asim: improve comments and test set up for sma

This commit improves comments and test setup for tests under sma.

Author: wenyihu6

pkg/kv/kvserver/asim/tests/testdata/non_rand/sma/add_node.txt

@@ -27,21 +27,16 @@ set_span_config
 [0,10000): num_replicas=3 num_voters=3
 ----
 
+# Speed up up-replication by bumping the rebalancing snapshot rate.
+setting rebalancing_snapshot_rate_mib=512
+----
 
 # 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.
-eval duration=20m samples=1 seed=42 metrics=(replicas)
+eval duration=12m samples=1 seed=42 metrics=(replicas)
 ----
 replicas#1: first: [s1=301, s2=0, s3=0] (stddev=141.89, mean=100.33, sum=301)
-replicas#1: last:  [s1=301, s2=271, s3=267] (stddev=15.17, mean=279.67, sum=839)
+replicas#1: last:  [s1=300, s2=300, s3=301] (stddev=0.47, mean=300.33, sum=901)
 replicas#1: thrash_pct: [s1=0%, s2=0%, s3=0%]  (sum=0%)
-artifacts[sma-count]: 58e379d72ddc641d
-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
+artifacts[sma-count]: c99b633edc560571
 ==========================

pkg/kv/kvserver/asim/tests/testdata/non_rand/sma/conformance.txt

@@ -5,7 +5,7 @@ gen_cluster nodes=5 region=(a,b) nodes_per_region=(2,3)
 ----
 
 # Generate 10 ranges, one replica will be placed on each node initially.
-gen_ranges ranges=5 repl_factor=4
+gen_ranges ranges=10 repl_factor=4
 ----
 
 # Update the span config to require a non-voter in the `region=a` locality.
@@ -26,9 +26,32 @@ set_span_config delay=5m
 assertion type=conformance under=0 over=0 unavailable=0 violating=0
 ----
 
-eval duration=10m cfgs=(sma-count,mma-only)
+eval duration=1m cfgs=(sma-count,mma-only) metrics=(replicas) full=true
 ----
-artifacts[sma-count]: 79c6ee6ca7f1ab55
+replicas#1: first: [s1=9, s2=9, s3=9, s4=9, s5=8] (stddev=0.40, mean=8.80, sum=44)
+replicas#1: last:  [s1=11, s2=10, s3=9, s4=5, s5=9] (stddev=2.04, mean=8.80, sum=44)
+replicas#1: thrash_pct: [s1=239%, s2=170%, s3=67%, s4=0%, s5=0%]  (sum=476%)
+artifacts[sma-count]: a6973dd2f74e933f
 ==========================
-artifacts[mma-only]: 79c6ee6ca7f1ab55
+replicas#1: first: [s1=9, s2=9, s3=9, s4=9, s5=8] (stddev=0.40, mean=8.80, sum=44)
+replicas#1: last:  [s1=11, s2=11, s3=9, s4=5, s5=8] (stddev=2.23, mean=8.80, sum=44)
+replicas#1: thrash_pct: [s1=105%, s2=37%, s3=67%, s4=0%, s5=0%]  (sum=209%)
+artifacts[mma-only]: 76036bbb784a6b13
+==========================
+Cluster Set Up
+	n1(a,a_1,8vcpu): {s1:(256GiB)}
+	n2(a,a_1,8vcpu): {s2:(256GiB)}
+	n3(b,b_1,8vcpu): {s3:(256GiB)}
+	n4(b,b_1,8vcpu): {s4:(256GiB)}
+	n5(b,b_1,8vcpu): {s5:(256GiB)}
+Key Space
+	[0,10000): 10(rf=4), 0MiB, [s1:(8,2*),s2:(8,2*),s3:(8,2*),s4:(8,2*),s5:(8,2*)]
+Event
+	[0,10000): 3voters,1nonvoters [replicas:{2:region=a}{2:region=b}] [voters:{1:region=a}{2:region=b}]
+	set LBRebalancingMode to 2
+	[0,10000): 3voters,1nonvoters [replicas:{2:region=a}{2:region=b}] [voters:{2:region=a}{1:region=b}] at 11:05:00
+Workload Set Up
+	empty
+Changed Settings
+	empty
 ==========================

pkg/kv/kvserver/asim/tests/testdata/non_rand/sma/decommission_conformance.txt

@@ -26,9 +26,9 @@ set_liveness node=4 liveness=decommissioning delay=5m
 assertion type=conformance under=0 over=0 unavailable=0 violating=0
 ----
 
-eval duration=20m cfgs=(sma-count,mma-only)
+eval duration=6m cfgs=(sma-count,mma-only)
 ----
-artifacts[sma-count]: 8e9d6ada04cb9561
+artifacts[sma-count]: b29d203c32d8eff1
 ==========================
-artifacts[mma-only]: 8d6fbe46b5c14575
+artifacts[mma-only]: 7cb8843ee1554cf5
 ==========================

pkg/kv/kvserver/asim/tests/testdata/non_rand/sma/fulldisk.txt

@@ -1,3 +1,8 @@
+# This test verifies that the allocator correctly handles disk fullness by
+#  shedding replicas from stores that exceed the storage capacity threshold. It
+# sets up a cluster where one store (s5) has significantly less capacity (100 GiB)
+# than others (512 GiB), causing it to hit the disk fullness threshold and
+# continuously shed replicas to maintain disk usage less than 95%.
 skip_under_ci
 ----
 
@@ -20,29 +25,29 @@ set_capacity store=5 capacity=107374182400
 # We will repeatedly hit the disk fullness threshold which causes shedding
 # replicas on store 5. We should see s5 hovering right around 92.5-95%
 # (the storage capacity threshold value).
-eval duration=30m seed=42 metrics=(replicas,disk_fraction_used) cfgs=(sma-count,mma-only,mma-count)
+eval duration=20m seed=42 metrics=(replicas,disk_fraction_used) cfgs=(sma-count,mma-only,mma-count)
 ----
 disk_fraction_used#1: first: [s1=0.20, s2=0.20, s3=0.20, s4=0.20, s5=1.05] (stddev=0.34, mean=0.37, sum=2)
-disk_fraction_used#1: last:  [s1=0.30, s2=0.30, s3=0.30, s4=0.31, s5=0.95] (stddev=0.26, mean=0.43, sum=2)
-disk_fraction_used#1: thrash_pct: [s1=30%, s2=31%, s3=32%, s4=30%, s5=124%]  (sum=246%)
+disk_fraction_used#1: last:  [s1=0.27, s2=0.27, s3=0.27, s4=0.27, s5=0.92] (stddev=0.26, mean=0.40, sum=2)
+disk_fraction_used#1: thrash_pct: [s1=19%, s2=20%, s3=20%, s4=19%, s5=74%]  (sum=153%)
 replicas#1: first: [s1=300, s2=300, s3=300, s4=300, s5=300] (stddev=0.00, mean=300.00, sum=1500)
-replicas#1: last:  [s1=325, s2=324, s3=322, s4=329, s5=200] (stddev=50.05, mean=300.00, sum=1500)
-replicas#1: thrash_pct: [s1=227%, s2=237%, s3=240%, s4=228%, s5=91%]  (sum=1023%)
-artifacts[sma-count]: e501ccc056a9e929
+replicas#1: last:  [s1=324, s2=322, s3=323, s4=319, s5=212] (stddev=44.03, mean=300.00, sum=1500)
+replicas#1: thrash_pct: [s1=179%, s2=195%, s3=190%, s4=185%, s5=56%]  (sum=805%)
+artifacts[sma-count]: 6a137567c06e77bf
 ==========================
 disk_fraction_used#1: first: [s1=0.20, s2=0.20, s3=0.20, s4=0.20, s5=1.05] (stddev=0.34, mean=0.37, sum=2)
-disk_fraction_used#1: last:  [s1=0.31, s2=0.31, s3=0.31, s4=0.29, s5=0.91] (stddev=0.24, mean=0.43, sum=2)
-disk_fraction_used#1: thrash_pct: [s1=2%, s2=1%, s3=1%, s4=0%, s5=67%]  (sum=71%)
+disk_fraction_used#1: last:  [s1=0.28, s2=0.28, s3=0.28, s4=0.26, s5=0.90] (stddev=0.25, mean=0.40, sum=2)
+disk_fraction_used#1: thrash_pct: [s1=1%, s2=0%, s3=1%, s4=0%, s5=47%]  (sum=48%)
 replicas#1: first: [s1=300, s2=300, s3=300, s4=300, s5=300] (stddev=0.00, mean=300.00, sum=1500)
-replicas#1: last:  [s1=331, s2=330, s3=331, s4=315, s5=193] (stddev=53.84, mean=300.00, sum=1500)
-replicas#1: thrash_pct: [s1=14%, s2=4%, s3=9%, s4=0%, s5=0%]  (sum=26%)
-artifacts[mma-only]: 2f2f9597d62f93d7
+replicas#1: last:  [s1=329, s2=330, s3=329, s4=305, s5=207] (stddev=47.45, mean=300.00, sum=1500)
+replicas#1: thrash_pct: [s1=4%, s2=0%, s3=6%, s4=0%, s5=0%]  (sum=11%)
+artifacts[mma-only]: 1487ef171680ff9f
 ==========================
 disk_fraction_used#1: first: [s1=0.20, s2=0.20, s3=0.20, s4=0.20, s5=1.05] (stddev=0.34, mean=0.37, sum=2)
-disk_fraction_used#1: last:  [s1=0.30, s2=0.30, s3=0.31, s4=0.30, s5=0.90] (stddev=0.24, mean=0.42, sum=2)
-disk_fraction_used#1: thrash_pct: [s1=37%, s2=35%, s3=36%, s4=35%, s5=175%]  (sum=318%)
+disk_fraction_used#1: last:  [s1=0.27, s2=0.27, s3=0.27, s4=0.27, s5=0.92] (stddev=0.26, mean=0.40, sum=2)
+disk_fraction_used#1: thrash_pct: [s1=23%, s2=23%, s3=24%, s4=24%, s5=129%]  (sum=223%)
 replicas#1: first: [s1=300, s2=300, s3=300, s4=300, s5=300] (stddev=0.00, mean=300.00, sum=1500)
-replicas#1: last:  [s1=326, s2=328, s3=329, s4=326, s5=191] (stddev=54.51, mean=300.00, sum=1500)
-replicas#1: thrash_pct: [s1=277%, s2=263%, s3=272%, s4=260%, s5=176%]  (sum=1247%)
-artifacts[mma-count]: 3118ca7a9543fdd3
+replicas#1: last:  [s1=323, s2=323, s3=324, s4=317, s5=213] (stddev=43.57, mean=300.00, sum=1500)
+replicas#1: thrash_pct: [s1=220%, s2=220%, s3=236%, s4=231%, s5=171%]  (sum=1077%)
+artifacts[mma-count]: 4d36e982fd0dff3e
 ==========================

pkg/kv/kvserver/asim/tests/testdata/non_rand/sma/lease_preferences.txt

@@ -1,5 +1,10 @@
-# This test demonstrates setting and asserting on lease preferences. The
-# cluster topology is identical to example_conformance.
+# This test verifies lease preference handling in a 5-node cluster (2 in region a, 3 in region b).
+# Tests three configurations:
+# 1. Satisfiable: voters in region b, lease preference for region b (no violations)
+# 2. Less-preferred: voters in region b, preference for region a then b (1 lease-less-preferred)
+# 3. Impossible: voters in region b, preference for region a (1 lease-violating)
+#
+# Expected: 1 lease-violating and 1 lease-less-preferred violation.
 gen_cluster nodes=5 region=(a,b) nodes_per_region=(2,3)
 ----
 
@@ -28,9 +33,21 @@ set_span_config
 assertion type=conformance lease-violating=1 lease-less-preferred=1
 ----
 
-eval duration=10m cfgs=(sma-count,mma-only)
+eval duration=1m cfgs=(sma-count,mma-only) metrics=(replicas,leases)
 ----
-artifacts[sma-count]: 35051da9f3ec651d
+leases#1: first: [s1=4, s2=0, s3=0, s4=0, s5=0] (stddev=1.60, mean=0.80, sum=4)
+leases#1: last:  [s1=1, s2=0, s3=0, s4=2, s5=1] (stddev=0.75, mean=0.80, sum=4)
+leases#1: thrash_pct: [s1=0%, s2=0%, s3=0%, s4=0%, s5=0%]  (sum=0%)
+replicas#1: first: [s1=4, s2=4, s3=4, s4=0, s5=0] (stddev=1.96, mean=2.40, sum=12)
+replicas#1: last:  [s1=1, s2=1, s3=4, s4=3, s5=3] (stddev=1.20, mean=2.40, sum=12)
+replicas#1: thrash_pct: [s1=0%, s2=0%, s3=0%, s4=0%, s5=0%]  (sum=0%)
+artifacts[sma-count]: eaa5c7254461b39d
 ==========================
-artifacts[mma-only]: 339579f79e6b26fd
+leases#1: first: [s1=4, s2=0, s3=0, s4=0, s5=0] (stddev=1.60, mean=0.80, sum=4)
+leases#1: last:  [s1=1, s2=0, s3=0, s4=1, s5=2] (stddev=0.75, mean=0.80, sum=4)
+leases#1: thrash_pct: [s1=0%, s2=0%, s3=0%, s4=0%, s5=0%]  (sum=0%)
+replicas#1: first: [s1=4, s2=4, s3=4, s4=0, s5=0] (stddev=1.96, mean=2.40, sum=12)
+replicas#1: last:  [s1=1, s2=1, s3=4, s4=3, s5=3] (stddev=1.20, mean=2.40, sum=12)
+replicas#1: thrash_pct: [s1=0%, s2=0%, s3=0%, s4=0%, s5=0%]  (sum=0%)
+artifacts[mma-only]: 2216e9f740949e3d
 ==========================

pkg/kv/kvserver/asim/tests/testdata/non_rand/sma/liveness.txt

@@ -1,3 +1,6 @@
+# This test verifies node liveness handling in a 7-node cluster with 700 ranges (RF=3).
+# Node 7 is dead initially, node 6 becomes decommissioning after 3 minutes.
+# Expected: Dead/decommissioning nodes lose all replicas and leases.
 skip_under_ci
 ----
 

pkg/kv/kvserver/asim/tests/testdata/non_rand/sma/rebalancing.txt

@@ -1,3 +1,5 @@
+# This test verifies that the allocator can rebalance qps load across a 7-node
+# cluster. mma-only fails here since it doesn't balance based on leases.
 skip_under_ci
 ----
 
@@ -16,18 +18,11 @@ gen_ranges ranges=7 placement_type=skewed
 gen_load rate=7000 rw_ratio=0.95 access_skew=false min_block=128 max_block=256
 ----
 
-# Add two assertions, the first is a balance assertion. The balance assertion
-# requires that when simulation is evaluated that during last 6 ticks (60
-# seconds) the max/mean QPS of the cluster does not exceed 1.15.
+# Add two assertions (balanced, stable steady state) on QPS of the cluster.
 assertion stat=qps type=balance ticks=6 upper_bound=1.15
 ----
 asserting: max_{stores}(qps)/mean_{stores}(qps) ≤ 1.15 at each of last 6 ticks
 
-# The second is a steady state assertion. The steady state assertion requires
-# that during the last 6 ticks (60 seconds), the value of QPS per-store doesn't
-# increase or decrease by more than 5% of the mean. This type of assertion is
-# useful when a stat is balanced but not necessarily stable.
-#
 # TODO(tbg): at this point, six ticks is 3s, not 60s. Update assertion API
 # to take a duration, not ticks.
 assertion stat=qps type=steady ticks=6 upper_bound=0.05
@@ -39,7 +34,7 @@ asserting: |qps(t)/mean_{T}(qps) - 1| ≤ 0.05 ∀ t∈T and each store (T=last
 # Following the evaluation, the samples are checked individually against the
 # existing assertions, added above. If any assertion fails, the reason is
 # printed. If no assertions fail, then OK is printed.
-eval duration=3m samples=2 seed=42 metrics=(qps,replica_moves) cfgs=(sma-count,mma-only)
+eval duration=3m samples=2 seed=42 metrics=(qps,replica_moves) cfgs=(sma-count,mma-only) full=true
 ----
 qps#1: last:  [s1=996, s2=1001, s3=999, s4=994, s5=1001, s6=1002, s7=1007] (stddev=3.93, mean=1000.00, sum=7000)
 qps#1: thrash_pct: [s1=12%, s2=9%, s3=4%, s4=8%, s5=5%, s6=6%, s7=4%]  (sum=48%)
@@ -73,10 +68,26 @@ failed assertion sample 2
 	max/mean=4.00 tick=4
 	max/mean=4.00 tick=5
 ==========================
+Cluster Set Up
+	n1(AU_EAST,AU_EAST_1,8vcpu): {s1:(256GiB)}
+	n2(AU_EAST,AU_EAST_1,8vcpu): {s2:(256GiB)}
+	n3(AU_EAST,AU_EAST_1,8vcpu): {s3:(256GiB)}
+	n4(AU_EAST,AU_EAST_1,8vcpu): {s4:(256GiB)}
+	n5(AU_EAST,AU_EAST_1,8vcpu): {s5:(256GiB)}
+	n6(AU_EAST,AU_EAST_1,8vcpu): {s6:(256GiB)}
+	n7(AU_EAST,AU_EAST_1,8vcpu): {s7:(256GiB)}
+Key Space
+	[0,10000): 7(rf=3), 0MiB, [s1:(7,4*),s2:(6,2*),s3:(3,0*),s4:(2,0*),s5:(1,0*),s6:(1,1*),s7:(1,0*)]
+Event
+	set LBRebalancingMode to 2
+Workload Set Up
+	[0,10000): 95%r large-block [128-256B/op, 7000ops/s]
+Changed Settings
+	empty
+==========================
 
-# The cluster settings and simulation variables can be modified to examine how
-# allocation behaves during uncommon scenarios. Update the gossip delay to be
-# unreasonably long (default 500ms).
+# Update the gossip delay to be unreasonably long (default 500ms) to assess
+# uncommon scenarios.
 setting gossip_delay=20s
 ----
 

pkg/kv/kvserver/asim/tests/testdata/non_rand/sma/splitting.txt

@@ -1,3 +1,8 @@
+# This test verifies load-based splitting behavior in a single-store cluster.
+# It tests both uniform and zipfian access patterns to ensure the split
+# algorithm works correctly under different load distributions. With a 10k QPS
+# load and 2.5k QPS split threshold, we expect approximately 4 splits for
+# uniform access and more splits for zipfian due to hotspotting.
 skip_under_ci
 ----