omdb: Add sled state to blueprint displays and diffs (#6545)

On `main`, omdb shows this for the confusing blueprint on dogfood from oxidecomputer/product-assurance#52: ``` root@oxz_switch0:/var/tmp/john# omdb nexus blueprints diff 430f5c6b-3156-4921-8ddc-74560989c8f4 3eb67393-bdbc-4957-98c2-36cc60e3e901 ... snip unchanged sleds ... MODIFIED SLEDS: sled 1efda86b-caef-489f-9792-589d7677e59a: physical disks from generation 1: ----------------------------------- vendor model serial ----------------------------------- - 1b96 WUS4C6432DSP3X3 A079DDFD - 1b96 WUS4C6432DSP3X3 A079DE08 - 1b96 WUS4C6432DSP3X3 A079DE11 - 1b96 WUS4C6432DSP3X3 A079DEA7 - 1b96 WUS4C6432DSP3X3 A079DEAF - 1b96 WUS4C6432DSP3X3 A079DF11 - 1b96 WUS4C6432DSP3X3 A079DFA7 - 1b96 WUS4C6432DSP3X3 A079DFCA - 1b96 WUS4C6432DSP3X3 A079E02E - 1b96 WUS4C6432DSP3X3 A079E076 omicron zones generation 3 -> 4: ------------------------------------------------------------------------------------------- zone type zone id disposition underlay IP ------------------------------------------------------------------------------------------- * crucible 09f14045-df78-447a-b7d8-217e0ca8ee09 - in service fd00:1122:3344:124::24 └─ + expunged * crucible 3ec0e848-39de-495e-be0e-88241e11d0fb - in service fd00:1122:3344:124::22 └─ + expunged * crucible 562c58dc-2408-415e-a005-eb80d1769d10 - in service fd00:1122:3344:124::28 └─ + expunged * crucible 568f842d-3cd3-4901-97f8-94991c2e9938 - in service fd00:1122:3344:124::25 └─ + expunged * crucible 5f3fbd1c-5513-4527-88b5-d07c8fbf71e0 - in service fd00:1122:3344:124::29 └─ + expunged * crucible 6632cd6f-ade4-415f-ad65-b510d4ead12d - in service fd00:1122:3344:124::23 └─ + expunged * crucible 6ca6aa76-c32b-402c-a1ac-751f12d5bdd9 - in service fd00:1122:3344:124::2b └─ + expunged * crucible 9a475569-439c-4749-b78f-eba2096b2131 - in service fd00:1122:3344:124::26 └─ + expunged * crucible bac40327-4eeb-429d-94e1-d3c0525266a2 - in service fd00:1122:3344:124::2a └─ + expunged * crucible eb71bb55-37fb-4fc4-bdee-a5382a480271 - in service fd00:1122:3344:124::27 └─ + expunged * internal_ntp bfea30f1-9ea6-496d-aeb9-ae126ea4f686 - in service fd00:1122:3344:124::21 └─ + expunged ADDED SLEDS: sled 05652dc1-b811-4cac-95e1-d32633f2ba75: COCKROACHDB SETTINGS: state fingerprint::::::::::::::::: d4d87aa2ad877a4cc2fddd0573952362739110de (unchanged) cluster.preserve_downgrade_option: "22.1" (unchanged) METADATA: * internal DNS version: 5 -> 6 external DNS version: 31 (unchanged) ``` As of this branch, we get parenthetical state information on the `sled $SLED_ID:` lines: ``` MODIFIED SLEDS: sled 1efda86b-caef-489f-9792-589d7677e59a (active -> decommissioned): physical disks from generation 1: ----------------------------------- vendor model serial ----------------------------------- - 1b96 WUS4C6432DSP3X3 A079DDFD - 1b96 WUS4C6432DSP3X3 A079DE08 - 1b96 WUS4C6432DSP3X3 A079DE11 - 1b96 WUS4C6432DSP3X3 A079DEA7 - 1b96 WUS4C6432DSP3X3 A079DEAF - 1b96 WUS4C6432DSP3X3 A079DF11 - 1b96 WUS4C6432DSP3X3 A079DFA7 - 1b96 WUS4C6432DSP3X3 A079DFCA - 1b96 WUS4C6432DSP3X3 A079E02E - 1b96 WUS4C6432DSP3X3 A079E076 omicron zones generation 3 -> 4: ------------------------------------------------------------------------------------------- zone type zone id disposition underlay IP ------------------------------------------------------------------------------------------- * crucible 09f14045-df78-447a-b7d8-217e0ca8ee09 - in service fd00:1122:3344:124::24 └─ + expunged * crucible 3ec0e848-39de-495e-be0e-88241e11d0fb - in service fd00:1122:3344:124::22 └─ + expunged * crucible 562c58dc-2408-415e-a005-eb80d1769d10 - in service fd00:1122:3344:124::28 └─ + expunged * crucible 568f842d-3cd3-4901-97f8-94991c2e9938 - in service fd00:1122:3344:124::25 └─ + expunged * crucible 5f3fbd1c-5513-4527-88b5-d07c8fbf71e0 - in service fd00:1122:3344:124::29 └─ + expunged * crucible 6632cd6f-ade4-415f-ad65-b510d4ead12d - in service fd00:1122:3344:124::23 └─ + expunged * crucible 6ca6aa76-c32b-402c-a1ac-751f12d5bdd9 - in service fd00:1122:3344:124::2b └─ + expunged * crucible 9a475569-439c-4749-b78f-eba2096b2131 - in service fd00:1122:3344:124::26 └─ + expunged * crucible bac40327-4eeb-429d-94e1-d3c0525266a2 - in service fd00:1122:3344:124::2a └─ + expunged * crucible eb71bb55-37fb-4fc4-bdee-a5382a480271 - in service fd00:1122:3344:124::27 └─ + expunged * internal_ntp bfea30f1-9ea6-496d-aeb9-ae126ea4f686 - in service fd00:1122:3344:124::21 └─ + expunged ADDED SLEDS: sled 05652dc1-b811-4cac-95e1-d32633f2ba75 (decommissioned): COCKROACHDB SETTINGS: state fingerprint::::::::::::::::: d4d87aa2ad877a4cc2fddd0573952362739110de (unchanged) cluster.preserve_downgrade_option: "22.1" (unchanged) METADATA: * internal DNS version: 5 -> 6 external DNS version: 31 (unchanged) ``` The empty sled added block is still kinda confusing, but I think the note that the added sled is starting out in the `decommissioned` state is at least a reasonable pointer to an explanation of what's going on. Fixes #6544.
oxidecomputer · Sep 13, 2024 · 4c72357 · 4c72357
1 parent 1dcc7f9
commit 4c72357
Show file tree

Hide file tree

Showing 12 changed files with 197 additions and 52 deletions.
diff --git a/dev-tools/omdb/tests/successes.out b/dev-tools/omdb/tests/successes.out
@@ -1265,9 +1265,9 @@ to:   blueprint ......<REDACTED_BLUEPRINT_ID>.......
 
  UNCHANGED SLEDS:
 
-  sled ..........<REDACTED_UUID>...........:
+  sled ..........<REDACTED_UUID>........... (active):
 
-  sled ..........<REDACTED_UUID>...........:
+  sled ..........<REDACTED_UUID>........... (active):
 
     omicron zones at generation 2:
     ---------------------------------------------------------------------------------------

diff --git a/nexus/reconfigurator/planning/src/planner.rs b/nexus/reconfigurator/planning/src/planner.rs
@@ -1861,6 +1861,7 @@ mod test {
         expunged_zones.generation = expunged_zones.generation.next();
 
         blueprint2a.blueprint_zones.remove(&decommissioned_sled_id);
+        blueprint2a.sled_state.remove(&decommissioned_sled_id);
 
         blueprint2a.external_dns_version =
             blueprint2a.external_dns_version.next();

diff --git a/nexus/reconfigurator/planning/tests/output/blueprint_builder_initial_diff.txt b/nexus/reconfigurator/planning/tests/output/blueprint_builder_initial_diff.txt
@@ -2,7 +2,7 @@ from: collection 094d362b-7d79-49e7-a244-134276cca8fe
 to:   blueprint  e4aeb3b3-272f-4967-be34-2d34daa46aa1
  UNCHANGED SLEDS:
 
-  sled 08c7046b-c9c4-4368-881f-19a72df22143:
+  sled 08c7046b-c9c4-4368-881f-19a72df22143 (active):
 
     physical disks at generation 1:
     ----------------------------------------------------------------------
@@ -39,7 +39,7 @@ to:   blueprint  e4aeb3b3-272f-4967-be34-2d34daa46aa1
     nexus          b2573120-9c91-4ed7-8b4f-a7bfe8dbc807   in service    fd00:1122:3344:103::22
 
 
-  sled 84ac367e-9b03-4e9d-a846-df1a08deee6c:
+  sled 84ac367e-9b03-4e9d-a846-df1a08deee6c (active):
 
     physical disks at generation 1:
     ----------------------------------------------------------------------
@@ -76,7 +76,7 @@ to:   blueprint  e4aeb3b3-272f-4967-be34-2d34daa46aa1
     nexus          a9a6a974-8953-4783-b815-da46884f2c02   in service    fd00:1122:3344:101::22
 
 
-  sled be7f4375-2a6b-457f-b1a4-3074a715e5fe:
+  sled be7f4375-2a6b-457f-b1a4-3074a715e5fe (active):
 
     physical disks at generation 1:
     ----------------------------------------------------------------------

diff --git a/nexus/reconfigurator/planning/tests/output/planner_basic_add_sled_2_3.txt b/nexus/reconfigurator/planning/tests/output/planner_basic_add_sled_2_3.txt
@@ -3,7 +3,7 @@ to:   blueprint 4171ad05-89dd-474b-846b-b007e4346366
 
  UNCHANGED SLEDS:
 
-  sled 41f45d9f-766e-4ca6-a881-61ee45c80f57:
+  sled 41f45d9f-766e-4ca6-a881-61ee45c80f57 (active):
 
     physical disks at generation 1:
     ----------------------------------------------------------------------
@@ -40,7 +40,7 @@ to:   blueprint 4171ad05-89dd-474b-846b-b007e4346366
     nexus          cc816cfe-3869-4dde-b596-397d41198628   in service    fd00:1122:3344:103::22
 
 
-  sled 43677374-8d2f-4deb-8a41-eeea506db8e0:
+  sled 43677374-8d2f-4deb-8a41-eeea506db8e0 (active):
 
     physical disks at generation 1:
     ----------------------------------------------------------------------
@@ -77,7 +77,7 @@ to:   blueprint 4171ad05-89dd-474b-846b-b007e4346366
     nexus          3eda924f-22a9-4f3e-9a1b-91d1c47601ab   in service    fd00:1122:3344:101::22
 
 
-  sled 590e3034-d946-4166-b0e5-2d0034197a07:
+  sled 590e3034-d946-4166-b0e5-2d0034197a07 (active):
 
     physical disks at generation 1:
     ----------------------------------------------------------------------
@@ -116,7 +116,7 @@ to:   blueprint 4171ad05-89dd-474b-846b-b007e4346366
 
  ADDED SLEDS:
 
-  sled b59ec570-2abb-4017-80ce-129d94e7a025:
+  sled b59ec570-2abb-4017-80ce-129d94e7a025 (active):
 
     physical disks at generation 1:
     ----------------------------------------------------------------------

diff --git a/nexus/reconfigurator/planning/tests/output/planner_basic_add_sled_3_5.txt b/nexus/reconfigurator/planning/tests/output/planner_basic_add_sled_3_5.txt
@@ -3,7 +3,7 @@ to:   blueprint f432fcd5-1284-4058-8b4a-9286a3de6163
 
  UNCHANGED SLEDS:
 
-  sled 41f45d9f-766e-4ca6-a881-61ee45c80f57:
+  sled 41f45d9f-766e-4ca6-a881-61ee45c80f57 (active):
 
     physical disks at generation 1:
     ----------------------------------------------------------------------
@@ -40,7 +40,7 @@ to:   blueprint f432fcd5-1284-4058-8b4a-9286a3de6163
     nexus          cc816cfe-3869-4dde-b596-397d41198628   in service    fd00:1122:3344:103::22
 
 
-  sled 43677374-8d2f-4deb-8a41-eeea506db8e0:
+  sled 43677374-8d2f-4deb-8a41-eeea506db8e0 (active):
 
     physical disks at generation 1:
     ----------------------------------------------------------------------
@@ -77,7 +77,7 @@ to:   blueprint f432fcd5-1284-4058-8b4a-9286a3de6163
     nexus          3eda924f-22a9-4f3e-9a1b-91d1c47601ab   in service    fd00:1122:3344:101::22
 
 
-  sled 590e3034-d946-4166-b0e5-2d0034197a07:
+  sled 590e3034-d946-4166-b0e5-2d0034197a07 (active):
 
     physical disks at generation 1:
     ----------------------------------------------------------------------
@@ -116,7 +116,7 @@ to:   blueprint f432fcd5-1284-4058-8b4a-9286a3de6163
 
  MODIFIED SLEDS:
 
-  sled b59ec570-2abb-4017-80ce-129d94e7a025:
+  sled b59ec570-2abb-4017-80ce-129d94e7a025 (active):
 
     physical disks at generation 1:
     ----------------------------------------------------------------------

diff --git a/nexus/reconfigurator/planning/tests/output/planner_decommissions_sleds_1_2.txt b/nexus/reconfigurator/planning/tests/output/planner_decommissions_sleds_1_2.txt
@@ -3,7 +3,7 @@ to:   blueprint 1ac2d88f-27dd-4506-8585-6b2be832528e
 
  MODIFIED SLEDS:
 
-  sled a1b477db-b629-48eb-911d-1ccdafca75b9:
+  sled a1b477db-b629-48eb-911d-1ccdafca75b9 (active -> decommissioned):
 
     physical disks from generation 1:
     ----------------------------------------------------------------------
@@ -53,7 +53,7 @@ to:   blueprint 1ac2d88f-27dd-4506-8585-6b2be832528e
      └─                                                   + expunged                           
 
 
-  sled d67ce8f0-a691-4010-b414-420d82e80527:
+  sled d67ce8f0-a691-4010-b414-420d82e80527 (active):
 
     physical disks at generation 1:
     ----------------------------------------------------------------------
@@ -91,7 +91,7 @@ to:   blueprint 1ac2d88f-27dd-4506-8585-6b2be832528e
 +   nexus          ff9ce09c-afbf-425b-bbfa-3d8fb254f98e   in service    fd00:1122:3344:101::2d
 
 
-  sled fefcf4cf-f7e7-46b3-b629-058526ce440e:
+  sled fefcf4cf-f7e7-46b3-b629-058526ce440e (active):
 
     physical disks at generation 1:
     ----------------------------------------------------------------------

diff --git a/nexus/reconfigurator/planning/tests/output/planner_decommissions_sleds_bp2.txt b/nexus/reconfigurator/planning/tests/output/planner_decommissions_sleds_bp2.txt
@@ -1,7 +1,7 @@
 blueprint  1ac2d88f-27dd-4506-8585-6b2be832528e
 parent:    516e80a3-b362-4fac-bd3c-4559717120dd
 
-  sled: d67ce8f0-a691-4010-b414-420d82e80527
+  sled: d67ce8f0-a691-4010-b414-420d82e80527 (active)
 
     physical disks at generation 1:
     ----------------------------------------------------------------------
@@ -40,7 +40,7 @@ parent:    516e80a3-b362-4fac-bd3c-4559717120dd
 
 
 
-  sled: fefcf4cf-f7e7-46b3-b629-058526ce440e
+  sled: fefcf4cf-f7e7-46b3-b629-058526ce440e (active)
 
     physical disks at generation 1:
     ----------------------------------------------------------------------

diff --git a/nexus/reconfigurator/planning/tests/output/planner_nonprovisionable_1_2.txt b/nexus/reconfigurator/planning/tests/output/planner_nonprovisionable_1_2.txt
@@ -3,7 +3,7 @@ to:   blueprint 9f71f5d3-a272-4382-9154-6ea2e171a6c6
 
  UNCHANGED SLEDS:
 
-  sled 2d1cb4f2-cf44-40fc-b118-85036eb732a9:
+  sled 2d1cb4f2-cf44-40fc-b118-85036eb732a9 (active):
 
     physical disks at generation 1:
     ----------------------------------------------------------------------
@@ -42,7 +42,7 @@ to:   blueprint 9f71f5d3-a272-4382-9154-6ea2e171a6c6
 
  MODIFIED SLEDS:
 
-  sled 48d95fef-bc9f-4f50-9a53-1e075836291d:
+  sled 48d95fef-bc9f-4f50-9a53-1e075836291d (active -> decommissioned):
 
     physical disks from generation 1:
     ----------------------------------------------------------------------
@@ -92,7 +92,7 @@ to:   blueprint 9f71f5d3-a272-4382-9154-6ea2e171a6c6
      └─                                                   + expunged                           
 
 
-  sled 68d24ac5-f341-49ea-a92a-0381b52ab387:
+  sled 68d24ac5-f341-49ea-a92a-0381b52ab387 (active):
 
     physical disks from generation 1:
     ----------------------------------------------------------------------
@@ -129,7 +129,7 @@ to:   blueprint 9f71f5d3-a272-4382-9154-6ea2e171a6c6
     nexus          6464d025-4652-4948-919e-740bec5699b1   expunged      fd00:1122:3344:102::22
 
 
-  sled 75bc286f-2b4b-482c-9431-59272af529da:
+  sled 75bc286f-2b4b-482c-9431-59272af529da (active):
 
     physical disks at generation 1:
     ----------------------------------------------------------------------
@@ -169,7 +169,7 @@ to:   blueprint 9f71f5d3-a272-4382-9154-6ea2e171a6c6
 +   nexus          59950bc8-1497-44dd-8cbf-b6502ba921b2   in service    fd00:1122:3344:104::2f
 
 
-  sled affab35f-600a-4109-8ea0-34a067a4e0bc:
+  sled affab35f-600a-4109-8ea0-34a067a4e0bc (active):
 
     physical disks at generation 1:
     ----------------------------------------------------------------------

diff --git a/nexus/reconfigurator/planning/tests/output/planner_nonprovisionable_2_2a.txt b/nexus/reconfigurator/planning/tests/output/planner_nonprovisionable_2_2a.txt
@@ -3,7 +3,7 @@ to:   blueprint 9f71f5d3-a272-4382-9154-6ea2e171a6c6
 
  UNCHANGED SLEDS:
 
-  sled 75bc286f-2b4b-482c-9431-59272af529da:
+  sled 75bc286f-2b4b-482c-9431-59272af529da (active):
 
     physical disks at generation 1:
     ----------------------------------------------------------------------
@@ -43,7 +43,7 @@ to:   blueprint 9f71f5d3-a272-4382-9154-6ea2e171a6c6
     nexus          59950bc8-1497-44dd-8cbf-b6502ba921b2   in service    fd00:1122:3344:104::2f
 
 
-  sled affab35f-600a-4109-8ea0-34a067a4e0bc:
+  sled affab35f-600a-4109-8ea0-34a067a4e0bc (active):
 
     physical disks at generation 1:
     ----------------------------------------------------------------------
@@ -85,7 +85,7 @@ to:   blueprint 9f71f5d3-a272-4382-9154-6ea2e171a6c6
 
  REMOVED SLEDS:
 
-  sled 68d24ac5-f341-49ea-a92a-0381b52ab387:
+  sled 68d24ac5-f341-49ea-a92a-0381b52ab387 (was active):
 
     omicron zones from generation 2:
     ------------------------------------------------------------------------------------------
@@ -108,7 +108,7 @@ to:   blueprint 9f71f5d3-a272-4382-9154-6ea2e171a6c6
 
  MODIFIED SLEDS:
 
-  sled 2d1cb4f2-cf44-40fc-b118-85036eb732a9:
+  sled 2d1cb4f2-cf44-40fc-b118-85036eb732a9 (active):
 
     physical disks at generation 1:
     ----------------------------------------------------------------------
@@ -144,7 +144,7 @@ to:   blueprint 9f71f5d3-a272-4382-9154-6ea2e171a6c6
      └─                                                   + quiesced                           
 
 
-  sled 48d95fef-bc9f-4f50-9a53-1e075836291d:
+  sled 48d95fef-bc9f-4f50-9a53-1e075836291d (decommissioned):
 
     omicron zones generation 3 -> 4:
     ------------------------------------------------------------------------------------------

diff --git a/nexus/reconfigurator/planning/tests/output/planner_nonprovisionable_bp2.txt b/nexus/reconfigurator/planning/tests/output/planner_nonprovisionable_bp2.txt
@@ -1,7 +1,7 @@
 blueprint  9f71f5d3-a272-4382-9154-6ea2e171a6c6
 parent:    4d4e6c38-cd95-4c4e-8f45-6af4d686964b
 
-  sled: 2d1cb4f2-cf44-40fc-b118-85036eb732a9
+  sled: 2d1cb4f2-cf44-40fc-b118-85036eb732a9 (active)
 
     physical disks at generation 1:
     ----------------------------------------------------------------------
@@ -39,7 +39,7 @@ parent:    4d4e6c38-cd95-4c4e-8f45-6af4d686964b
 
 
 
-  sled: 75bc286f-2b4b-482c-9431-59272af529da
+  sled: 75bc286f-2b4b-482c-9431-59272af529da (active)
 
     physical disks at generation 1:
     ----------------------------------------------------------------------
@@ -80,7 +80,7 @@ parent:    4d4e6c38-cd95-4c4e-8f45-6af4d686964b
 
 
 
-  sled: affab35f-600a-4109-8ea0-34a067a4e0bc
+  sled: affab35f-600a-4109-8ea0-34a067a4e0bc (active)
 
     physical disks at generation 1:
     ----------------------------------------------------------------------

diff --git a/nexus/types/src/deployment.rs b/nexus/types/src/deployment.rs
@@ -234,18 +234,20 @@ impl Blueprint {
     pub fn diff_since_blueprint(&self, before: &Blueprint) -> BlueprintDiff {
         BlueprintDiff::new(
             DiffBeforeMetadata::Blueprint(Box::new(before.metadata())),
+            before.sled_state.clone(),
             before
                 .blueprint_zones
                 .iter()
                 .map(|(sled_id, zones)| (*sled_id, zones.clone().into()))
                 .collect(),
-            self.metadata(),
-            self.blueprint_zones.clone(),
             before
                 .blueprint_disks
                 .iter()
                 .map(|(sled_id, disks)| (*sled_id, disks.clone().into()))
                 .collect(),
+            self.metadata(),
+            self.sled_state.clone(),
+            self.blueprint_zones.clone(),
             self.blueprint_disks.clone(),
         )
     }
@@ -260,6 +262,14 @@ impl Blueprint {
     /// disposition, so it is assumed that all zones in the collection have the
     /// [`InService`](BlueprintZoneDisposition::InService) disposition.
     pub fn diff_since_collection(&self, before: &Collection) -> BlueprintDiff {
+        // We'll assume any sleds present in a collection were active; if they
+        // were decommissioned they wouldn't be present.
+        let before_state = before
+            .sled_agents
+            .keys()
+            .map(|sled_id| (*sled_id, SledState::Active))
+            .collect();
+
         let before_zones = before
             .omicron_zones
             .iter()
@@ -288,10 +298,12 @@ impl Blueprint {
 
         BlueprintDiff::new(
             DiffBeforeMetadata::Collection { id: before.id },
+            before_state,
             before_zones,
+            before_disks,
             self.metadata(),
+            self.sled_state.clone(),
             self.blueprint_zones.clone(),
-            before_disks,
             self.blueprint_disks.clone(),
         )
     }
@@ -438,6 +450,16 @@ impl<'a> fmt::Display for BlueprintDisplay<'a> {
                 disks.rows(BpDiffState::Unchanged).collect(),
             );
 
+            // Look up the sled state
+            let sled_state = self
+                .blueprint
+                .sled_state
+                .get(sled_id)
+                .map(|state| state.to_string())
+                .unwrap_or_else(|| {
+                    "blueprint error: unknown sled state".to_string()
+                });
+
             // Construct the zones subtable
             match self.blueprint.blueprint_zones.get(sled_id) {
                 Some(zones) => {
@@ -450,10 +472,13 @@ impl<'a> fmt::Display for BlueprintDisplay<'a> {
                     );
                     writeln!(
                         f,
-                        "\n  sled: {sled_id}\n\n{disks_table}\n\n{zones_tab}\n"
+                        "\n  sled: {sled_id} ({sled_state})\n\n{disks_table}\n\n{zones_tab}\n"
                     )?;
                 }
-                None => writeln!(f, "\n  sled: {sled_id}\n\n{disks_table}\n")?,
+                None => writeln!(
+                    f,
+                    "\n  sled: {sled_id} ({sled_state})\n\n{disks_table}\n"
+                )?,
             }
             seen_sleds.insert(sled_id);
         }