[4/n] [sled-agent] add code to read mupdate overrides (#8081)

Don't make any decisions based on this at the moment -- simply read it
for now. Decisions will happen in upcoming commits.

The logic here became quite complex, including tests, and I'm wondering
if it would make sense to extract this into something more generic now
or in the future.

Author: sunshowers

Cargo.lock

@@ -365,6 +365,7 @@ byteorder = "1.5.0"
 bytes = "1.10.1"
 camino = { version = "1.1", features = ["serde1"] }
 camino-tempfile = "1.3.0"
+camino-tempfile-ext = { version = "0.3.0", features = ["assert-color"] }
 cancel-safe-futures = "0.1.5"
 cargo_metadata = "0.19.2"
 chacha20poly1305 = "0.10.1"

Cargo.toml

@@ -148,6 +148,7 @@ impl BootstrapAgentStartup {
             config.switch_zone_maghemite_links.clone(),
             long_running_task_handles.storage_manager.clone(),
             long_running_task_handles.zone_bundler.clone(),
+            long_running_task_handles.zone_image_resolver.clone(),
         );
 
         // Inform the hardware monitor that the service manager is ready

sled-agent/src/bootstrap/pre_server.rs

@@ -23,12 +23,15 @@ use crate::sled_agent::SledAgent;
 use crate::storage_monitor::{StorageMonitor, StorageMonitorHandle};
 use crate::zone_bundle::ZoneBundler;
 use bootstore::schemes::v0 as bootstore;
+use illumos_utils::zpool::ZpoolName;
 use key_manager::{KeyManager, StorageKeyRequester};
 use sled_agent_types::zone_bundle::CleanupContext;
+use sled_agent_zone_images::{ZoneImageSourceResolver, ZoneImageZpools};
 use sled_hardware::{HardwareManager, SledMode, UnparsedDisk};
 use sled_storage::config::MountConfig;
 use sled_storage::disk::RawSyntheticDisk;
 use sled_storage::manager::{StorageHandle, StorageManager};
+use sled_storage::resources::AllDisks;
 use slog::{Logger, info};
 use std::net::Ipv6Addr;
 use tokio::sync::oneshot;
@@ -59,6 +62,13 @@ pub struct LongRunningTaskHandles {
 
     // A reference to the object used to manage zone bundles
     pub zone_bundler: ZoneBundler,
+
+    /// Resolver for zone image sources.
+    ///
+    /// This isn't really implemented in the backend as a task per se, but it
+    /// looks like one from the outside, and is convenient to put here. (If it
+    /// had any async involved within it, it would be a task.)
+    pub zone_image_resolver: ZoneImageSourceResolver,
 }
 
 /// Spawn all long running tasks
@@ -95,18 +105,21 @@ pub async fn spawn_all_longrunning_tasks(
     // Wait for the boot disk so that we can work with any ledgers,
     // such as those needed by the bootstore and sled-agent
     info!(log, "Waiting for boot disk");
-    let (disk_id, _) = storage_manager.wait_for_boot_disk().await;
+    let (disk_id, boot_zpool) = storage_manager.wait_for_boot_disk().await;
     info!(log, "Found boot disk {:?}", disk_id);
 
+    let all_disks = storage_manager.get_latest_disks().await;
     let bootstore = spawn_bootstore_tasks(
         log,
-        &mut storage_manager,
+        &all_disks,
         &hardware_manager,
         global_zone_bootstrap_ip,
     )
     .await;
 
     let zone_bundler = spawn_zone_bundler_tasks(log, &mut storage_manager);
+    let zone_image_resolver =
+        make_zone_image_resolver(log, &all_disks, &boot_zpool);
 
     (
         LongRunningTaskHandles {
@@ -116,6 +129,7 @@ pub async fn spawn_all_longrunning_tasks(
             hardware_manager,
             bootstore,
             zone_bundler,
+            zone_image_resolver,
         },
         sled_agent_started_tx,
         service_manager_ready_tx,
@@ -195,13 +209,12 @@ fn spawn_hardware_monitor(
 
 async fn spawn_bootstore_tasks(
     log: &Logger,
-    storage_handle: &mut StorageHandle,
+    all_disks: &AllDisks,
     hardware_manager: &HardwareManager,
     global_zone_bootstrap_ip: Ipv6Addr,
 ) -> bootstore::NodeHandle {
-    let iter_all = storage_handle.get_latest_disks().await;
     let config = new_bootstore_config(
-        &iter_all,
+        all_disks,
         hardware_manager.baseboard(),
         global_zone_bootstrap_ip,
     )
@@ -233,6 +246,18 @@ fn spawn_zone_bundler_tasks(
     ZoneBundler::new(log, storage_handle.clone(), CleanupContext::default())
 }
 
+fn make_zone_image_resolver(
+    log: &Logger,
+    all_disks: &AllDisks,
+    boot_zpool: &ZpoolName,
+) -> ZoneImageSourceResolver {
+    let zpools = ZoneImageZpools {
+        root: &all_disks.mount_config().root,
+        all_m2_zpools: all_disks.all_m2_zpools(),
+    };
+    ZoneImageSourceResolver::new(log, &zpools, boot_zpool)
+}
+
 async fn upsert_synthetic_disks_if_needed(
     log: &Logger,
     storage_manager: &StorageHandle,

sled-agent/src/long_running_tasks.rs

@@ -929,6 +929,7 @@ impl ServiceManager {
         switch_zone_maghemite_links: Vec<PhysicalLink>,
         storage: StorageHandle,
         zone_bundler: ZoneBundler,
+        zone_image_resolver: ZoneImageSourceResolver,
     ) -> Self {
         Self::new_inner(
             log,
@@ -940,6 +941,7 @@ impl ServiceManager {
             switch_zone_maghemite_links,
             storage,
             zone_bundler,
+            zone_image_resolver,
             RealSystemApi::new(),
         )
     }
@@ -955,6 +957,7 @@ impl ServiceManager {
         switch_zone_maghemite_links: Vec<PhysicalLink>,
         storage: StorageHandle,
         zone_bundler: ZoneBundler,
+        zone_image_resolver: ZoneImageSourceResolver,
         system_api: Box<dyn SystemApi>,
     ) -> Self {
         let log = log.new(o!("component" => "ServiceManager"));
@@ -990,7 +993,7 @@ impl ServiceManager {
                     .switch_zone_bootstrap_ip,
                 storage,
                 zone_bundler,
-                zone_image_resolver: ZoneImageSourceResolver::new(),
+                zone_image_resolver,
                 ledger_directory_override: OnceLock::new(),
                 system_api,
             }),
@@ -5141,6 +5144,7 @@ mod illumos_tests {
 
     use nexus_sled_agent_shared::inventory::OmicronZoneImageSource;
     use omicron_uuid_kinds::OmicronZoneUuid;
+    use sled_agent_zone_images::ZoneImageZpools;
     use sled_storage::manager_test_harness::StorageManagerTestHarness;
     use std::{
         net::{Ipv6Addr, SocketAddrV6},
@@ -5342,6 +5346,7 @@ mod illumos_tests {
         log: slog::Logger,
         storage_test_harness: StorageManagerTestHarness,
         zone_bundler: ZoneBundler,
+        zone_image_resolver: ZoneImageSourceResolver,
         test_config: &'a TestConfig,
     }
 
@@ -5354,10 +5359,21 @@ mod illumos_tests {
                 Default::default(),
             );
 
+            let mut storage_manager = storage_test_harness.handle().clone();
+            let all_disks = storage_manager.get_latest_disks().await;
+            let (_, boot_zpool) = storage_manager.wait_for_boot_disk().await;
+            let zpools = ZoneImageZpools {
+                root: &all_disks.mount_config().root,
+                all_m2_zpools: all_disks.all_m2_zpools(),
+            };
+            let zone_image_resolver =
+                ZoneImageSourceResolver::new(&log, &zpools, &boot_zpool);
+
             LedgerTestHelper {
                 log,
                 storage_test_harness,
                 zone_bundler,
+                zone_image_resolver,
                 test_config,
             }
         }
@@ -5392,6 +5408,7 @@ mod illumos_tests {
                 vec![],
                 self.storage_test_harness.handle().clone(),
                 self.zone_bundler.clone(),
+                self.zone_image_resolver.clone(),
                 system,
             );
             self.test_config.override_paths(&mgr);

sled-agent/src/services.rs

@@ -10,6 +10,19 @@ workspace = true
 [dependencies]
 anyhow.workspace = true
 camino.workspace = true
+id-map.workspace = true
+illumos-utils.workspace = true
 nexus-sled-agent-shared.workspace = true
+omicron-common.workspace = true
 omicron-workspace-hack.workspace = true
+serde_json.workspace = true
 sled-storage.workspace = true
+slog.workspace = true
+slog-error-chain.workspace = true
+thiserror.workspace = true
+
+[dev-dependencies]
+camino-tempfile-ext.workspace = true
+dropshot.workspace = true
+omicron-uuid-kinds.workspace = true
+pretty_assertions.workspace = true

sled-agent/zone-images/Cargo.toml

@@ -7,6 +7,8 @@
 //! This contains a subset of zone image code at the moment: you're encouraged
 //! to move more code into this crate as appropriate.
 
+mod mupdate_override;
 mod source_resolver;
 
+pub(crate) use mupdate_override::*;
 pub use source_resolver::*;