[db] omdb db disk for physical disks (#4141)

New omdb command and database schema updates

This bumps the database version to 5.0.0
This adds a new subcommand: `physical <UUID>` to `omdb db disks`.  
This will show the disk resources that reside on a physical disk.

This change requires additional indexes in the zpool and dataset tables as
we lookup which zpool is on a physical disk, and we lookup which
datasets are on a given zpool. The usage here is just for the omdb tool,
but those same searches will need to happen when Omicron needs to figure out
what resources are on a physical disk.

Tested both initial install, and schema update
---------

Co-authored-by: Alan Hanson <[email protected]>

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

@@ -24,13 +24,16 @@ use clap::ValueEnum;
 use diesel::expression::SelectableHelper;
 use diesel::query_dsl::QueryDsl;
 use diesel::ExpressionMethods;
+use nexus_db_model::Dataset;
 use nexus_db_model::Disk;
 use nexus_db_model::DnsGroup;
 use nexus_db_model::DnsName;
 use nexus_db_model::DnsVersion;
 use nexus_db_model::DnsZone;
 use nexus_db_model::Instance;
+use nexus_db_model::Region;
 use nexus_db_model::Sled;
+use nexus_db_model::Zpool;
 use nexus_db_queries::context::OpContext;
 use nexus_db_queries::db;
 use nexus_db_queries::db::identity::Asset;
@@ -45,6 +48,7 @@ use omicron_common::api::external::Generation;
 use omicron_common::postgres_config::PostgresConfigWithUrl;
 use std::cmp::Ordering;
 use std::collections::BTreeMap;
+use std::collections::HashSet;
 use std::fmt::Display;
 use std::num::NonZeroU32;
 use std::sync::Arc;
@@ -96,6 +100,8 @@ enum DiskCommands {
     Info(DiskInfoArgs),
     /// Summarize current disks
     List,
+    /// Determine what crucible resources are on the given physical disk.
+    Physical(DiskPhysicalArgs),
 }
 
 #[derive(Debug, Args)]
@@ -104,6 +110,12 @@ struct DiskInfoArgs {
     uuid: Uuid,
 }
 
+#[derive(Debug, Args)]
+struct DiskPhysicalArgs {
+    /// The UUID of the physical disk
+    uuid: Uuid,
+}
+
 #[derive(Debug, Args)]
 struct DnsArgs {
     #[command(subcommand)]
@@ -214,6 +226,12 @@ impl DbArgs {
             DbCommands::Disks(DiskArgs { command: DiskCommands::List }) => {
                 cmd_db_disk_list(&datastore, self.fetch_limit).await
             }
+            DbCommands::Disks(DiskArgs {
+                command: DiskCommands::Physical(uuid),
+            }) => {
+                cmd_db_disk_physical(&opctx, &datastore, self.fetch_limit, uuid)
+                    .await
+            }
             DbCommands::Dns(DnsArgs { command: DnsCommands::Show }) => {
                 cmd_db_dns_show(&opctx, &datastore, self.fetch_limit).await
             }
@@ -509,6 +527,151 @@ async fn cmd_db_disk_info(
     Ok(())
 }
 
+/// Run `omdb db disk physical <UUID>`.
+async fn cmd_db_disk_physical(
+    opctx: &OpContext,
+    datastore: &DataStore,
+    limit: NonZeroU32,
+    args: &DiskPhysicalArgs,
+) -> Result<(), anyhow::Error> {
+    // We start by finding any zpools that are using the physical disk.
+    use db::schema::zpool::dsl as zpool_dsl;
+    let zpools = zpool_dsl::zpool
+        .filter(zpool_dsl::time_deleted.is_null())
+        .filter(zpool_dsl::physical_disk_id.eq(args.uuid))
+        .select(Zpool::as_select())
+        .load_async(datastore.pool_for_tests().await?)
+        .await
+        .context("loading zpool from pysical disk id")?;
+
+    let mut sled_ids = HashSet::new();
+    let mut dataset_ids = HashSet::new();
+
+    // The current plan is a single zpool per physical disk, so we expect that
+    // this will have a single item.  However, If single zpool per disk ever
+    // changes, this code will still work.
+    for zp in zpools {
+        // zpool has the sled id, record that so we can find the serial number.
+        sled_ids.insert(zp.sled_id);
+
+        // Next, we find all the datasets that are on our zpool.
+        use db::schema::dataset::dsl as dataset_dsl;
+        let datasets = dataset_dsl::dataset
+            .filter(dataset_dsl::time_deleted.is_null())
+            .filter(dataset_dsl::pool_id.eq(zp.id()))
+            .select(Dataset::as_select())
+            .load_async(datastore.pool_for_tests().await?)
+            .await
+            .context("loading dataset")?;
+
+        // Add all the datasets ids that are using this pool.
+        for ds in datasets {
+            dataset_ids.insert(ds.id());
+        }
+    }
+
+    // If we do have more than one sled ID, then something is wrong, but
+    // go ahead and print out whatever we have found.
+    for sid in sled_ids {
+        let (_, my_sled) = LookupPath::new(opctx, datastore)
+            .sled_id(sid)
+            .fetch()
+            .await
+            .context("failed to look up sled")?;
+
+        println!(
+            "Physical disk: {} found on sled: {}",
+            args.uuid,
+            my_sled.serial_number()
+        );
+    }
+
+    let mut volume_ids = HashSet::new();
+    // Now, take the list of datasets we found and search all the regions
+    // to see if any of them are on the dataset.  If we find a region that
+    // is on one of our datasets, then record the volume ID of that region.
+    for did in dataset_ids.clone().into_iter() {
+        use db::schema::region::dsl as region_dsl;
+        let regions = region_dsl::region
+            .filter(region_dsl::dataset_id.eq(did))
+            .select(Region::as_select())
+            .load_async(datastore.pool_for_tests().await?)
+            .await
+            .context("loading region")?;
+
+        for rs in regions {
+            volume_ids.insert(rs.volume_id());
+        }
+    }
+
+    // At this point, we have a list of volume IDs that contain a region
+    // that is part of a dataset on a pool on our disk.  The final step is
+    // to find the virtual disks associated with these volume IDs and
+    // display information about those disks.
+    use db::schema::disk::dsl;
+    let disks = dsl::disk
+        .filter(dsl::time_deleted.is_null())
+        .filter(dsl::volume_id.eq_any(volume_ids))
+        .limit(i64::from(u32::from(limit)))
+        .select(Disk::as_select())
+        .load_async(datastore.pool_for_tests().await?)
+        .await
+        .context("loading disks")?;
+
+    check_limit(&disks, limit, || "listing disks".to_string());
+
+    #[derive(Tabled)]
+    #[tabled(rename_all = "SCREAMING_SNAKE_CASE")]
+    struct DiskRow {
+        name: String,
+        id: String,
+        state: String,
+        instance_name: String,
+    }
+
+    let mut rows = Vec::new();
+
+    for disk in disks {
+        // If the disk is attached to an instance, determine the name of the
+        // instance.
+        let instance_name =
+            if let Some(instance_uuid) = disk.runtime().attach_instance_id {
+                // Get the instance this disk is attached to
+                use db::schema::instance::dsl as instance_dsl;
+                let instance = instance_dsl::instance
+                    .filter(instance_dsl::id.eq(instance_uuid))
+                    .limit(1)
+                    .select(Instance::as_select())
+                    .load_async(datastore.pool_for_tests().await?)
+                    .await
+                    .context("loading requested instance")?;
+
+                if let Some(instance) = instance.into_iter().next() {
+                    instance.name().to_string()
+                } else {
+                    "???".to_string()
+                }
+            } else {
+                "-".to_string()
+            };
+
+        rows.push(DiskRow {
+            name: disk.name().to_string(),
+            id: disk.id().to_string(),
+            state: disk.runtime().disk_state,
+            instance_name: instance_name,
+        });
+    }
+
+    let table = tabled::Table::new(rows)
+        .with(tabled::settings::Style::empty())
+        .with(tabled::settings::Padding::new(0, 1, 0, 0))
+        .to_string();
+
+    println!("{}", table);
+    Ok(())
+}
+
 // SERVICES
 
 #[derive(Tabled)]

dev-tools/omdb/tests/env.out

@@ -7,7 +7,7 @@ sim-b6d65341 [::1]:REDACTED_PORT -    REDACTED_UUID_REDACTED_UUID_REDACTED
 ---------------------------------------------
 stderr:
 note: using database URL postgresql://root@[::1]:REDACTED_PORT/omicron?sslmode=disable
-note: database schema version matches expected (4.0.0)
+note: database schema version matches expected (5.0.0)
 =============================================
 EXECUTING COMMAND: omdb ["db", "--db-url", "junk", "sleds"]
 termination: Exited(2)
@@ -172,7 +172,7 @@ stderr:
 note: database URL not specified.  Will search DNS.
 note: (override with --db-url or OMDB_DB_URL)
 note: using database URL postgresql://root@[::1]:REDACTED_PORT/omicron?sslmode=disable
-note: database schema version matches expected (4.0.0)
+note: database schema version matches expected (5.0.0)
 =============================================
 EXECUTING COMMAND: omdb ["--dns-server", "[::1]:REDACTED_PORT", "db", "sleds"]
 termination: Exited(0)
@@ -185,5 +185,5 @@ stderr:
 note: database URL not specified.  Will search DNS.
 note: (override with --db-url or OMDB_DB_URL)
 note: using database URL postgresql://root@[::1]:REDACTED_PORT/omicron?sslmode=disable
-note: database schema version matches expected (4.0.0)
+note: database schema version matches expected (5.0.0)
 =============================================

dev-tools/omdb/tests/successes.out

@@ -8,7 +8,7 @@ external oxide-dev.test               2   <REDACTED_TIMESTAMP> create silo: "tes
 ---------------------------------------------
 stderr:
 note: using database URL postgresql://root@[::1]:REDACTED_PORT/omicron?sslmode=disable
-note: database schema version matches expected (4.0.0)
+note: database schema version matches expected (5.0.0)
 =============================================
 EXECUTING COMMAND: omdb ["db", "dns", "diff", "external", "2"]
 termination: Exited(0)
@@ -24,7 +24,7 @@ changes:                    names added: 1, names removed: 0
 ---------------------------------------------
 stderr:
 note: using database URL postgresql://root@[::1]:REDACTED_PORT/omicron?sslmode=disable
-note: database schema version matches expected (4.0.0)
+note: database schema version matches expected (5.0.0)
 =============================================
 EXECUTING COMMAND: omdb ["db", "dns", "names", "external", "2"]
 termination: Exited(0)
@@ -36,7 +36,7 @@ External zone: oxide-dev.test
 ---------------------------------------------
 stderr:
 note: using database URL postgresql://root@[::1]:REDACTED_PORT/omicron?sslmode=disable
-note: database schema version matches expected (4.0.0)
+note: database schema version matches expected (5.0.0)
 =============================================
 EXECUTING COMMAND: omdb ["db", "services", "list-instances"]
 termination: Exited(0)
@@ -52,7 +52,7 @@ Nexus          REDACTED_UUID_REDACTED_UUID_REDACTED [::ffff:127.0.0.1]:REDACTED_
 ---------------------------------------------
 stderr:
 note: using database URL postgresql://root@[::1]:REDACTED_PORT/omicron?sslmode=disable
-note: database schema version matches expected (4.0.0)
+note: database schema version matches expected (5.0.0)
 =============================================
 EXECUTING COMMAND: omdb ["db", "services", "list-by-sled"]
 termination: Exited(0)
@@ -71,7 +71,7 @@ sled: sim-b6d65341 (id REDACTED_UUID_REDACTED_UUID_REDACTED)
 ---------------------------------------------
 stderr:
 note: using database URL postgresql://root@[::1]:REDACTED_PORT/omicron?sslmode=disable
-note: database schema version matches expected (4.0.0)
+note: database schema version matches expected (5.0.0)
 =============================================
 EXECUTING COMMAND: omdb ["db", "sleds"]
 termination: Exited(0)
@@ -82,7 +82,7 @@ sim-b6d65341 [::1]:REDACTED_PORT -    REDACTED_UUID_REDACTED_UUID_REDACTED
 ---------------------------------------------
 stderr:
 note: using database URL postgresql://root@[::1]:REDACTED_PORT/omicron?sslmode=disable
-note: database schema version matches expected (4.0.0)
+note: database schema version matches expected (5.0.0)
 =============================================
 EXECUTING COMMAND: omdb ["nexus", "background-tasks", "doc"]
 termination: Exited(0)

nexus/db-model/src/schema.rs

@@ -1130,7 +1130,7 @@ table! {
 ///
 /// This should be updated whenever the schema is changed. For more details,
 /// refer to: schema/crdb/README.adoc
-pub const SCHEMA_VERSION: SemverVersion = SemverVersion::new(4, 0, 0);
+pub const SCHEMA_VERSION: SemverVersion = SemverVersion::new(5, 0, 0);
 
 allow_tables_to_appear_in_same_query!(
     system_update,

schema/crdb/5.0.0/up1.sql

@@ -0,0 +1 @@
+CREATE INDEX IF NOT EXISTS lookup_zpool_by_disk ON omicron.public.zpool (physical_disk_id, id) WHERE physical_disk_id IS NOT NULL AND time_deleted IS NULL;

schema/crdb/5.0.0/up2.sql

@@ -0,0 +1 @@
+CREATE INDEX IF NOT EXISTS lookup_dataset_by_zpool ON omicron.public.dataset (pool_id, id) WHERE pool_id IS NOT NULL AND time_deleted IS NULL;

schema/crdb/dbinit.sql

@@ -388,6 +388,12 @@ CREATE TABLE IF NOT EXISTS omicron.public.zpool (
     total_size INT NOT NULL
 );
 
+/* Create an index on the physical disk id */
+CREATE INDEX IF NOT EXISTS lookup_zpool_by_disk on omicron.public.zpool (
+    physical_disk_id,
+    id
+) WHERE physical_disk_id IS NOT NULL AND time_deleted IS NULL;
+
 CREATE TYPE IF NOT EXISTS omicron.public.dataset_kind AS ENUM (
   'crucible',
   'cockroach',
@@ -437,6 +443,12 @@ CREATE INDEX IF NOT EXISTS lookup_dataset_by_size_used on omicron.public.dataset
     size_used
 ) WHERE size_used IS NOT NULL AND time_deleted IS NULL;
 
+/* Create an index on the zpool id */
+CREATE INDEX IF NOT EXISTS lookup_dataset_by_zpool on omicron.public.dataset (
+    pool_id,
+    id
+) WHERE pool_id IS NOT NULL AND time_deleted IS NULL;
+
 /*
  * A region of space allocated to Crucible Downstairs, within a dataset.
  */
@@ -2562,7 +2574,7 @@ INSERT INTO omicron.public.db_metadata (
     version,
     target_version
 ) VALUES
-    ( TRUE, NOW(), NOW(), '4.0.0', NULL)
+    ( TRUE, NOW(), NOW(), '5.0.0', NULL)
 ON CONFLICT DO NOTHING;
 
 COMMIT;