XXX hubris factory wip

Author: jclulow

Cargo.lock

@@ -21,7 +21,6 @@ schemars = { workspace = true }
 sea-query = { workspace = true }
 serde = { workspace = true }
 serde_json = { workspace = true }
-serde_yaml = { workspace = true }
 slog = { workspace = true }
 slog-term = { workspace = true }
 smf = { workspace = true }

factory/hubris/Cargo.toml

@@ -0,0 +1,33 @@
+<?xml version='1.0'?>
+<!DOCTYPE service_bundle SYSTEM '/usr/share/lib/xml/dtd/service_bundle.dtd.1'>
+<service_bundle type='manifest' name='export'>
+    <service name='site/buildomat/hubris-factory' type='service' version='0'>
+	    <create_default_instance enabled='false' />
+        <dependency name='multi-user' grouping='require_all'
+          restart_on='none' type='service'>
+            <service_fmri value='svc:/milestone/multi-user'/>
+        </dependency>
+
+        <exec_method name='start' type='method' timeout_seconds='60'
+		exec='/opt/buildomat/hubris/lib/buildomat-factory-hubris -f /opt/buildomat/hubris/etc/config.toml -d /opt/buildomat/hubris/var/db.sqlite3' />
+
+        <exec_method name='stop' type='method' timeout_seconds='60'
+          exec=':kill'/>
+
+        <property_group name='startd' type='framework'>
+            <propval name='duration' type='astring' value='child'/>
+        </property_group>
+
+        <stability value='Unstable'/>
+
+        <template>
+            <common_name>
+                <loctext xml:lang='C'>buildomat hubris factory</loctext>
+            </common_name>
+            <description>
+                <loctext xml:lang='C'>buildomat hubris factory</loctext>
+            </description>
+        </template>
+
+    </service>
+</service_bundle>

factory/hubris/smf/hubris-factory.xml

@@ -0,0 +1,35 @@
+<?xml version='1.0'?>
+<!DOCTYPE service_bundle SYSTEM '/usr/share/lib/xml/dtd/service_bundle.dtd.1'>
+<service_bundle type='manifest' name='export'>
+    <service name='site/buildomat/hubris-agent' type='service' version='0'>
+        <dependency name='multi-user' grouping='require_all'
+          restart_on='none' type='service'>
+            <service_fmri value='svc:/milestone/multi-user'/>
+        </dependency>
+
+        <exec_method name='start' type='method' timeout_seconds='60'
+          exec='/opt/buildomat/hubris/lib/start.sh'>
+          <method_context>
+            <method_credential user='build' group='build' />
+          </method_context>
+        </exec_method>
+        <exec_method name='stop' type='method' timeout_seconds='60'
+          exec=':kill'/>
+
+        <!--<property_group name='startd' type='framework'>
+            <propval name='duration' type='astring' value='child'/>
+        </property_group>-->
+
+        <stability value='Unstable'/>
+
+        <template>
+            <common_name>
+                <loctext xml:lang='C'>buildomat agent</loctext>
+            </common_name>
+            <description>
+                <loctext xml:lang='C'>buildomat agent</loctext>
+            </description>
+        </template>
+
+    </service>
+</service_bundle>

factory/hubris/smf/hubris.xml

@@ -17,6 +17,7 @@ pub(crate) struct ConfigFile {
     //pub template: ConfigFileSlotTemplate,
     //#[serde(default)]
     pub target: HashMap<String, ConfigFileTarget>,
+    pub slot: HashMap<String, ConfigFileSlot>,
     //pub slots: u32,
     //pub software_dir: String,
     pub nodename: String,
@@ -37,21 +38,29 @@ pub(crate) struct ConfigFileFactory {
 #[derive(Deserialize, Debug, Clone)]
 #[serde(deny_unknown_fields)]
 pub(crate) struct ConfigFileTarget {
-    pub enable: bool,
+    pub slot: String,
+}
+
+#[derive(Deserialize, Debug, Clone)]
+#[serde(deny_unknown_fields)]
+pub(crate) struct ConfigFileSlot {
+    pub vid: u16,
+    pub pid: u16,
+    pub serial: String,
+    pub kind: Kind,
+}
+
+#[derive(Deserialize, Debug, Clone)]
+#[serde(rename = "snake_case")]
+pub(crate) enum Kind {
+    Basic,
+    Rot,
 }
 
 impl ConfigFile {
     pub fn for_instance(&self, id: &InstanceId) -> Result<InstanceInSlot> {
         Ok(InstanceInSlot { config: self, id: id.clone() })
     }
-
-    pub fn targets(&self) -> Vec<String> {
-        self.target
-            .iter()
-            .filter(|(_, t)| t.enable)
-            .map(|(id, _)| id.to_string())
-            .collect()
-    }
 }
 
 pub(crate) struct InstanceInSlot<'a> {

factory/hubris/src/config.rs

@@ -216,6 +216,10 @@ async fn factory_task_one(log: &Logger, c: &Arc<Central>) -> Result<()> {
     let res = c
         .client
         .factory_lease()
+        /*
+         * XXX The supported target list should come from the slot manager, not
+         * just the configuration.
+         */
         .body_map(|b| b.supported_targets(c.config.targets()))
         .send()
         .await?

factory/hubris/src/factory.rs

@@ -0,0 +1,35 @@
+/*
+ * Copyright 2024 Oxide Computer Company
+ */
+
+use std::collections::BTreeMap;
+
+use serde::Serialize;
+
+/*
+ * Humility accepts a JSON file describing a list of probes, so that they can be
+ * selected by name using the -t (target) option.  We write out such a file and
+ * include the path in the the job environment (HUMILITY_ENVIRONMENT) so that it
+ * can be used by humility commands invoked in the job.
+ */
+type EnvironmentFile = BTreeMap<String, Environment>;
+
+#[derive(Serialize)]
+pub struct Environment {
+    pub description: String,
+    pub probe: String,
+    pub archive: String,
+}
+
+#[derive(Serialize)]
+#[serde(untagged)]
+pub enum Archive {
+    Single(String),
+    Slots(TwoSlots),
+}
+
+#[derive(Serialize)]
+pub struct TwoSlots {
+    imagea: String,
+    imageb: String,
+}

factory/hubris/src/humility.rs

@@ -15,6 +15,7 @@ mod factory;
 mod unix;
 mod usb;
 mod workload;
+mod humility;
 //mod net;
 //mod nocloud;
 //mod propolis;

factory/hubris/src/main.rs

@@ -18,7 +18,6 @@ use anyhow::{anyhow, bail, Result};
 use buildomat_common::OutputExt;
 use slog::{debug, error, info, o, warn, Logger};
 
-const FMRI: &str = "svc:/site/buildomat/hubris-agent:default";
 const SERVICE: &str = "site/buildomat/hubris-agent";
 
 pub(crate) async fn workload_worker(c: Arc<Central>) -> Result<()> {
@@ -56,7 +55,7 @@ async fn workload_worker_one(
      * service bundle.
      */
     let svc = loc
-        .get_service("site/buildomat/hubris-agent")?
+        .get_service(SERVICE)?
         .ok_or_else(|| anyhow!("could not locate service {SERVICE:?}"))?;
 
     /*
@@ -338,7 +337,7 @@ async fn instance_worker_one(
     let mut scf = smf::Scf::new()?;
     let loc = scf.scope_local()?;
     let svc = loc
-        .get_service("site/buildomat/hubris-agent")?
+        .get_service(SERVICE)?
         .ok_or_else(|| anyhow!("could not locate service {SERVICE:?}"))?;
 
     match i.state {
@@ -352,7 +351,30 @@ async fn instance_worker_one(
             scrub_tmp(log, build_user)?;
 
             /*
-             * XXX Make sure we have an MCU-Link available.
+             * XXX We need to confirm that power control is available for the
+             * things that we intend to control here, and that the devices under
+             * test begin in the OFF state.
+             */
+
+            /*
+             * XXX Make sure we have hardware devices available.
+             *
+             * We need configuration here; looms, like in manufacturing,
+             * to be able to specify a "gimlet", "sidecar", and "psc" target.
+             *
+             * We should select a loom here (which would include a list of
+             * vid/pid/serial pairs for dongles) and commit to it for the
+             * duration of this "instance".
+             *
+             * We should power up each device in sequence and ensure they are
+             * flashed with stock images, confirm on some level that they are
+             * working, and then power them back off.
+             *
+             * We need to be able to generate a humility environment file here
+             * as part of the hand-off to the job.
+             *
+             * XXX Really this should actually happen before we've even promised
+             * to start the job -- this should be part of the "slot" definition.
              */
             let mculink =
                 c.usb.mculink().ok_or_else(|| anyhow!("no MCU-Link found?"))?;
@@ -418,10 +440,9 @@ async fn instance_worker_one(
         InstanceState::Configured => {
             let mut scf = smf::Scf::new()?;
             let loc = scf.scope_local()?;
-            let svc =
-                loc.get_service("site/buildomat/hubris-agent")?.ok_or_else(
-                    || anyhow!("could not locate service {SERVICE:?}"),
-                )?;
+            let svc = loc.get_service(SERVICE)?.ok_or_else(|| {
+                anyhow!("could not locate service {SERVICE:?}")
+            })?;
 
             let smfi = svc.get_instance(&svcname)?.ok_or_else(|| {
                 anyhow!("could not locate SMF instance {svcname:?}")
@@ -475,12 +496,14 @@ async fn instance_worker_one(
                          * This is a terminal state and we can delete the
                          * service instance.
                          */
+                        info!(log, "deleting SMF service");
+                        smfi.delete()?;
                     }
                     other => {
                         /*
                          * Try to disable the service.
                          */
-                        warn!(log, "SMF state: {other:?} --> disable!");
+                        info!(log, "disabling SMF service");
                         smfi.disable(true)?;
                         return Ok(DoNext::Sleep);
                     }
@@ -490,13 +513,24 @@ async fn instance_worker_one(
             };
 
             /*
-             * XXX we need to ensure all processes owned by the builder
-             * uid are killed...  see: sigsend(2) / sigsendset(2)
+             * While we have disabled the service, it is possible that some
+             * processes could have been created outside the contract.  Ensure
+             * all processes for the build user are terminated and then remove
+             * any files left behind.
              */
             kill_all(log, build_user)?;
             dataset_destroy(log, build_user)?;
             scrub_tmp(log, build_user)?;
 
+            /*
+             * XXX We need to confirm that we are able to shut off the power for
+             * the devices under test here.
+             */
+
+            /*
+             * XXX Change ownership of any USB devices back to root
+             */
+
             c.db.instance_new_state(&id, InstanceState::Destroyed)?;
             Ok(DoNext::Immediate)
         }