Create a new version of StartSledAgentRequest (#4407)

In preparation for sled-addition, a new version of `StartSledAgentRequest` was created and the old version coverted to V0. Deterministic conversion from v0 to v1 is possible because we know that `ntp_servers` and `dns_servers` fields are not in use. We also know the values of the new fields. `is_lrtq_learner` must be false because we only support RSS addition on existing sleds. We then implement a deserialize method that will return the new version regardless of what's on disk. This is similar to how we handled the bootstore upgrade for `EarlyNetworkConfig`. Also similar to that, we extended the format of `StartSledAgentRequest` to ease future deserializations. In those deserializations we can deserialize the "header" and use [serde_json::value::RawValue](https://docs.rs/serde_json/latest/serde_json/value/struct.RawValue.html) in order to defer deserialization of the `body` field until we know the `schema_version`.
oxidecomputer · Nov 7, 2023 · 3445996 · 3445996
1 parent bd05e19
commit 3445996
Show file tree

Hide file tree

Showing 9 changed files with 270 additions and 119 deletions.
diff --git a/common/src/ledger.rs b/common/src/ledger.rs
@@ -7,7 +7,7 @@
 use async_trait::async_trait;
 use camino::{Utf8Path, Utf8PathBuf};
 use serde::{de::DeserializeOwned, Serialize};
-use slog::{debug, warn, Logger};
+use slog::{error, info, warn, Logger};
 
 #[derive(thiserror::Error, Debug)]
 pub enum Error {
@@ -84,8 +84,11 @@ impl<T: Ledgerable> Ledger<T> {
         // Read all the ledgers that we can.
         let mut ledgers = vec![];
         for path in paths.iter() {
-            if let Ok(ledger) = T::read_from(log, &path).await {
-                ledgers.push(ledger);
+            match T::read_from(log, &path).await {
+                Ok(ledger) => ledgers.push(ledger),
+                Err(err) => {
+                    error!(log, "Failed to read ledger: {err}"; "path" => %path)
+                }
             }
         }
 
@@ -169,8 +172,8 @@ pub trait Ledgerable: DeserializeOwned + Serialize + Send + Sync {
     /// Reads from `path` as a json-serialized version of `Self`.
     async fn read_from(log: &Logger, path: &Utf8Path) -> Result<Self, Error> {
         if path.exists() {
-            debug!(log, "Reading ledger from {}", path);
-            serde_json::from_str(
+            info!(log, "Reading ledger from {}", path);
+            <Self as Ledgerable>::deserialize(
                 &tokio::fs::read_to_string(&path)
                     .await
                     .map_err(|err| Error::io_path(&path, err))?,
@@ -180,7 +183,7 @@ pub trait Ledgerable: DeserializeOwned + Serialize + Send + Sync {
                 err,
             })
         } else {
-            debug!(log, "No ledger in {path}");
+            warn!(log, "No ledger in {path}");
             Err(Error::NotFound)
         }
     }
@@ -191,7 +194,7 @@ pub trait Ledgerable: DeserializeOwned + Serialize + Send + Sync {
         log: &Logger,
         path: &Utf8Path,
     ) -> Result<(), Error> {
-        debug!(log, "Writing ledger to {}", path);
+        info!(log, "Writing ledger to {}", path);
         let as_str = serde_json::to_string(&self).map_err(|err| {
             Error::JsonSerialize { path: path.to_path_buf(), err }
         })?;
@@ -200,6 +203,10 @@ pub trait Ledgerable: DeserializeOwned + Serialize + Send + Sync {
             .map_err(|err| Error::io_path(&path, err))?;
         Ok(())
     }
+
+    fn deserialize(s: &str) -> Result<Self, serde_json::Error> {
+        serde_json::from_str(s)
+    }
 }
 
 #[cfg(test)]

diff --git a/schema/rss-sled-plan.json b/schema/rss-sled-plan.json
@@ -588,33 +588,46 @@
       "description": "Configuration information for launching a Sled Agent.",
       "type": "object",
       "required": [
-        "dns_servers",
+        "body",
+        "generation",
+        "schema_version"
+      ],
+      "properties": {
+        "body": {
+          "$ref": "#/definitions/StartSledAgentRequestBody"
+        },
+        "generation": {
+          "description": "The current generation number of data as stored in CRDB.\n\nThe initial generation is set during RSS time and then only mutated by Nexus. For now, we don't actually anticipate mutating this data, but we leave open the possiblity.",
+          "type": "integer",
+          "format": "uint64",
+          "minimum": 0.0
+        },
+        "schema_version": {
+          "type": "integer",
+          "format": "uint32",
+          "minimum": 0.0
+        }
+      }
+    },
+    "StartSledAgentRequestBody": {
+      "description": "This is the actual app level data of `StartSledAgentRequest`\n\nWe nest it below the \"header\" of `generation` and `schema_version` so that we can perform partial deserialization of `EarlyNetworkConfig` to only read the header and defer deserialization of the body once we know the schema version. This is possible via the use of [`serde_json::value::RawValue`] in future (post-v1) deserialization paths.",
+      "type": "object",
+      "required": [
         "id",
-        "ntp_servers",
+        "is_lrtq_learner",
         "rack_id",
         "subnet",
         "use_trust_quorum"
       ],
       "properties": {
-        "dns_servers": {
-          "description": "The external DNS servers to use",
-          "type": "array",
-          "items": {
-            "type": "string",
-            "format": "ip"
-          }
-        },
         "id": {
           "description": "Uuid of the Sled Agent to be created.",
           "type": "string",
           "format": "uuid"
         },
-        "ntp_servers": {
-          "description": "The external NTP servers to use",
-          "type": "array",
-          "items": {
-            "type": "string"
-          }
+        "is_lrtq_learner": {
+          "description": "Is this node an LRTQ learner node?\n\nWe only put the node into learner mode if `use_trust_quorum` is also true.",
+          "type": "boolean"
         },
         "rack_id": {
           "description": "Uuid of the rack to which this sled agent belongs.",

diff --git a/schema/persistent-sled-agent-request.json → schema/start-sled-agent-request.json b/schema/persistent-sled-agent-request.json → schema/start-sled-agent-request.json
@@ -1,13 +1,27 @@
 {
   "$schema": "http://json-schema.org/draft-07/schema#",
-  "title": "PersistentSledAgentRequest",
+  "title": "StartSledAgentRequest",
+  "description": "Configuration information for launching a Sled Agent.",
   "type": "object",
   "required": [
-    "request"
+    "body",
+    "generation",
+    "schema_version"
   ],
   "properties": {
-    "request": {
-      "$ref": "#/definitions/StartSledAgentRequest"
+    "body": {
+      "$ref": "#/definitions/StartSledAgentRequestBody"
+    },
+    "generation": {
+      "description": "The current generation number of data as stored in CRDB.\n\nThe initial generation is set during RSS time and then only mutated by Nexus. For now, we don't actually anticipate mutating this data, but we leave open the possiblity.",
+      "type": "integer",
+      "format": "uint64",
+      "minimum": 0.0
+    },
+    "schema_version": {
+      "type": "integer",
+      "format": "uint32",
+      "minimum": 0.0
     }
   },
   "definitions": {
@@ -32,37 +46,25 @@
         }
       }
     },
-    "StartSledAgentRequest": {
-      "description": "Configuration information for launching a Sled Agent.",
+    "StartSledAgentRequestBody": {
+      "description": "This is the actual app level data of `StartSledAgentRequest`\n\nWe nest it below the \"header\" of `generation` and `schema_version` so that we can perform partial deserialization of `EarlyNetworkConfig` to only read the header and defer deserialization of the body once we know the schema version. This is possible via the use of [`serde_json::value::RawValue`] in future (post-v1) deserialization paths.",
       "type": "object",
       "required": [
-        "dns_servers",
         "id",
-        "ntp_servers",
+        "is_lrtq_learner",
         "rack_id",
         "subnet",
         "use_trust_quorum"
       ],
       "properties": {
-        "dns_servers": {
-          "description": "The external DNS servers to use",
-          "type": "array",
-          "items": {
-            "type": "string",
-            "format": "ip"
-          }
-        },
         "id": {
           "description": "Uuid of the Sled Agent to be created.",
           "type": "string",
           "format": "uuid"
         },
-        "ntp_servers": {
-          "description": "The external NTP servers to use",
-          "type": "array",
-          "items": {
-            "type": "string"
-          }
+        "is_lrtq_learner": {
+          "description": "Is this node an LRTQ learner node?\n\nWe only put the node into learner mode if `use_trust_quorum` is also true.",
+          "type": "boolean"
         },
         "rack_id": {
           "description": "Uuid of the rack to which this sled agent belongs.",