Planner wait conditions

dev-tools/omdb/src/bin/omdb/nexus.rs

@@ -45,6 +45,7 @@ use nexus_types::deployment::ClickhouseMode;
 use nexus_types::deployment::ClickhousePolicy;
 use nexus_types::deployment::OximeterReadMode;
 use nexus_types::deployment::OximeterReadPolicy;
+use nexus_types::deployment::WaitCondition;
 use nexus_types::internal_api::background::AbandonedVmmReaperStatus;
 use nexus_types::internal_api::background::BlueprintPlannerStatus;
 use nexus_types::internal_api::background::BlueprintRendezvousStatus;
@@ -1218,6 +1219,13 @@ fn print_task_abandoned_vmm_reaper(details: &serde_json::Value) {
 }
 
 fn print_task_blueprint_planner(details: &serde_json::Value) {
+    fn print_waiting_on(waiting_on: &[WaitCondition]) {
+        let n = waiting_on.len();
+        if n > 0 {
+            println!("    waiting on {n} events: {waiting_on:?}");
+        }
+    }
+
     let status =
         match serde_json::from_value::<BlueprintPlannerStatus>(details.clone())
         {
@@ -1237,20 +1245,32 @@ fn print_task_blueprint_planner(details: &serde_json::Value) {
         BlueprintPlannerStatus::Error(error) => {
             println!("    task did not complete successfully: {error}");
         }
-        BlueprintPlannerStatus::Unchanged { parent_blueprint_id } => {
+        BlueprintPlannerStatus::Unchanged {
+            parent_blueprint_id,
+            waiting_on,
+        } => {
             println!("    plan unchanged from parent {parent_blueprint_id}");
+            print_waiting_on(&waiting_on);
         }
-        BlueprintPlannerStatus::Planned { parent_blueprint_id, error } => {
+        BlueprintPlannerStatus::Planned {
+            parent_blueprint_id,
+            error,
+            waiting_on,
+        } => {
             println!(
                 "    planned new blueprint from parent {parent_blueprint_id}, \
                      but could not make it the target: {error}"
             );
+            print_waiting_on(&waiting_on);
         }
-        BlueprintPlannerStatus::Targeted { blueprint_id, .. } => {
+        BlueprintPlannerStatus::Targeted {
+            blueprint_id, waiting_on, ..
+        } => {
             println!(
                 "    planned new blueprint {blueprint_id}, \
                      and made it the current target"
             );
+            print_waiting_on(&waiting_on);
         }
     }
 }

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

@@ -26,6 +26,7 @@ use nexus_reconfigurator_simulation::{BlueprintId, SimState};
 use nexus_types::deployment::OmicronZoneNic;
 use nexus_types::deployment::PlanningInput;
 use nexus_types::deployment::SledFilter;
+use nexus_types::deployment::ZoneExpungeReason;
 use nexus_types::deployment::execution;
 use nexus_types::deployment::execution::blueprint_external_dns_config;
 use nexus_types::deployment::execution::blueprint_internal_dns_config;
@@ -1249,7 +1250,11 @@ fn cmd_blueprint_edit(
         BlueprintEditCommands::ExpungeZone { zone_id } => {
             let sled_id = sled_with_zone(&builder, &zone_id)?;
             builder
-                .sled_expunge_zone(sled_id, zone_id)
+                .sled_expunge_zone(
+                    sled_id,
+                    zone_id,
+                    &ZoneExpungeReason::ManualEdit,
+                )
                 .context("failed to expunge zone")?;
             format!("expunged zone {zone_id} from sled {sled_id}")
         }

dev-tools/reconfigurator-cli/tests/output/cmds-example-stdout

@@ -505,8 +505,7 @@ INFO sufficient ExternalDns zones exist in plan, desired_count: 0, current_count
 WARN failed to place all new desired Nexus zones, placed: 0, wanted_to_place: 3
 INFO sufficient Oximeter zones exist in plan, desired_count: 0, current_count: 0
 WARN cannot issue more SP updates (no current artifacts)
-INFO some zones not yet up-to-date, sled_id: 89d02b1b-478c-401a-8e28-7a26f74fa41b
-INFO will ensure cockroachdb setting, setting: cluster.preserve_downgrade_option, value: DoNotModify
+INFO some zones not yet up-to-date
 generated blueprint 86db3308-f817-4626-8838-4085949a6a41 based on parent blueprint ade5749d-bdf3-4fab-a8ae-00bea01b3a5a
 
 > blueprint-list

dev-tools/reconfigurator-cli/tests/output/cmds-expunge-newly-added-external-dns-stdout

@@ -980,8 +980,7 @@ INFO added zone to sled, sled_id: 711ac7f8-d19e-4572-bdb9-e9b50f6e362a, kind: Ex
 INFO sufficient Nexus zones exist in plan, desired_count: 3, current_count: 3
 INFO sufficient Oximeter zones exist in plan, desired_count: 0, current_count: 0
 WARN cannot issue more SP updates (no current artifacts)
-INFO some zones not yet up-to-date, sled_id: 711ac7f8-d19e-4572-bdb9-e9b50f6e362a
-INFO will ensure cockroachdb setting, setting: cluster.preserve_downgrade_option, value: DoNotModify
+INFO some zones not yet up-to-date
 generated blueprint 9c998c1d-1a7b-440a-ae0c-40f781dea6e2 based on parent blueprint 366b0b68-d80e-4bc1-abd3-dc69837847e0
 
 > blueprint-diff 366b0b68-d80e-4bc1-abd3-dc69837847e0 9c998c1d-1a7b-440a-ae0c-40f781dea6e2

dev-tools/reconfigurator-cli/tests/output/cmds-expunge-newly-added-internal-dns-stdout

@@ -1013,8 +1013,7 @@ INFO sufficient ExternalDns zones exist in plan, desired_count: 3, current_count
 INFO sufficient Nexus zones exist in plan, desired_count: 3, current_count: 3
 INFO sufficient Oximeter zones exist in plan, desired_count: 0, current_count: 0
 WARN cannot issue more SP updates (no current artifacts)
-INFO some zones not yet up-to-date, sled_id: 2b8f0cb3-0295-4b3c-bc58-4fe88b57112c
-INFO will ensure cockroachdb setting, setting: cluster.preserve_downgrade_option, value: DoNotModify
+INFO some zones not yet up-to-date
 generated blueprint af934083-59b5-4bf6-8966-6fb5292c29e1 based on parent blueprint 58d5e830-0884-47d8-a7cd-b2b3751adeb4
 
 > blueprint-diff 58d5e830-0884-47d8-a7cd-b2b3751adeb4 af934083-59b5-4bf6-8966-6fb5292c29e1

dev-tools/reconfigurator-cli/tests/output/target-release-stdout

@@ -170,7 +170,6 @@ INFO sufficient Nexus zones exist in plan, desired_count: 3, current_count: 3
 INFO sufficient Oximeter zones exist in plan, desired_count: 0, current_count: 0
 INFO configuring SP update, artifact_version: 1.0.0, artifact_hash: 7e6667e646ad001b54c8365a3d309c03f89c59102723d38d01697ee8079fe670, expected_inactive_version: NoValidVersion, expected_active_version: 0.0.1, component: sp, sp_slot: 0, sp_type: Sled, serial_number: serial0, part_number: model0
 INFO reached maximum number of pending SP updates, max: 1
-INFO will ensure cockroachdb setting, setting: cluster.preserve_downgrade_option, value: DoNotModify
 generated blueprint 8da82a8e-bf97-4fbd-8ddd-9f6462732cf1 based on parent blueprint dbcbd3d6-41ff-48ae-ac0b-1becc9b2fd21
 
 > blueprint-diff dbcbd3d6-41ff-48ae-ac0b-1becc9b2fd21 8da82a8e-bf97-4fbd-8ddd-9f6462732cf1
@@ -352,7 +351,6 @@ INFO sufficient Nexus zones exist in plan, desired_count: 3, current_count: 3
 INFO sufficient Oximeter zones exist in plan, desired_count: 0, current_count: 0
 INFO SP update not yet completed (will keep it), artifact_version: 1.0.0, artifact_hash: 7e6667e646ad001b54c8365a3d309c03f89c59102723d38d01697ee8079fe670, expected_inactive_version: NoValidVersion, expected_active_version: 0.0.1, component: sp, sp_slot: 0, sp_type: Sled, serial_number: serial0, part_number: model0
 INFO reached maximum number of pending SP updates, max: 1
-INFO will ensure cockroachdb setting, setting: cluster.preserve_downgrade_option, value: DoNotModify
 generated blueprint 58d5e830-0884-47d8-a7cd-b2b3751adeb4 based on parent blueprint 8da82a8e-bf97-4fbd-8ddd-9f6462732cf1
 
 > blueprint-diff 8da82a8e-bf97-4fbd-8ddd-9f6462732cf1 58d5e830-0884-47d8-a7cd-b2b3751adeb4
@@ -537,7 +535,6 @@ INFO SP update completed (will remove it and re-evaluate board), artifact_versio
 INFO skipping board for SP update, serial_number: serial0, part_number: model0
 INFO configuring SP update, artifact_version: 1.0.0, artifact_hash: 7e6667e646ad001b54c8365a3d309c03f89c59102723d38d01697ee8079fe670, expected_inactive_version: NoValidVersion, expected_active_version: 0.0.1, component: sp, sp_slot: 1, sp_type: Sled, serial_number: serial1, part_number: model1
 INFO reached maximum number of pending SP updates, max: 1
-INFO will ensure cockroachdb setting, setting: cluster.preserve_downgrade_option, value: DoNotModify
 generated blueprint af934083-59b5-4bf6-8966-6fb5292c29e1 based on parent blueprint 58d5e830-0884-47d8-a7cd-b2b3751adeb4
 
 > blueprint-diff 58d5e830-0884-47d8-a7cd-b2b3751adeb4 af934083-59b5-4bf6-8966-6fb5292c29e1
@@ -729,7 +726,6 @@ INFO sufficient Oximeter zones exist in plan, desired_count: 0, current_count: 0
 INFO SP update impossible (will remove it and re-evaluate board), artifact_version: 1.0.0, artifact_hash: 7e6667e646ad001b54c8365a3d309c03f89c59102723d38d01697ee8079fe670, expected_inactive_version: NoValidVersion, expected_active_version: 0.0.1, component: sp, sp_slot: 1, sp_type: Sled, serial_number: serial1, part_number: model1
 INFO configuring SP update, artifact_version: 1.0.0, artifact_hash: 7e6667e646ad001b54c8365a3d309c03f89c59102723d38d01697ee8079fe670, expected_inactive_version: Version(ArtifactVersion("0.5.0")), expected_active_version: 0.0.1, component: sp, sp_slot: 1, sp_type: Sled, serial_number: serial1, part_number: model1
 INFO reached maximum number of pending SP updates, max: 1
-INFO will ensure cockroachdb setting, setting: cluster.preserve_downgrade_option, value: DoNotModify
 generated blueprint df06bb57-ad42-4431-9206-abff322896c7 based on parent blueprint af934083-59b5-4bf6-8966-6fb5292c29e1
 
 > blueprint-diff af934083-59b5-4bf6-8966-6fb5292c29e1 df06bb57-ad42-4431-9206-abff322896c7
@@ -922,7 +918,6 @@ INFO skipping board for SP update, serial_number: serial1, part_number: model1
 INFO skipping board for SP update, serial_number: serial0, part_number: model0
 INFO configuring SP update, artifact_version: 1.0.0, artifact_hash: 7e6667e646ad001b54c8365a3d309c03f89c59102723d38d01697ee8079fe670, expected_inactive_version: NoValidVersion, expected_active_version: 0.0.1, component: sp, sp_slot: 2, sp_type: Sled, serial_number: serial2, part_number: model2
 INFO ran out of boards for SP update
-INFO will ensure cockroachdb setting, setting: cluster.preserve_downgrade_option, value: DoNotModify
 generated blueprint 7f976e0d-d2a5-4eeb-9e82-c82bc2824aba based on parent blueprint df06bb57-ad42-4431-9206-abff322896c7
 
 > blueprint-diff df06bb57-ad42-4431-9206-abff322896c7 7f976e0d-d2a5-4eeb-9e82-c82bc2824aba

live-tests/tests/test_nexus_add_remove.rs

@@ -19,6 +19,7 @@ use nexus_reconfigurator_planning::planner::Planner;
 use nexus_reconfigurator_preparation::PlanningInputFromDb;
 use nexus_sled_agent_shared::inventory::ZoneKind;
 use nexus_types::deployment::SledFilter;
+use nexus_types::deployment::ZoneExpungeReason;
 use omicron_common::address::NEXUS_INTERNAL_PORT;
 use omicron_test_utils::dev::poll::CondCheckError;
 use omicron_test_utils::dev::poll::wait_for_condition;
@@ -123,7 +124,11 @@ async fn test_nexus_add_remove(lc: &LiveTestContext) {
         &nexus,
         &|builder: &mut BlueprintBuilder| {
             builder
-                .sled_expunge_zone(sled_id, new_zone.id)
+                .sled_expunge_zone(
+                    sled_id,
+                    new_zone.id,
+                    &ZoneExpungeReason::Test,
+                )
                 .context("expunging zone")?;
             Ok(())
         },

nexus/reconfigurator/planning/src/blueprint_builder/builder.rs

@@ -14,7 +14,6 @@ use crate::blueprint_editor::ExternalSnatNetworkingChoice;
 use crate::blueprint_editor::NoAvailableDnsSubnets;
 use crate::blueprint_editor::SledEditError;
 use crate::blueprint_editor::SledEditor;
-use crate::planner::ZoneExpungeReason;
 use crate::planner::rng::PlannerRng;
 use anyhow::Context as _;
 use anyhow::anyhow;
@@ -45,6 +44,7 @@ use nexus_types::deployment::PendingMgsUpdates;
 use nexus_types::deployment::PlanningInput;
 use nexus_types::deployment::SledFilter;
 use nexus_types::deployment::SledResources;
+use nexus_types::deployment::ZoneExpungeReason;
 use nexus_types::deployment::ZpoolFilter;
 use nexus_types::deployment::ZpoolName;
 use nexus_types::deployment::blueprint_zone_type;
@@ -337,6 +337,13 @@ impl fmt::Display for Operation {
                     ZoneExpungeReason::ClickhouseSingleNodeDisabled => {
                         "clickhouse single-node disabled via policy"
                     }
+                    ZoneExpungeReason::ManualEdit => {
+                        "blueprint edited manually"
+                    }
+                    ZoneExpungeReason::Test => "for testing purposes",
+                    ZoneExpungeReason::UpdatedSource { from, to } => {
+                        &format!("updating from image source {from:?} → {to:?}")
+                    }
                 };
                 write!(
                     f,
@@ -1741,6 +1748,7 @@ impl<'a> BlueprintBuilder<'a> {
         &mut self,
         sled_id: SledUuid,
         zone_id: OmicronZoneUuid,
+        _reason: &ZoneExpungeReason,
     ) -> Result<SledEditCounts, Error> {
         let editor = self.sled_editors.get_mut(&sled_id).ok_or_else(|| {
             Error::Planner(anyhow!(