assembly of reconfigurator state should have limits (#9197)

Author: davepacheco

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

@@ -65,6 +65,10 @@ enum ReconfiguratorCommands {
 
 #[derive(Debug, Args, Clone)]
 struct ExportArgs {
+    /// maximum number of blueprints to save
+    #[clap(long, default_value_t = 1000)]
+    nmax_blueprints: usize,
+
     /// where to save the output
     output_file: Utf8PathBuf,
 }
@@ -147,21 +151,30 @@ async fn cmd_reconfigurator_export(
 ) -> anyhow::Result<UnstableReconfiguratorState> {
     // See Nexus::blueprint_planning_context().
     eprint!("assembling reconfigurator state ... ");
+    let limit = export_args.nmax_blueprints;
     let state = nexus_reconfigurator_preparation::reconfigurator_state_load(
-        opctx, datastore,
+        opctx, datastore, limit,
     )
     .await?;
     eprintln!("done");
 
+    if state.blueprints.len() >= limit {
+        eprintln!(
+            "warning: reached limit of {limit} while fetching blueprints"
+        );
+        eprintln!("warning: saving only the most recent {limit}");
+    }
+
     let output_path = &export_args.output_file;
+    eprint!("saving to {} ... ", output_path);
     let file = std::fs::OpenOptions::new()
         .create_new(true)
         .write(true)
         .open(&output_path)
         .with_context(|| format!("open {:?}", output_path))?;
     serde_json::to_writer_pretty(&file, &state)
         .with_context(|| format!("write {:?}", output_path))?;
-    eprintln!("wrote {}", output_path);
+    eprintln!("done");
     Ok(state)
 }
 
@@ -199,7 +212,7 @@ async fn cmd_reconfigurator_archive(
 
     let mut ndeleted = 0;
 
-    eprintln!("removing non-target blueprints ...");
+    eprintln!("removing saved, non-target blueprints ...");
     for blueprint in &saved_state.blueprints {
         if blueprint.id == target_blueprint_id {
             continue;
@@ -245,6 +258,15 @@ async fn cmd_reconfigurator_archive(
         eprintln!("done ({ndeleted} blueprint{plural} deleted)",);
     }
 
+    if saved_state.blueprints.len() >= archive_args.nmax_blueprints {
+        eprintln!(
+            "warning: Only tried deleting the most recent {} blueprints\n\
+             warning: because that's all that was fetched and saved.\n\
+             warning: You may want to run this tool again to archive more.",
+            saved_state.blueprints.len(),
+        );
+    }
+
     Ok(())
 }
 

dev-tools/omdb/tests/successes.out

@@ -1806,7 +1806,7 @@ stderr:
 note: using database URL postgresql://root@[::1]:REDACTED_PORT/omicron?sslmode=disable
 note: database schema version matches expected (<redacted database version>)
 assembling reconfigurator state ... done
-wrote <TMP_PATH_REDACTED>
+saving to <TMP_PATH_REDACTED> ... done
 =============================================
 EXECUTING COMMAND: omdb ["--destructive", "db", "db-metadata", "force-mark-nexus-quiesced", "--skip-confirmation", "--skip-blueprint-validation", "..........<REDACTED_UUID>..........."]
 termination: Exited(0)

nexus/db-queries/src/db/datastore/inventory.rs

@@ -4078,6 +4078,29 @@ impl DataStore {
             internal_dns_generation_status,
         })
     }
+
+    pub async fn inventory_collections_latest(
+        &self,
+        opctx: &OpContext,
+        count: u8,
+    ) -> anyhow::Result<Vec<InvCollection>> {
+        let limit: i64 = i64::from(count);
+        let conn = self
+            .pool_connection_authorized(opctx)
+            .await
+            .context("getting connection")?;
+
+        use nexus_db_schema::schema::inv_collection::dsl;
+        let collections = dsl::inv_collection
+            .select(InvCollection::as_select())
+            .order_by(dsl::time_started.desc())
+            .limit(limit)
+            .load_async(&*conn)
+            .await
+            .context("failed to list collections")?;
+
+        Ok(collections)
+    }
 }
 
 #[derive(Debug)]

nexus/reconfigurator/cli-integration-tests/tests/integration/blueprint_edit.rs

@@ -116,7 +116,7 @@ async fn test_blueprint_edit(cptestctx: &ControlPlaneTestContext) {
 
     // Assemble state that we can load into reconfigurator-cli.
     let state1 = nexus_reconfigurator_preparation::reconfigurator_state_load(
-        &opctx, datastore,
+        &opctx, datastore, 20,
     )
     .await
     .expect("failed to assemble reconfigurator state");

nexus/reconfigurator/preparation/src/lib.rs

@@ -12,7 +12,6 @@ use nexus_db_model::Generation;
 use nexus_db_queries::context::OpContext;
 use nexus_db_queries::db::DataStore;
 use nexus_db_queries::db::datastore::DataStoreDnsTest;
-use nexus_db_queries::db::datastore::DataStoreInventoryTest;
 use nexus_db_queries::db::datastore::Discoverability;
 use nexus_db_queries::db::datastore::SQL_BATCH_SIZE;
 use nexus_db_queries::db::pagination::Paginator;
@@ -57,6 +56,7 @@ use omicron_uuid_kinds::OmicronZoneUuid;
 use slog::Logger;
 use slog::error;
 use slog_error_chain::InlineErrorChain;
+use std::cmp::Reverse;
 use std::collections::BTreeMap;
 use std::collections::BTreeSet;
 
@@ -406,12 +406,13 @@ async fn fetch_all_service_ip_pool_ranges(
     Ok(ranges)
 }
 
-/// Loads state for import into `reconfigurator-cli`
+/// Loads state for debugging or import into `reconfigurator-cli`
 ///
-/// This is only to be used in omdb or tests.
+/// This is used in omdb, tests, and in Nexus to collect support bundles
 pub async fn reconfigurator_state_load(
     opctx: &OpContext,
     datastore: &DataStore,
+    nmax_blueprints: usize,
 ) -> Result<UnstableReconfiguratorState, anyhow::Error> {
     opctx.check_complex_operations_allowed()?;
     let planner_config = datastore
@@ -420,8 +421,11 @@ pub async fn reconfigurator_state_load(
         .map_or_else(PlannerConfig::default, |c| c.config.planner_config);
     let planning_input =
         PlanningInputFromDb::assemble(opctx, datastore, planner_config).await?;
+
+    // We'll grab the most recent several inventory collections.
+    const NCOLLECTIONS: u8 = 5;
     let collection_ids = datastore
-        .inventory_collections()
+        .inventory_collections_latest(opctx, NCOLLECTIONS)
         .await
         .context("listing collections")?
         .into_iter()
@@ -439,11 +443,13 @@ pub async fn reconfigurator_state_load(
         .collect::<Vec<Collection>>()
         .await;
 
+    // Grab the latest target blueprint.
     let target_blueprint = datastore
         .blueprint_target_get_current(opctx)
         .await
         .context("failed to read current target blueprint")?;
 
+    // Paginate through the list of all blueprints.
     let mut blueprint_ids = Vec::new();
     let mut paginator = Paginator::new(
         SQL_BATCH_SIZE,
@@ -460,6 +466,13 @@ pub async fn reconfigurator_state_load(
         blueprint_ids.extend(batch.into_iter());
     }
 
+    // We'll only grab the most recent blueprints that fit within the limit that
+    // we were given.  This is a heuristic intended to grab what's most likely
+    // to be useful even when the system has a large number of blueprints.  But
+    // the intent is that callers provide a limit that should be large enough to
+    // cover everything.
+    blueprint_ids.sort_by_key(|bpm| Reverse(bpm.time_created));
+    let blueprint_ids = blueprint_ids.into_iter().take(nmax_blueprints);
     let blueprints = futures::stream::iter(blueprint_ids)
         .filter_map(|bpm| async move {
             let blueprint_id = bpm.id.into_untyped_uuid();