From 305b758040879a4b9935d1d99046b1c1ac16ec9d Mon Sep 17 00:00:00 2001
From: Sean Klein <sean@oxide.computer>
Date: Sun, 30 Apr 2023 00:21:40 -0400
Subject: [PATCH] Read the CRDB address from the running zone

---
 illumos-utils/src/running_zone.rs |  4 ++++
 sled-agent/src/params.rs          |  2 ++
 sled-agent/src/services.rs        | 21 +++++++++++++++++----
 3 files changed, 23 insertions(+), 4 deletions(-)

diff --git a/illumos-utils/src/running_zone.rs b/illumos-utils/src/running_zone.rs
index f26aa82721b..eda6b00ef92 100644
--- a/illumos-utils/src/running_zone.rs
+++ b/illumos-utils/src/running_zone.rs
@@ -147,6 +147,10 @@ impl RunningZone {
         format!("{}/root", self.inner.zonepath.display())
     }
 
+    pub fn control_interface(&self) -> AddrObject {
+        AddrObject::new(self.inner.get_control_vnic_name(), "omicron6").unwrap()
+    }
+
     /// Runs a command within the Zone, return the output.
     pub fn run_cmd<I, S>(&self, args: I) -> Result<String, RunCommandError>
     where
diff --git a/sled-agent/src/params.rs b/sled-agent/src/params.rs
index 0b80c434b4d..84ba8a083a8 100644
--- a/sled-agent/src/params.rs
+++ b/sled-agent/src/params.rs
@@ -521,6 +521,8 @@ impl std::fmt::Display for ZoneType {
 )]
 pub struct ServiceZoneRequest {
     // The UUID of the zone to be initialized.
+    // TODO: Should this be removed? If we have UUIDs on the services, what's
+    // the point of this?
     pub id: Uuid,
     // The type of the zone to be created.
     pub zone_type: ZoneType,
diff --git a/sled-agent/src/services.rs b/sled-agent/src/services.rs
index da20294c4d2..c7d9c59a921 100644
--- a/sled-agent/src/services.rs
+++ b/sled-agent/src/services.rs
@@ -169,6 +169,9 @@ pub enum Error {
     #[error("Services already configured for this Sled Agent")]
     ServicesAlreadyConfigured,
 
+    #[error("Failed to get address: {0}")]
+    GetAddressFailure(#[from] illumos_utils::zone::GetAddressError),
+
     #[error("NTP zone not ready")]
     NtpZoneNotReady,
 
@@ -1871,13 +1874,18 @@ impl ServiceManager {
 
     pub async fn cockroachdb_initialize(&self) -> Result<(), Error> {
         let log = &self.inner.log;
-        // TODO: Can we confirm this address is okay?
-        let host = &format!("[::1]:{COCKROACH_PORT}");
         let dataset_zones = self.inner.dataset_zones.lock().await;
         for zone in dataset_zones.iter() {
             // TODO: We could probably store the ZoneKind in the running zone to
-            // make this a bit safer.
-            if zone.name().contains("cockroach") {
+            // make this "comparison to existing zones by name" mechanism a bit
+            // safer.
+            if zone.name().contains(&ZoneType::CockroachDb.to_string()) {
+                let address = Zones::get_address(
+                    Some(zone.name()),
+                    &zone.control_interface(),
+                )?
+                .network();
+                let host = &format!("[{address}]:{COCKROACH_PORT}");
                 info!(log, "Initializing CRDB Cluster");
                 zone.run_cmd(&[
                     "/opt/oxide/cockroachdb/bin/cockroach",
@@ -1909,6 +1917,11 @@ impl ServiceManager {
                 ])
                 .map_err(|err| Error::CockroachInit { err })?;
                 info!(log, "Formatting CRDB - Completed");
+
+                // In the single-sled case, if there are multiple CRDB nodes on
+                // a single device, we'd still only want to send the
+                // initialization requests to a single dataset.
+                return Ok(());
             }
         }