤╰╮
+ 280 ┤ ╰─╮
+ 267 ┤   ╰╮
+ 255 ┤    ╰────╮
+ 242 ┤         ╰─────╮
+ 229 ┤               ╰─────╮
+ 217 ┤                     ╰───────╮
+ 204 ┤                             ╰──────────╮
+ 191 ┤                                        ╰────╮
+ 178 ┤                                             ╰───────────╮ ╭╮
+ 166 ┤                                                         ╰─╯╰──────╮╭──╮ ╭╮
+ 153 ┤                                                                   ╰╯  ╰─╯╰────
                                           replicas
 initial store values: [s1=300, s2=300, s3=300, s4=300, s5=300] (stddev=0.00, mean=300.00, sum=1500)
-last store values: [s1=347, s2=342, s3=338, s4=336, s5=152] (stddev=75.59, mean=303.00, sum=1515)
+last store values: [s1=342, s2=336, s3=339, s4=339, s5=153] (stddev=74.42, mean=301.80, sum=1509)
 
 # Plot the % of disk storage capacity used. We should see s5 hovering right
 # around 92.5-95% (the storage capacity threshold value).
 plot stat=disk_fraction_used
 ----
- 1.05 ┼╮                                                        ╭╮  ╭╮  ╭╮
- 0.99 ┤╰╮                   ╭╮ ╭╮ ╭╮ ╭╮ ╭╮     ╭╮ ╭╮ ╭─╮ ╭╮  ╭╮ │╰╮ ││  ││ ╭╮  ╭─╮ ╭─╮
- 0.94 ┤ ╰╮╭╮ ╭╮╭─╮╭─╮╭─╮╭─╮╭╯╰╮│╰╮│╰─╯╰╮│╰╮╭─╮ │╰╮││ │ │ │╰╮╭╯╰╮│ ╰─╯╰╮╭╯╰╮│╰╮ │ │ │ │
- 0.88 ┤  ╰╯╰─╯╰╯ ╰╯ ╰╯ ╰╯ ╰╯  ╰╯ ╰╯    ╰╯ ╰╯ │╭╯ ╰╯╰─╯ ╰─╯ ╰╯  ╰╯     ╰╯  ╰╯ ╰─╯ ╰─╯ ╰
- 0.82 ┤                                      ╰╯
+ 1.05 ┼─╮
+ 0.99 ┤ ╰─╮
+ 0.94 ┤   ╰─────────────────╮╭──╮╭──╮╭────────────────────╮╭────╮ ╭───────╮╭──╮╭──────
+ 0.88 ┤                     ╰╯  ╰╯  ╰╯                    ╰╯    ╰─╯       ╰╯  ╰╯
+ 0.82 ┤
  0.77 ┤
  0.71 ┤
  0.65 ┤
  0.60 ┤
  0.54 ┤
  0.49 ┤
  0.43 ┤
- 0.37 ┤                                                           ╭╮╭─╭───────────────
- 0.32 ┤                                    ╭──╭───────────────────╯╰──╯╰─╯
- 0.26 ┤              ╭╮╭╮╭─╮╭─────────────────╯╯─╯
- 0.20 ┼────────────────╯╰╯╯╰╯╯
+ 0.37 ┤                                                            ╭╭╭────────────────
+ 0.32 ┤                                       ╭╭─────────────────────╯╰╯
+ 0.26 ┤                ╭───────────────────────╯╯
+ 0.20 ┼────────────────╯╯╯
                                       disk_fraction_used
 initial store values: [s1=0.20, s2=0.20, s3=0.20, s4=0.20, s5=1.05] (stddev=0.34, mean=0.37, sum=2)
-last store values: [s1=0.41, s2=0.41, s3=0.40, s4=0.40, s5=0.93] (stddev=0.21, mean=0.51, sum=3)
+last store values: [s1=0.41, s2=0.40, s3=0.41, s4=0.41, s5=0.94] (stddev=0.22, mean=0.51, sum=3)