swarm-builder: an agent option whose eval failed is left out of the reply, not read as its default

docs/agent-lifecycle/persistence.md

@@ -476,9 +476,11 @@ first start afterwards has hive-priv, as root, chown the agent's `state`,
 builder reads `config.services.hyperhive.agent.user.privateUsers` from the
 config it evaluates in its sandbox and puts it in the agent's closure record
 (`private_users`). hive-c0re reads the record when it writes the container's
-flags and sends the value to hive-priv with each start. A record or builder
+flags and sends the value to hive-priv with each start. An agent whose module
-from before the field reads false; a record hive-c0re can't read, or none,
+doesn't declare the option reads false. When the builder can't read the
-leaves the agent as it is.
+option (its eval fails or times out), the record carries no flag, and so does a
+record or builder from before the field. A record without a flag, a record
+hive-c0re can't read, or none, leaves the agent as it is.
 
 - **On disk the data belongs to one host uid:gid**,
   `services.hyperhive.deploy.hive-controller.userns.dataOwner`

docs/swarm/builder.md

@@ -282,6 +282,30 @@ The record's `cached` field says whether the builder pushed the closure to
 [the binary cache](#binary-cache). Each hive reads its own agents' keys to
 [report whether it can pull them and download them in a rebuild](#what-hives-do-with-the-cache). The closure stays unrooted in the builder host's store.
 
+The built reply and the record also carry two options of the agent's evaluated
+config, `private_users` (`services.hyperhive.agent.user.privateUsers`) and
+`nix_builds` (`services.hyperhive.agent.nixBuilds`). The builder reads each with
+a `nix eval` in its sandbox, and an eval has three outcomes:
+
+- **A value:** the field holds it.
+- **The module doesn't declare the option** (`nix` says the flake doesn't
+  provide the attribute): the field holds the option's default, off for
+  `private_users` and `host` for `nix_builds`.
+- **Any other failure, or a timeout:** the value is unknown, and the reply and
+  the record leave the field out. hive-c0re then sends nothing for that option,
+  and hive-priv keeps what the container's last write chose, as it does for an
+  agent with no record.
+
+Mixed versions:
+
+| builder | controller and hive-c0re | what happens                                                                                                   |
+| ------- | ------------------------ | -------------------------------------------------------------------------------------------------------------- |
+| new     | new                      | an unread option leaves the field out, and the last write stands                                               |
+| old     | new                      | the old builder always sends a value, which the record carries                                                 |
+| new     | old                      | the old side reads a missing field as the default, so an unread option behaves as it did before this behaviour |
+
+A record or reply from before the fields reads as unknown.
+
 The controller gives the `builder` forge account read on `internal/meta` when it
 updates the flake, and again before each build, since the repo is private. A
 grant the forge refuses fails the build with no verdict.

docs/trust-boundary/security.md

@@ -211,12 +211,15 @@ the option from the evaluated config into the agent's closure record
 flags and sends the mode to hive-priv, and the flags apply at the container's
 next start.
 
-- A record or builder from before the field reads `host`.
+- An agent whose module doesn't declare the option reads `host`. When the builder
+  can't read the option (its eval fails or times out), the record carries no
+  mode, and so does a record or builder from before the field. hive-c0re then
+  sends no mode, and hive-priv keeps what the container's last write chose.
 - A mode this build doesn't know (a record written by a newer swarm) reads as
   `none`, never `host`: an agent that asked for a mode its hive can't give gets
   no builds, not builds outside its limits.
-- A record hive-c0re can't read, or none, sends no mode, and hive-priv keeps
+- A record hive-c0re can't read, or none, sends no mode either, and hive-priv
-  what the container's last write chose.
+  keeps what the container's last write chose.
 - A hive-c0re or hive-priv from before the option ignores it, so the agent keeps
   host builds until you upgrade both.
 

hive-c0re/src/attic/closure.rs

@@ -173,8 +173,8 @@ mod tests {
             system: "x86_64-linux".to_owned(),
             built_unix: 1,
             cached,
-            private_users: false,
+            private_users: Some(false),
-            nix_builds: hive_types::NixBuilds::Host,
+            nix_builds: Some(hive_types::NixBuilds::Host),
         }
     }
 

hive-c0re/src/attic/probe.rs

@@ -245,8 +245,8 @@ mod tests {
             system: "x86_64-linux".to_owned(),
             built_unix: 1,
             cached: true,
-            private_users: false,
+            private_users: Some(false),
-            nix_builds: hive_types::NixBuilds::Host,
+            nix_builds: Some(hive_types::NixBuilds::Host),
         }
     }
 

hive-c0re/src/job_queue/exec/tests.rs

@@ -274,8 +274,8 @@ fn closure(cached: bool) -> AgentClosure {
         system: "x86_64-linux".to_owned(),
         built_unix: 1,
         cached,
-        private_users: false,
+        private_users: Some(false),
-        nix_builds: hive_types::NixBuilds::Host,
+        nix_builds: Some(hive_types::NixBuilds::Host),
     }
 }
 

hive-c0re/src/lifecycle/host_config.rs

@@ -518,13 +518,14 @@ fn socket_bind(agent_name: &str, socket_dir: &Path, private_users: Option<bool>)
 }
 
 /// What hive-priv is told about the agent's user namespace: the record's flag,
-/// or `None` (leave the agent as it is) when there is no record or it cannot be
+/// or `None` (leave the agent as it is) when there is no record, it cannot be
-/// read, so a swarm outage never turns a namespaced agent off or on.
+/// read, or it carries no flag because the builder could not read the option,
+/// so neither a swarm outage nor a failed eval turns a namespaced agent off or on.
 fn private_users_of(
     record: &Result<Option<swarm_wire::agent_closures::AgentClosure>>,
 ) -> Option<bool> {
     match record {
-        Ok(record) => record.as_ref().map(|r| r.private_users),
+        Ok(record) => record.as_ref().and_then(|r| r.private_users),
         Err(e) => {
             tracing::warn!(error = %format!("{e:#}"), "closure record unreadable; the agent's user namespace is left as it is");
             None
@@ -533,13 +534,15 @@ fn private_users_of(
 }
 
 /// What hive-priv is told about the agent's nix builds: the record's mode, or
-/// `None` (leave the agent as the last write set it) when there is no record or
+/// `None` (leave the agent as the last write set it) when there is no record,
-/// it cannot be read, so a swarm outage never turns an agent's host builds on.
+/// it cannot be read, or it carries no mode because the builder could not read
+/// the option, so neither a swarm outage nor a failed eval turns an agent's
+/// host builds on.
 fn nix_builds_of(
     record: &Result<Option<swarm_wire::agent_closures::AgentClosure>>,
 ) -> Option<NixBuilds> {
     match record {
-        Ok(record) => record.as_ref().map(|r| r.nix_builds),
+        Ok(record) => record.as_ref().and_then(|r| r.nix_builds),
         Err(e) => {
             tracing::warn!(error = %format!("{e:#}"), "closure record unreadable; the agent's nix builds are left as they are");
             None
@@ -579,13 +582,13 @@ mod tests {
     use crate::coordinator::HiveEnv;
     use hive_types::NixBuilds;
 
-    fn record_with(private_users: bool) -> swarm_wire::agent_closures::AgentClosure {
+    fn record_with(private_users: Option<bool>) -> swarm_wire::agent_closures::AgentClosure {
-        record_with_builds(private_users, NixBuilds::Host)
+        record_with_builds(private_users, Some(NixBuilds::Host))
     }
 
     fn record_with_builds(
-        private_users: bool,
+        private_users: Option<bool>,
-        nix_builds: NixBuilds,
+        nix_builds: Option<NixBuilds>,
     ) -> swarm_wire::agent_closures::AgentClosure {
         swarm_wire::agent_closures::AgentClosure {
             meta_rev: "a".repeat(40),
@@ -613,48 +616,58 @@ mod tests {
         assert!(!b.read_only);
     }
 
-    /// The flag in the record is what hive-priv is told, both ways. No record
+    /// The flag in the record is what hive-priv is told, both ways. No record,
-    /// and an unreadable one say nothing, so a swarm outage changes no agent's
+    /// an unreadable one, and a record whose builder could not read the option
-    /// namespace; a record from before the field reads false.
+    /// (or that predates the field) say nothing, so neither a swarm outage nor
+    /// a failed eval changes an agent's namespace.
     #[test]
     fn the_records_flag_is_what_the_start_path_passes() {
-        assert_eq!(
+        for flag in [true, false] {
-            super::private_users_of(&Ok(Some(record_with(true)))),
+            assert_eq!(
-            Some(true)
+                super::private_users_of(&Ok(Some(record_with(Some(flag))))),
-        );
+                Some(flag)
-        assert_eq!(
+            );
-            super::private_users_of(&Ok(Some(record_with(false)))),
+        }
-            Some(false)
-        );
         assert_eq!(super::private_users_of(&Ok(None)), None);
         assert_eq!(super::private_users_of(&Err(anyhow::anyhow!("down"))), None);
+        assert_eq!(
+            super::private_users_of(&Ok(Some(record_with(None)))),
+            None,
+            "an unread flag keeps the last write"
+        );
         let old = serde_json::json!({
             "meta_rev": "a", "store_path": "/nix/store/x",
             "system": "s", "built_unix": 1,
         });
         let old = serde_json::from_value(old).expect("decodes");
-        assert_eq!(super::private_users_of(&Ok(Some(old))), Some(false));
+        assert_eq!(super::private_users_of(&Ok(Some(old))), None);
     }
 
-    /// The record's mode is what hive-priv is told, both ways. No record and an
+    /// The record's mode is what hive-priv is told, both ways. No record, an
-    /// unreadable one say nothing, so a swarm outage changes no agent; a record
+    /// unreadable one, and a record whose builder could not read the option (or
-    /// from before the field reads as host builds.
+    /// that predates the field) say nothing, so neither a swarm outage nor a
+    /// failed eval changes an agent.
     #[test]
     fn the_records_nix_builds_is_what_the_start_path_passes() {
         for mode in [NixBuilds::Host, NixBuilds::Disabled] {
             assert_eq!(
-                super::nix_builds_of(&Ok(Some(record_with_builds(false, mode)))),
+                super::nix_builds_of(&Ok(Some(record_with_builds(None, Some(mode))))),
                 Some(mode)
             );
         }
         assert_eq!(super::nix_builds_of(&Ok(None)), None);
         assert_eq!(super::nix_builds_of(&Err(anyhow::anyhow!("down"))), None);
+        assert_eq!(
+            super::nix_builds_of(&Ok(Some(record_with_builds(None, None)))),
+            None,
+            "an unread mode keeps the last write"
+        );
         let old = serde_json::json!({
             "meta_rev": "a", "store_path": "/nix/store/x",
             "system": "s", "built_unix": 1,
         });
         let old = serde_json::from_value(old).expect("decodes");
-        assert_eq!(super::nix_builds_of(&Ok(Some(old))), Some(NixBuilds::Host));
+        assert_eq!(super::nix_builds_of(&Ok(Some(old))), None);
         // A mode a newer swarm wrote, which this build does not know, decodes
         // and builds nothing.
         let newer = serde_json::json!({

swarm-builder/src/main.rs

@@ -862,8 +862,8 @@ mod tests {
                     out_path: "/nix/store/x".to_owned(),
                     system: "s".to_owned(),
                     cached: false,
-                    private_users: false,
+                    private_users: Some(false),
-                    nix_builds: hive_types::NixBuilds::Host,
+                    nix_builds: Some(hive_types::NixBuilds::Host),
                 }
             },
         )
@@ -885,8 +885,8 @@ mod tests {
                     out_path: "/nix/store/x".to_owned(),
                     system: "s".to_owned(),
                     cached: false,
-                    private_users: false,
+                    private_users: Some(false),
-                    nix_builds: hive_types::NixBuilds::Host,
+                    nix_builds: Some(hive_types::NixBuilds::Host),
                 }
             },
         )
@@ -1454,7 +1454,9 @@ mod tests {
 
     #[tokio::test]
     async fn a_meta_build_is_answered_with_the_path_the_system_and_the_private_users_flag() {
-        let fake = meta_nix("echo true");
+        let fake = meta_nix(
+            "for a in \"$@\"; do last=$a; done; case \"$last\" in *nixBuilds) echo '\"host\"';; *) echo true;; esac",
+        );
         let (env, mut sent) = env();
         let request = msg(Some("_INBOX.5"), &meta_json("atlas", REV));
         serve_meta(fake.builder(), env, vec![request]).await;
@@ -1490,12 +1492,17 @@ mod tests {
         );
     }
 
+    /// An option the module does not declare is false; one whose eval failed, or
+    /// printed no bool, is left out of the reply, for the controller to record as
+    /// unknown. `echo false` is the control that the field is read at all.
     #[tokio::test]
-    async fn private_users_is_false_unless_nix_eval_printed_true() {
+    async fn private_users_is_left_out_unless_nix_eval_printed_a_bool_or_the_option_is_missing() {
-        for eval_says in [
+        let missing = "echo \"error: flake 'f' does not provide attribute 'a'\" >&2; exit 1";
-            "echo false",
+        for (eval_says, want) in [
-            "echo null",
+            ("echo false", Some(false)),
-            "echo 'error: no such option' >&2; exit 1",
+            (missing, Some(false)),
+            ("echo null", None),
+            ("echo 'error: no such option' >&2; exit 1", None),
         ] {
             let fake = meta_nix(eval_says);
             let (env, mut sent) = env();
@@ -1503,7 +1510,12 @@ mod tests {
             serve_meta(fake.builder(), env, vec![request]).await;
             let (_, body) = next(&mut sent).await;
             assert_eq!(body["state"], "built", "{eval_says}");
-            assert_eq!(body["private_users"], false, "{eval_says}");
+            assert_eq!(
+                body.get("private_users")
+                    .and_then(serde_json::Value::as_bool),
+                want,
+                "{eval_says}: {body}"
+            );
         }
     }
 

swarm-builder/src/nix_build.rs

@@ -186,58 +186,50 @@ impl Builder {
     }
 
     /// `nix eval --json` of `option` in the agent's evaluated config, run in this
-    /// sandbox against the flake the build just evaluated. `None` when the eval
+    /// sandbox against the flake the build just evaluated. Logs why when the
-    /// fails or times out, and logs why along with what the caller falls back to.
+    /// answer is not a value.
-    async fn eval_option(
+    async fn eval_option(&self, job: &BuildFromMeta, token: &str, option: &str) -> Eval {
-        &self,
-        job: &BuildFromMeta,
-        token: &str,
-        option: &str,
-        fallback: &str,
-    ) -> Option<Vec<u8>> {
         let mut cmd = self.nix_command(token);
         cmd.args(EVAL_FLAGS)
             .arg(format!("{}.{option}", self.meta_config(job)));
         match run_raw(cmd, self.deadline).await {
-            Ok((status, out, _)) if status.success() => Some(out),
+            Ok((status, out, _)) if status.success() => Eval::Value(out),
+            Ok((_, _, err)) if attribute_missing(&String::from_utf8_lossy(&err)) => {
+                tracing::info!(agent = %job.agent, "{option} is not declared by the agent's module; its default applies");
+                Eval::Missing
+            }
             Ok((status, _, err)) => {
-                tracing::info!(agent = %job.agent, %status, error = %tail(&String::from_utf8_lossy(&err), status), "{option} not read; {fallback}");
+                tracing::info!(agent = %job.agent, %status, error = %tail(&String::from_utf8_lossy(&err), status), "{option} not read; the reply carries no value for it");
-                None
+                Eval::Unknown
             }
             Err(why) => {
-                tracing::warn!(agent = %job.agent, %why, "{option} not read; {fallback}");
+                tracing::warn!(agent = %job.agent, %why, "{option} not read; the reply carries no value for it");
-                None
+                Eval::Unknown
             }
         }
     }
 
     /// Whether the agent's config sets `services.hyperhive.agent.user.privateUsers`.
-    /// Anything but a clean `true` is false: an agent whose module predates the
+    /// An agent whose module predates the option is out of a user namespace;
-    /// option, a failed eval and a timeout all keep the agent out of a user
+    /// a failed eval, a timeout and an output that is no bool are `None`.
-    /// namespace.
+    async fn private_users(&self, job: &BuildFromMeta, token: &str) -> Option<bool> {
-    async fn private_users(&self, job: &BuildFromMeta, token: &str) -> bool {
+        match self.eval_option(job, token, PRIVATE_USERS_OPTION).await {
-        self.eval_option(
+            Eval::Value(out) => parse_flag(&out),
-            job,
+            Eval::Missing => Some(false),
-            token,
+            Eval::Unknown => None,
-            PRIVATE_USERS_OPTION,
+        }
-            "the agent stays out of a user namespace",
-        )
-        .await
-        .is_some_and(|out| parse_flag(&out))
     }
 
     /// The agent's `services.hyperhive.agent.nixBuilds`. An agent whose module
-    /// predates the option, a failed eval and a timeout all read as
+    /// predates the option reads as [`NixBuilds::Host`], which is what every
-    /// [`NixBuilds::Host`], which is what every agent did before the option.
+    /// agent did before the option; a failed eval, a timeout and an output that
-    async fn nix_builds(&self, job: &BuildFromMeta, token: &str) -> NixBuilds {
+    /// is no string are `None`.
-        self.eval_option(
+    async fn nix_builds(&self, job: &BuildFromMeta, token: &str) -> Option<NixBuilds> {
-            job,
+        match self.eval_option(job, token, NIX_BUILDS_OPTION).await {
-            token,
+            Eval::Value(out) => parse_nix_builds(&out),
-            NIX_BUILDS_OPTION,
+            Eval::Missing => Some(NixBuilds::default()),
-            "the agent builds through the host daemon",
+            Eval::Unknown => None,
-        )
+        }
-        .await
-        .map_or_else(NixBuilds::default, |out| parse_nix_builds(&out))
     }
 
     /// The config commit as a flake reference, fetched through the PR's head
@@ -281,15 +273,35 @@ const PRIVATE_USERS_OPTION: &str = "services.hyperhive.agent.user.privateUsers";
 /// The agent option holding where its nix builds run.
 const NIX_BUILDS_OPTION: &str = "services.hyperhive.agent.nixBuilds";
 
-/// `nix eval --json` of a bool: true only for exactly `true`.
+/// What `nix eval --json` of an agent option came to.
-fn parse_flag(out: &[u8]) -> bool {
+enum Eval {
-    out.trim_ascii() == b"true"
+    /// It printed a value.
+    Value(Vec<u8>),
+    /// The config has no such attribute: the agent's module predates the option.
+    Missing,
+    /// It failed another way or timed out, so the value is not known.
+    Unknown,
 }
 
-/// `nix eval --json` of a [`NixBuilds`]: the spelling, or the default when the
+/// Whether `stderr` is `nix` saying the attribute path is not in the flake's
+/// outputs, which for an option path means the module does not declare it.
+fn attribute_missing(stderr: &str) -> bool {
+    stderr.contains("does not provide attribute")
+}
+
+/// `nix eval --json` of a bool: the bool, or `None` for anything else.
+fn parse_flag(out: &[u8]) -> Option<bool> {
+    match out.trim_ascii() {
+        b"true" => Some(true),
+        b"false" => Some(false),
+        _ => None,
+    }
+}
+
+/// `nix eval --json` of a [`NixBuilds`]: the spelling, or `None` when the
 /// output is not a string.
-fn parse_nix_builds(out: &[u8]) -> NixBuilds {
+fn parse_nix_builds(out: &[u8]) -> Option<NixBuilds> {
-    serde_json::from_slice(out).unwrap_or_default()
+    serde_json::from_slice(out).ok()
 }
 
 /// `nix flake lock` of the flake in the working directory. The flake's config
@@ -577,17 +589,27 @@ mod tests {
         );
     }
 
-    /// The reply carries what the evaluated config says about the option, and
+    /// The shell for a `nix eval` of an option the module does not declare, as
-    /// anything but a clean `true` reads as false. The `true` arm is the
+    /// `nix` 2.34 words it.
-    /// control for the others.
+    const ATTRIBUTE_MISSING: &str = "echo \"error: flake 'git+https://forge.example/internal/meta.git' does not provide attribute 'packages.x86_64-linux.nixosConfigurations.atlas.config.services.hyperhive.agent.user.privateUsers'\" >&2; exit 1";
+
+    /// The reply carries what the evaluated config says about the option. An
+    /// agent whose module predates the option is false; an eval that fails
+    /// another way, or prints no bool, leaves the field out. The `true` arm is
+    /// the control for the others.
     #[tokio::test]
     async fn a_meta_build_reports_the_agents_private_users_option() {
         for (i, (answer, want)) in [
-            ("echo true", true),
+            ("echo true", Some(true)),
-            ("echo false", false),
+            ("echo false", Some(false)),
-            ("echo null", false),
+            (ATTRIBUTE_MISSING, Some(false)),
-            ("echo true; exit 1", false),
+            ("echo null", None),
-            ("echo error: attribute missing >&2; exit 1", false),
+            ("echo true; exit 1", None),
+            ("echo error: attribute missing >&2; exit 1", None),
+            (
+                "echo 'error: infinite recursion encountered' >&2; exit 1",
+                None,
+            ),
         ]
         .into_iter()
         .enumerate()
@@ -606,17 +628,19 @@ mod tests {
     }
 
     /// The reply carries the option's spelling. `host` is the control for
-    /// `none`; a mode this build does not know reads as `none`, and anything
+    /// `none`; a mode this build does not know reads as `none`, an agent whose
-    /// that is no spelling at all (a failed eval, an older module) as `host`.
+    /// module predates the option as `host`, and a failed eval, or output that
+    /// is no spelling at all, leaves the field out.
     #[tokio::test]
     async fn a_meta_build_reports_the_agents_nix_builds_option() {
         for (i, (answer, want)) in [
-            ("echo '\"host\"'", NixBuilds::Host),
+            ("echo '\"host\"'", Some(NixBuilds::Host)),
-            ("echo '\"none\"'", NixBuilds::Disabled),
+            ("echo '\"none\"'", Some(NixBuilds::Disabled)),
-            ("echo '\"container\"'", NixBuilds::Disabled),
+            ("echo '\"container\"'", Some(NixBuilds::Disabled)),
-            ("echo null", NixBuilds::Host),
+            (ATTRIBUTE_MISSING, Some(NixBuilds::Host)),
-            ("echo '\"none\"'; exit 1", NixBuilds::Host),
+            ("echo null", None),
-            ("echo error: attribute missing >&2; exit 1", NixBuilds::Host),
+            ("echo '\"none\"'; exit 1", None),
+            ("echo error: attribute missing >&2; exit 1", None),
         ]
         .into_iter()
         .enumerate()
@@ -636,6 +660,33 @@ mod tests {
         }
     }
 
+    /// An eval that does not finish in time is unknown for both options, not
+    /// their defaults. The build itself finishes, which is the control.
+    #[tokio::test]
+    async fn an_option_eval_that_times_out_leaves_both_options_out_of_the_reply() {
+        let nix = fake_nix(
+            "eval-timeout",
+            "if [ \"$1\" = eval ]; then exec sleep 30; else echo /nix/store/x; fi\n",
+        );
+        let b = Builder {
+            nix,
+            deadline: Duration::from_secs(1),
+            ..builder()
+        };
+        let reply = b.build_from_meta(&meta_job(), "t").await;
+        assert!(
+            matches!(
+                &reply,
+                FromMetaReply::Built {
+                    private_users: None,
+                    nix_builds: None,
+                    ..
+                }
+            ),
+            "{reply:?}"
+        );
+    }
+
     #[test]
     fn the_option_is_read_from_the_evaluated_config_not_the_toplevel() {
         assert_eq!(
@@ -647,8 +698,11 @@ mod tests {
              ?rev=0123456789abcdef0123456789abcdef01234567\
              #nixosConfigurations.atlas.config.services.hyperhive.agent.user.privateUsers"
         );
-        assert!(parse_flag(b"true\n") && !parse_flag(b"false\n") && !parse_flag(b""));
+        assert_eq!(parse_flag(b"true\n"), Some(true));
+        assert_eq!(parse_flag(b"false\n"), Some(false));
+        assert_eq!(parse_flag(b""), None);
     }
+
     #[test]
     fn a_meta_job_that_would_bend_the_reference_is_refused() {
         for bad in [
@@ -685,8 +739,8 @@ mod tests {
                 out_path: format!("{} {}", BUILD_FLAGS.join(" "), b.meta_attr(&meta_job())),
                 system: "x86_64-linux".to_owned(),
                 cached: false,
-                private_users: false,
+                private_users: None,
-                nix_builds: NixBuilds::Host,
+                nix_builds: None,
             }
         );
     }

swarm-builder/src/push.rs

@@ -266,8 +266,8 @@ mod tests {
             out_path: OUT.to_owned(),
             system: "x86_64-linux".to_owned(),
             cached: false,
-            private_users: false,
+            private_users: None,
-            nix_builds: NixBuilds::Disabled,
+            nix_builds: Some(NixBuilds::Disabled),
         }
     }
 
@@ -304,7 +304,8 @@ mod tests {
             matches!(
                 reply,
                 FromMetaReply::Built {
-                    nix_builds: NixBuilds::Disabled,
+                    private_users: None,
+                    nix_builds: Some(NixBuilds::Disabled),
                     ..
                 }
             ),

swarm-controller/src/cache_miss.rs

@@ -564,8 +564,8 @@ mod tests {
             system: "x86_64-linux".to_owned(),
             built_unix: 7,
             cached,
-            private_users: false,
+            private_users: Some(false),
-            nix_builds: NixBuilds::Host,
+            nix_builds: Some(NixBuilds::Host),
         }
     }
 
@@ -652,8 +652,8 @@ mod tests {
             out_path: out_path.to_owned(),
             system: "x86_64-linux".to_owned(),
             cached,
-            private_users: false,
+            private_users: Some(false),
-            nix_builds: NixBuilds::Host,
+            nix_builds: Some(NixBuilds::Host),
         }
     }
 
@@ -916,15 +916,15 @@ mod tests {
             out_path,
             system,
             cached,
-            private_users: true,
+            private_users: Some(true),
-            nix_builds: NixBuilds::Disabled,
+            nix_builds: Some(NixBuilds::Disabled),
         };
         let world = Fake::missing(&["iris"]).answering("iris", answer);
         pass_at(&world, &mut Backoff::default(), Instant::now()).await;
         let writes = world.writes();
         assert_eq!(writes.len(), 1);
-        assert!(writes[0].1.private_users);
+        assert_eq!(writes[0].1.private_users, Some(true));
-        assert_eq!(writes[0].1.nix_builds, NixBuilds::Disabled);
+        assert_eq!(writes[0].1.nix_builds, Some(NixBuilds::Disabled));
     }
     #[tokio::test]
     async fn another_path_leaves_a_record_that_moved_on() {

swarm-controller/src/meta_build.rs

@@ -492,10 +492,14 @@ mod tests {
         }
 
         fn built_cached(cached: bool) -> Self {
-            Self::built_with(cached, false, NixBuilds::Host)
+            Self::built_with(cached, Some(false), Some(NixBuilds::Host))
         }
 
-        fn built_with(cached: bool, private_users: bool, nix_builds: NixBuilds) -> Self {
+        fn built_with(
+            cached: bool,
+            private_users: Option<bool>,
+            nix_builds: Option<NixBuilds>,
+        ) -> Self {
             Self::answering(Ok(FromMetaReply::Built {
                 out_path: "/nix/store/x-toplevel".to_owned(),
                 system: "x86_64-linux".to_owned(),
@@ -629,11 +633,15 @@ mod tests {
         }
     }
 
-    /// The record says what the builder said about the agent's user namespace,
+    /// The record says what the builder said about the agent's user namespace
-    /// both ways; a builder that said nothing leaves it off.
+    /// and nix builds, both ways; an option the builder left unread stays unread.
     #[tokio::test]
     async fn the_record_carries_the_builders_user_namespace_and_nix_builds_answers() {
-        for (private_users, nix_builds) in [(true, NixBuilds::Disabled), (false, NixBuilds::Host)] {
+        for (private_users, nix_builds) in [
+            (Some(true), Some(NixBuilds::Disabled)),
+            (Some(false), Some(NixBuilds::Host)),
+            (None, None),
+        ] {
             let queue = FakeQueue::built_with(true, private_users, nix_builds);
             let outcome = run_with(&FakeHead::reading(Ok(Some(REV))), &queue).await;
             assert_eq!(outcome, Outcome::Done);
@@ -927,8 +935,8 @@ mod tests {
             system: "x86_64-linux".to_owned(),
             built_unix: 1,
             cached: false,
-            private_users: true,
+            private_users: Some(true),
-            nix_builds: NixBuilds::Disabled,
+            nix_builds: Some(NixBuilds::Disabled),
         };
         client.record("atlas", &closure).await.expect("written");
         closure.store_path = "/nix/store/b".to_owned();

swarm-wire/src/agent_closures.rs

@@ -40,15 +40,15 @@ pub struct AgentClosure {
     pub cached: bool,
     /// Whether the agent's evaluated config sets
     /// `services.hyperhive.agent.user.privateUsers`: the agent runs in its own
-    /// user namespace. A record written before the field existed reads as
+    /// user namespace. `None` when the builder could not read it, and for a
-    /// false, which keeps the agent out of one.
+    /// record written before the field existed: the record says nothing, and
-    #[serde(default)]
+    /// the reader leaves the agent as its last start set it.
-    pub private_users: bool,
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub private_users: Option<bool>,
     /// The agent's `services.hyperhive.agent.nixBuilds`: where its nix builds
-    /// run. A record written before the field existed reads as
+    /// run. `None` as for `private_users`.
-    /// [`NixBuilds::Host`], which is what every agent did before it.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
-    #[serde(default)]
+    pub nix_builds: Option<NixBuilds>,
-    pub nix_builds: NixBuilds,
 }
 
 #[cfg(test)]
@@ -69,8 +69,8 @@ mod tests {
             system: "x86_64-linux".to_owned(),
             built_unix: 7,
             cached: true,
-            private_users: true,
+            private_users: Some(true),
-            nix_builds: NixBuilds::Disabled,
+            nix_builds: Some(NixBuilds::Disabled),
         };
         let wire = serde_json::json!({
             "meta_rev": "abc",
@@ -89,7 +89,7 @@ mod tests {
     }
 
     #[test]
-    fn a_record_without_the_fields_reads_as_not_cached_not_namespaced_and_host_builds() {
+    fn a_record_without_the_fields_reads_as_not_cached_and_says_nothing_of_the_options() {
         let old = serde_json::json!({
             "meta_rev": "abc",
             "store_path": "/nix/store/x",
@@ -98,7 +98,41 @@ mod tests {
         });
         let record = serde_json::from_value::<AgentClosure>(old).expect("decodes");
         assert!(!record.cached);
-        assert!(!record.private_users);
+        assert_eq!(record.private_users, None);
-        assert_eq!(record.nix_builds, NixBuilds::Host);
+        assert_eq!(record.nix_builds, None);
+    }
+
+    /// An option the builder could not read is left out of the record, not
+    /// written as `null`: a reader that predates the option still decodes it.
+    #[test]
+    fn an_unread_option_is_left_out_of_the_record() {
+        // The shape a reader before the option had: a bool that defaults to false.
+        #[derive(serde::Deserialize)]
+        struct Before {
+            #[serde(default)]
+            private_users: bool,
+        }
+        let record = AgentClosure {
+            meta_rev: "abc".to_owned(),
+            store_path: "/nix/store/x".to_owned(),
+            system: "x86_64-linux".to_owned(),
+            built_unix: 7,
+            cached: true,
+            private_users: None,
+            nix_builds: Some(NixBuilds::Host),
+        };
+        let wire = serde_json::to_value(&record).expect("encodes");
+        let fields = wire.as_object().expect("an object");
+        assert!(!fields.contains_key("private_users"), "{wire}");
+        assert_eq!(fields["nix_builds"], "host");
+        assert!(
+            !serde_json::from_value::<Before>(wire.clone())
+                .expect("decodes")
+                .private_users
+        );
+        assert_eq!(
+            serde_json::from_value::<AgentClosure>(wire).expect("decodes"),
+            record
+        );
     }
 }

swarm-wire/src/builder.rs

@@ -151,15 +151,15 @@ pub enum FromMetaReply {
         #[serde(default)]
         cached: bool,
         /// Whether the agent's evaluated config sets
-        /// `services.hyperhive.agent.user.privateUsers`. A builder that
+        /// `services.hyperhive.agent.user.privateUsers`. Absent (`None`) when
-        /// predates the field answers without it, which reads as false.
+        /// the builder could not read it: its eval failed or timed out. A
-        #[serde(default)]
+        /// builder that predates the field answers without it too.
-        private_users: bool,
+        #[serde(default, skip_serializing_if = "Option::is_none")]
-        /// The agent's `services.hyperhive.agent.nixBuilds`. A builder that
+        private_users: Option<bool>,
-        /// predates the field answers without it, which reads as
+        /// The agent's `services.hyperhive.agent.nixBuilds`. Absent (`None`)
-        /// [`NixBuilds::Host`].
+        /// when the builder could not read it, as for `private_users`.
-        #[serde(default)]
+        #[serde(default, skip_serializing_if = "Option::is_none")]
-        nix_builds: NixBuilds,
+        nix_builds: Option<NixBuilds>,
     },
     /// The build did not succeed: the last lines `nix` printed, or why it
     /// stopped (it timed out, or could not start). The controller holds the
@@ -273,8 +273,8 @@ mod tests {
                     out_path: "/nix/store/x".to_owned(),
                     system: "x86_64-linux".to_owned(),
                     cached: true,
-                    private_users: true,
+                    private_users: Some(true),
-                    nix_builds: NixBuilds::Disabled,
+                    nix_builds: Some(NixBuilds::Disabled),
                 },
                 json!({
                     "state": "built",
@@ -307,8 +307,7 @@ mod tests {
     }
 
     #[test]
-    fn a_built_reply_from_a_builder_without_the_fields_reads_as_not_cached_not_namespaced_and_host_builds()
+    fn a_built_reply_from_a_builder_without_the_fields_reads_as_not_cached_and_unknown() {
-     {
         let old = json!({"state": "built", "out_path": "/nix/store/x", "system": "s"});
         assert_eq!(
             serde_json::from_value::<FromMetaReply>(old).expect("decodes"),
@@ -316,12 +315,44 @@ mod tests {
                 out_path: "/nix/store/x".to_owned(),
                 system: "s".to_owned(),
                 cached: false,
-                private_users: false,
+                private_users: None,
-                nix_builds: NixBuilds::Host,
+                nix_builds: None,
             }
         );
     }
 
+    /// A value the builder could not read is left out of the reply, not sent as
+    /// `null`: a controller that predates the option still decodes it.
+    #[test]
+    fn an_unread_option_is_left_out_of_the_built_reply() {
+        // The shape a reader before the option had: a bool that defaults to false.
+        #[derive(serde::Deserialize)]
+        struct Before {
+            #[serde(default)]
+            private_users: bool,
+        }
+        let reply = FromMetaReply::Built {
+            out_path: "/nix/store/x".to_owned(),
+            system: "s".to_owned(),
+            cached: false,
+            private_users: None,
+            nix_builds: Some(NixBuilds::Host),
+        };
+        let wire = serde_json::to_value(&reply).expect("encodes");
+        let fields = wire.as_object().expect("an object");
+        assert!(!fields.contains_key("private_users"), "{wire}");
+        assert_eq!(fields["nix_builds"], "host");
+        assert!(
+            !serde_json::from_value::<Before>(wire.clone())
+                .expect("decodes")
+                .private_users
+        );
+        assert_eq!(
+            serde_json::from_value::<FromMetaReply>(wire).expect("decodes"),
+            reply
+        );
+    }
+
     #[test]
     fn the_subjects_are_pinned() {
         assert_eq!(SUBJECT, "$SWARM.builder.build");