propolis, aws: diagnostics for propolis factory, target-level overrides

Author: jclulow

Cargo.lock

@@ -299,13 +299,16 @@ async fn aws_worker_one(
                             i.id,
                             w.id
                         );
+                        let md = c.metadata(t).map_err(|e| {
+                            anyhow!("building worker metadata: {e}")
+                        })?;
                         c.client
                             .factory_worker_associate()
                             .worker(&w.id)
                             .body_map(|body| {
                                 body.private(&i.id)
                                     .ip(i.ip.clone())
-                                    .metadata(Some(c.metadata(t)))
+                                    .metadata(Some(md))
                             })
                             .send()
                             .await?;
@@ -540,6 +543,10 @@ async fn aws_worker_one(
             break;
         };
 
+        let md = c
+            .metadata(t)
+            .map_err(|e| anyhow!("building worker metadata: {e}"))?;
+
         let w = c
             .client
             .factory_worker_create()
@@ -562,9 +569,7 @@ async fn aws_worker_one(
         c.client
             .factory_worker_associate()
             .worker(&w.id)
-            .body_map(|body| {
-                body.private(&i.id).ip(i.ip).metadata(Some(c.metadata(t)))
-            })
+            .body_map(|body| body.private(&i.id).ip(i.ip).metadata(Some(md)))
             .send()
             .await?;
     }

factory/aws/src/aws.rs

@@ -4,6 +4,7 @@
 
 use std::collections::HashMap;
 
+use buildomat_types::config::ConfigFileDiag;
 use serde::Deserialize;
 
 #[derive(Deserialize, Debug)]
@@ -13,6 +14,8 @@ pub(crate) struct ConfigFile {
     pub general: ConfigFileGeneral,
     pub factory: ConfigFileFactory,
     pub target: HashMap<String, ConfigFileAwsTarget>,
+    #[serde(default)]
+    pub diag: ConfigFileDiag,
 }
 
 #[derive(Deserialize, Debug)]
@@ -33,8 +36,8 @@ pub(crate) struct ConfigFileAwsTarget {
     pub instance_type: String,
     pub root_size_gb: i32,
     pub ami: String,
-    pub post_job_diagnostic_script: Option<String>,
-    pub pre_job_diagnostic_script: Option<String>,
+    #[serde(default)]
+    pub diag: ConfigFileDiag,
 }
 
 #[derive(Deserialize, Debug)]
@@ -49,9 +52,6 @@ pub(crate) struct ConfigFileAws {
     pub key: String,
     pub security_group: String,
     pub limit_total: usize,
-    pub root_password_hash: Option<String>,
-    pub root_authorized_keys: Option<String>,
-    pub dump_to_rpool: Option<u32>,
 }
 
 impl ConfigFileAws {

factory/aws/src/config.rs

@@ -67,15 +67,12 @@ impl Central {
     fn metadata(
         &self,
         t: &config::ConfigFileAwsTarget,
-    ) -> metadata::FactoryMetadata {
-        metadata::FactoryMetadata::V1(metadata::FactoryMetadataV1 {
-            addresses: Default::default(),
-            root_password_hash: self.config.aws.root_password_hash.clone(),
-            root_authorized_keys: self.config.aws.root_authorized_keys.clone(),
-            dump_to_rpool: self.config.aws.dump_to_rpool,
-            post_job_diagnostic_script: t.post_job_diagnostic_script.clone(),
-            pre_job_diagnostic_script: t.pre_job_diagnostic_script.clone(),
-        })
+    ) -> Result<metadata::FactoryMetadata> {
+        /*
+         * Allow the per-target diagnostic configuration to override the base
+         * diagnostic configuration.
+         */
+        Ok(self.config.diag.apply_overrides(&t.diag)?.build()?)
     }
 }
 

factory/aws/src/main.rs

@@ -5,6 +5,7 @@
 use std::{collections::HashMap, net::Ipv4Addr, path::PathBuf};
 
 use anyhow::{bail, Result};
+use buildomat_types::config::ConfigFileDiag;
 use serde::Deserialize;
 
 use crate::db::types::InstanceId;
@@ -20,6 +21,9 @@ pub(crate) struct ConfigFile {
     pub slots: u32,
     pub software_dir: String,
     pub nodename: String,
+
+    #[serde(default)]
+    pub diag: ConfigFileDiag,
 }
 
 #[derive(Deserialize, Debug, Clone)]
@@ -58,6 +62,9 @@ pub(crate) struct ConfigFileSlotTemplate {
 pub(crate) struct ConfigFileTarget {
     pub image_path: Option<String>,
     pub image_zvol: Option<String>,
+
+    #[serde(default)]
+    pub diag: ConfigFileDiag,
 }
 
 impl ConfigFileTarget {

factory/propolis/src/config.rs

@@ -61,12 +61,31 @@ async fn factory_task_one(log: &Logger, c: &Arc<Central>) -> Result<()> {
                         log,
                         "associating instance {id} with worker {}", w.id,
                     );
+
+                    /*
+                     * We need to be able to load the target configuration in
+                     * order to get any potential per-target diagnostic
+                     * configuration.
+                     */
+                    let Some(targ) = c.config.target.get(&i.target) else {
+                        error!(
+                            log,
+                            "instance {id} for worker {} has target {} which \
+                            no longer exists?",
+                            w.id,
+                            i.target,
+                        );
+                        continue;
+                    };
+                    let md = c.metadata(targ)?;
+
                     c.client
                         .factory_worker_associate()
                         .worker(&w.id)
                         .body_map(|b| {
                             b.private(id.to_string())
                                 .ip(Some(t.ip().to_string()))
+                                .metadata(Some(md))
                         })
                         .send()
                         .await?;
@@ -240,7 +259,7 @@ async fn factory_task_one(log: &Logger, c: &Arc<Central>) -> Result<()> {
     /*
      * Locate target-specific configuration.
      */
-    if !c.config.target.contains_key(&lease.target) {
+    let Some(targ) = c.config.target.get(&lease.target) else {
         error!(log, "server wants target we do not support: {lease:?}");
         return Ok(());
     };
@@ -269,11 +288,14 @@ async fn factory_task_one(log: &Logger, c: &Arc<Central>) -> Result<()> {
     /*
      * Record the instance ID against the worker for which it was created:
      */
+    let md = c.metadata(targ)?;
     c.client
         .factory_worker_associate()
         .worker(&w.id)
         .body_map(|b| {
-            b.private(instance_id.to_string()).ip(Some(t.ip().to_string()))
+            b.private(instance_id.to_string())
+                .ip(Some(t.ip().to_string()))
+                .metadata(Some(md))
         })
         .send()
         .await?;

factory/propolis/src/factory.rs

@@ -6,6 +6,7 @@ use std::{sync::Arc, time::Duration};
 
 use anyhow::{bail, Context, Result};
 use buildomat_common::*;
+use buildomat_types::metadata;
 use getopts::Options;
 use slog::{info, o, Logger};
 
@@ -29,6 +30,19 @@ struct Central {
     serial: serial::Serial,
 }
 
+impl Central {
+    fn metadata(
+        &self,
+        t: &config::ConfigFileTarget,
+    ) -> Result<metadata::FactoryMetadata> {
+        /*
+         * Allow the per-target diagnostic configuration to override the base
+         * diagnostic configuration.
+         */
+        Ok(self.config.diag.apply_overrides(&t.diag)?.build()?)
+    }
+}
+
 /*
  * This factory runs on large AMD machines (e.g., 100+ SMT threads) and thus
  * ends up with far too many worker threads by default.

factory/propolis/src/main.rs

@@ -8,6 +8,7 @@ license = "MPL-2.0"
 doctest = false
 
 [dependencies]
+anyhow = { workspace = true }
 schemars = { workspace = true }
 serde = { workspace = true }
 serde_json = { workspace = true }