[reconfigurator-cli] Fixes and improvements for RoT update testing (#8904)

Follow up to #8893

Author: karencfv

dev-tools/reconfigurator-cli/src/lib.rs

@@ -10,6 +10,7 @@ use chrono::{DateTime, Utc};
 use clap::{ArgAction, ValueEnum};
 use clap::{Args, Parser, Subcommand};
 use daft::Diffable;
+use gateway_types::rot::RotSlot;
 use iddqd::IdOrdMap;
 use indent_write::fmt::IndentWriter;
 use internal_dns_types::diff::DnsDiff;
@@ -22,7 +23,7 @@ use nexus_reconfigurator_planning::blueprint_builder::BlueprintBuilder;
 use nexus_reconfigurator_planning::example::ExampleSystemBuilder;
 use nexus_reconfigurator_planning::planner::Planner;
 use nexus_reconfigurator_planning::system::{
-    SledBuilder, SledInventoryVisibility, SystemDescription,
+    RotStateOverrides, SledBuilder, SledInventoryVisibility, SystemDescription,
 };
 use nexus_reconfigurator_simulation::{BlueprintId, CollectionId, SimState};
 use nexus_reconfigurator_simulation::{SimStateBuilder, SimTufRepoSource};
@@ -553,6 +554,27 @@ struct SledUpdateRotArgs {
     /// sets the version reported for the RoT slot b
     #[clap(long, required_unless_present_any = &["slot_a"])]
     slot_b: Option<ExpectedVersion>,
+
+    /// sets whether we expect the "A" or "B" slot to be active
+    #[clap(long)]
+    active_slot: Option<RotSlot>,
+
+    /// sets the persistent boot preference written into the current
+    /// authoritative CFPA page (ping or pong).
+    #[clap(long)]
+    persistent_boot_preference: Option<RotSlot>,
+
+    /// sets the pending persistent boot preference written into the CFPA
+    /// scratch page that will become the persistent boot preference in the
+    /// authoritative CFPA page upon reboot, unless CFPA update of the
+    /// authoritative page fails for some reason
+    #[clap(long, num_args(0..=1))]
+    pending_persistent_boot_preference: Option<Option<RotSlot>>,
+
+    /// sets the transient boot preference, which overrides persistent
+    /// preference selection for a single boot (unimplemented)
+    #[clap(long, num_args(0..=1),)]
+    transient_boot_preference: Option<Option<RotSlot>>,
 }
 
 #[derive(Debug, Args)]
@@ -1706,6 +1728,31 @@ fn cmd_sled_update_rot(
     if let Some(slot_b) = &args.slot_b {
         labels.push(format!("slot b -> {}", slot_b));
     }
+    if let Some(active_slot) = &args.active_slot {
+        labels.push(format!("active slot -> {}", active_slot));
+    }
+
+    if let Some(persistent_boot_preference) = &args.persistent_boot_preference {
+        labels.push(format!(
+            "persistent boot preference -> {}",
+            persistent_boot_preference
+        ));
+    }
+
+    if let Some(pending_persistent_boot_preference) =
+        &args.pending_persistent_boot_preference
+    {
+        labels.push(format!(
+            "pending persistent boot preference -> {:?}",
+            pending_persistent_boot_preference
+        ));
+    }
+    if let Some(transient_boot_preference) = &args.transient_boot_preference {
+        labels.push(format!(
+            "transient boot preference -> {:?}",
+            transient_boot_preference
+        ));
+    }
 
     assert!(
         !labels.is_empty(),
@@ -1717,8 +1764,16 @@ fn cmd_sled_update_rot(
     let sled_id = args.sled_id.to_sled_id(system.description())?;
     system.description_mut().sled_update_rot_versions(
         sled_id,
-        args.slot_a,
-        args.slot_b,
+        RotStateOverrides {
+            active_slot_override: args.active_slot,
+            slot_a_version_override: args.slot_a,
+            slot_b_version_override: args.slot_b,
+            persistent_boot_preference_override: args
+                .persistent_boot_preference,
+            pending_persistent_boot_preference_override: args
+                .pending_persistent_boot_preference,
+            transient_boot_preference_override: args.transient_boot_preference,
+        },
     )?;
 
     sim.commit_and_bump(

dev-tools/reconfigurator-cli/tests/input/cmds-target-release.txt

@@ -52,7 +52,7 @@ blueprint-diff latest
 # reflect that update has completed as well. Like before, collect inventory from
 # it and use that collection for the next step. This should report that the
 # update completed, remove that update, and add one for another sled.
-sled-update-rot 98e6b7c2-2efa-41ca-b20a-0a4d61102fe6 --slot-a 1.0.0
+sled-update-rot 98e6b7c2-2efa-41ca-b20a-0a4d61102fe6 --slot-b 1.0.0 --active-slot b --persistent-boot-preference b
 inventory-generate
 blueprint-plan latest latest
 blueprint-diff latest
@@ -127,7 +127,7 @@ blueprint-diff latest
 
 # Now simulate the update completing successfully.
 # Like before, we should see a pending SP update for this sled.
-sled-update-rot 2b8f0cb3-0295-4b3c-bc58-4fe88b57112c --slot-a 1.0.0
+sled-update-rot 2b8f0cb3-0295-4b3c-bc58-4fe88b57112c --slot-b 1.0.0 --active-slot b --persistent-boot-preference b
 inventory-generate
 blueprint-plan latest latest
 blueprint-diff latest
@@ -175,9 +175,38 @@ inventory-generate
 blueprint-plan latest latest
 blueprint-diff latest
 
-# Update the RoT on the last sled.
+# Now let's simulate an RoT update that hasn't completed by setting the persistent
+# boot preference to a different slot than the active one. This should cause the
+# planner to mark it as impossible, and replace it.
+set ignore-impossible-mgs-updates-since now
+sled-update-rot d81c6a84-79b8-4958-ae41-ea46c9b19763 --slot-b 1.0.0 --active-slot a --persistent-boot-preference b
+inventory-generate
+blueprint-plan latest latest
+blueprint-diff latest
+
+# An RoT update cannot continue if pending persistent boot preference is set.
+# The slot B is set to a different version so it does not appear to be active
+# and on the correct version, and the corresponding checks are triggered.
+# Like before, this should cause the planner to mark it as impossible, and
+# replace it.
+set ignore-impossible-mgs-updates-since now
+sled-update-rot d81c6a84-79b8-4958-ae41-ea46c9b19763 --slot-b 1.1.0 --active-slot b --persistent-boot-preference b --pending-persistent-boot-preference b
+inventory-generate
+blueprint-plan latest latest
+blueprint-diff latest
+
+# We do the same for transient boot preference. Note that we need to set the
+# pending-persistent-boot-preference back to None
+set ignore-impossible-mgs-updates-since now
+sled-update-rot d81c6a84-79b8-4958-ae41-ea46c9b19763 --slot-b 1.1.0 --active-slot b --persistent-boot-preference b --transient-boot-preference b --pending-persistent-boot-preference
+inventory-generate
+blueprint-plan latest latest
+blueprint-diff latest
+
+# Update the RoT on the last sled. Note that we need to set the
+# transient-boot-preference back to None
 # There should be one last pending SP update.
-sled-update-rot d81c6a84-79b8-4958-ae41-ea46c9b19763 --slot-a 1.0.0
+sled-update-rot d81c6a84-79b8-4958-ae41-ea46c9b19763 --slot-b 1.0.0 --active-slot b --persistent-boot-preference b --transient-boot-preference
 inventory-generate
 blueprint-plan latest latest
 blueprint-diff latest

dev-tools/reconfigurator-cli/tests/input/cmds.txt

@@ -24,6 +24,14 @@ sled-update-rot dde1c0e2-b10d-4621-b420-f179f7a7a00a --slot-a 4.0.0 --slot-b inv
 sled-show dde1c0e2-b10d-4621-b420-f179f7a7a00a
 sled-update-rot dde1c0e2-b10d-4621-b420-f179f7a7a00a --slot-a 4.0.0 --slot-b 5.0.0
 sled-show dde1c0e2-b10d-4621-b420-f179f7a7a00a
+sled-update-rot dde1c0e2-b10d-4621-b420-f179f7a7a00a --slot-b 6.0.0 --active-slot b
+sled-show dde1c0e2-b10d-4621-b420-f179f7a7a00a
+sled-update-rot dde1c0e2-b10d-4621-b420-f179f7a7a00a --slot-b 6.0.0 --active-slot b --persistent-boot-preference b
+sled-show dde1c0e2-b10d-4621-b420-f179f7a7a00a
+sled-update-rot dde1c0e2-b10d-4621-b420-f179f7a7a00a --slot-b 6.0.0 --active-slot a --persistent-boot-preference a --pending-persistent-boot-preference b --transient-boot-preference a
+sled-show dde1c0e2-b10d-4621-b420-f179f7a7a00a
+sled-update-rot dde1c0e2-b10d-4621-b420-f179f7a7a00a --slot-a 7.0.0 --pending-persistent-boot-preference --transient-boot-preference
+sled-show dde1c0e2-b10d-4621-b420-f179f7a7a00a
 
 sled-update-sp dde1c0e2-b10d-4621-b420-f179f7a7a00a
 sled-update-sp dde1c0e2-b10d-4621-b420-f179f7a7a00a --active 1.0.0