feat(health): add health checks for GitOps Promoter (#23663)

Signed-off-by: Michael Crenshaw <[email protected]>

Author: crenshaw-dev

resource_customizations/promoter.argoproj.io/ArgoCDCommitStatus/health.lua

@@ -0,0 +1,50 @@
+local hs = {}
+hs.status = "Progressing"
+hs.message = "Initializing Argo CD commit status"
+
+-- Check for deletion timestamp
+if obj.metadata.deletionTimestamp then
+    hs.status = "Progressing"
+    hs.message = "Argo CD commit status is being deleted"
+    return hs
+end
+
+-- Check if status exists
+if not obj.status then
+    return hs
+end
+
+-- Check for reconciliation conditions
+local hasReadyCondition = false
+if obj.status.conditions then
+    for _, condition in ipairs(obj.status.conditions) do
+        if condition.type == "Ready" then
+            hasReadyCondition = true
+            if condition.observedGeneration and obj.metadata.generation and condition.observedGeneration ~= obj.metadata.generation then
+                hs.status = "Progressing"
+                hs.message = "Waiting for Argo CD commit status spec update to be observed"
+                return hs
+            end
+            if condition.status == "False" and condition.reason == "ReconciliationError" then
+                hs.status = "Degraded"
+                hs.message = "Argo CD commit status reconciliation failed: " .. (condition.message or "Unknown error")
+                return hs
+            end
+        end
+    end
+end
+if not hasReadyCondition then
+    hs.status = "Progressing"
+    hs.message = "Argo CD commit status is not ready yet"
+    return hs
+end
+
+if not obj.status.applicationsSelected or #obj.status.applicationsSelected == 0 then
+    hs.status = "Degraded"
+    hs.message = "Argo CD commit status has no applications configured"
+    return hs
+end
+
+hs.status = "Healthy"
+hs.message = "Argo CD commit status is healthy and is tracking " .. #obj.status.applicationsSelected .. " applications"
+return hs

resource_customizations/promoter.argoproj.io/ArgoCDCommitStatus/health_test.yaml

@@ -0,0 +1,29 @@
+tests:
+- healthStatus:
+    status: Progressing
+    message: Initializing Argo CD commit status
+  inputPath: testdata/no-status.yaml
+- healthStatus:
+    status: Progressing
+    message: Argo CD commit status is being deleted
+  inputPath: testdata/deleting.yaml
+- healthStatus:
+    status: Progressing
+    message: Waiting for Argo CD commit status spec update to be observed
+  inputPath: testdata/observed-generation-outdated.yaml
+- healthStatus:
+    status: Degraded
+    message: "Argo CD commit status reconciliation failed: Something went wrong"
+  inputPath: testdata/reconcile-error.yaml
+- healthStatus:
+    status: Progressing
+    message: Argo CD commit status is not ready yet
+  inputPath: testdata/no-ready-condition.yaml
+- healthStatus:
+    status: Degraded
+    message: Argo CD commit status has no applications configured
+  inputPath: testdata/no-applications.yaml
+- healthStatus:
+    status: Healthy
+    message: Argo CD commit status is healthy and is tracking 2 applications
+  inputPath: testdata/all-healthy.yaml

resource_customizations/promoter.argoproj.io/ArgoCDCommitStatus/testdata/all-healthy.yaml

@@ -0,0 +1,20 @@
+apiVersion: promoter.argoproj.io/v1alpha1
+kind: ArgoCDCommitStatus
+metadata:
+  name: test
+  generation: 2
+status:
+  conditions:
+    - type: Ready
+      status: True
+      observedGeneration: 2
+  applicationsSelected:
+    - name: app1
+      namespace: default
+      phase: success
+      sha: abc1234
+    - name: app2
+      namespace: default
+      phase: success
+      sha: def5678
+

resource_customizations/promoter.argoproj.io/ArgoCDCommitStatus/testdata/deleting.yaml

@@ -0,0 +1,5 @@
+apiVersion: promoter.argoproj.io/v1alpha1
+kind: ArgoCDCommitStatus
+metadata:
+  name: test
+  deletionTimestamp: "2025-07-04T12:00:00Z"

resource_customizations/promoter.argoproj.io/ArgoCDCommitStatus/testdata/no-applications.yaml

@@ -0,0 +1,12 @@
+apiVersion: promoter.argoproj.io/v1alpha1
+kind: ArgoCDCommitStatus
+metadata:
+  name: test
+  generation: 2
+status:
+  conditions:
+    - type: Ready
+      status: True
+      observedGeneration: 2
+  applicationsSelected: []
+

resource_customizations/promoter.argoproj.io/ArgoCDCommitStatus/testdata/no-ready-condition.yaml

@@ -0,0 +1,11 @@
+apiVersion: promoter.argoproj.io/v1alpha1
+kind: ArgoCDCommitStatus
+metadata:
+  name: test
+  generation: 2
+status:
+  conditions:
+    - type: SomeOtherCondition
+      status: True
+      observedGeneration: 2
+

resource_customizations/promoter.argoproj.io/ArgoCDCommitStatus/testdata/no-status.yaml

@@ -0,0 +1,4 @@
+apiVersion: promoter.argoproj.io/v1alpha1
+kind: ArgoCDCommitStatus
+metadata:
+  name: test

resource_customizations/promoter.argoproj.io/ArgoCDCommitStatus/testdata/observed-generation-outdated.yaml

@@ -0,0 +1,11 @@
+apiVersion: promoter.argoproj.io/v1alpha1
+kind: ArgoCDCommitStatus
+metadata:
+  name: test
+  generation: 2
+status:
+  conditions:
+    - type: Ready
+      status: True
+      observedGeneration: 1
+

resource_customizations/promoter.argoproj.io/ArgoCDCommitStatus/testdata/reconcile-error.yaml

@@ -0,0 +1,12 @@
+apiVersion: promoter.argoproj.io/v1alpha1
+kind: ArgoCDCommitStatus
+metadata:
+  name: test
+  generation: 2
+status:
+  conditions:
+    - type: Ready
+      status: "False"
+      reason: ReconciliationError
+      message: Something went wrong
+      observedGeneration: 2

resource_customizations/promoter.argoproj.io/ChangeTransferPolicy/health.lua

@@ -0,0 +1,121 @@
+local hs = {}
+hs.status = "Progressing"
+hs.message = "Initializing change transfer policy"
+
+-- Check for deletion timestamp
+if obj.metadata.deletionTimestamp then
+    hs.status = "Progressing"
+    hs.message = "Change transfer policy is being deleted"
+    return hs
+end
+
+-- Check if status exists
+if not obj.status then
+    return hs
+end
+
+-- Check for reconciliation conditions
+local hasReadyCondition = false
+if obj.status.conditions then
+    for _, condition in ipairs(obj.status.conditions) do
+        if condition.type == "Ready" then
+            hasReadyCondition = true
+            -- Check observedGeneration vs metadata.generation within the reconciliation condition
+            if condition.observedGeneration and obj.metadata.generation and condition.observedGeneration ~= obj.metadata.generation then
+                hs.status = "Progressing"
+                hs.message = "Waiting for change transfer policy spec update to be observed"
+                return hs
+            end
+            if condition.status == "False" and condition.reason == "ReconciliationError" then
+                hs.status = "Degraded"
+                hs.message = "Change transfer policy reconciliation failed: " .. (condition.message or "Unknown error")
+                return hs
+            end
+        end
+    end
+end
+if not hasReadyCondition then
+    hs.status = "Progressing"
+    hs.message = "Change transfer policy is not ready yet"
+    return hs
+end
+
+if not obj.status.active or not obj.status.active.dry or not obj.status.active.dry.sha or obj.status.active.dry.sha == "" then
+    hs.status = "Progressing"
+    hs.message = "The active commit DRY SHA is missing or empty."
+    return hs
+end
+if not obj.status.proposed or not obj.status.proposed.dry or not obj.status.proposed.dry.sha or obj.status.proposed.dry.sha == "" then
+    hs.status = "Progressing"
+    hs.message = "The proposed commit DRY SHA is missing or empty."
+    return hs
+end
+
+-- Helper function to get short SHA
+local function getShortSha(sha)
+    if not sha or sha == "" then
+        return ""
+    end
+    if string.len(sha) > 7 then
+        return string.sub(sha, 1, 7)
+    end
+    return sha
+end
+
+if obj.status.proposed.dry.sha ~= obj.status.active.dry.sha then
+    local pendingCount = 0
+    local successCount = 0
+    local failureCount = 0
+    local pendingKeys = {}
+    local failedKeys = {}
+
+    for _, status in ipairs(obj.status.proposed.commitStatuses or {}) do
+        if status.phase == "pending" then
+            pendingCount = pendingCount + 1
+            table.insert(pendingKeys, status.key)
+        elseif status.phase == "success" then
+            successCount = successCount + 1
+        elseif status.phase == "failure" then
+            failureCount = failureCount + 1
+            table.insert(failedKeys, status.key)
+        end
+    end
+
+    hs.status = "Progressing"
+    hs.message =
+        "Promotion in progress from '" .. getShortSha(obj.status.active.dry.sha) ..
+        "' to '" .. getShortSha(obj.status.proposed.dry.sha) .. "': " ..
+        pendingCount .. " pending, " .. successCount .. " successful, " .. failureCount .. " failed. "
+
+    if pendingCount > 0 then
+        hs.message = hs.message .. "Pending commit statuses: " .. table.concat(pendingKeys, ", ") .. ". "
+    end
+    if failureCount > 0 then
+        hs.message = hs.message .. "Failed commit statuses: " .. table.concat(failedKeys, ", ") .. ". "
+    end
+    return hs
+end
+
+local pendingCount = 0
+local failureCount = 0
+for _, status in ipairs(obj.status.active.commitStatuses or {}) do
+    if status.phase == "pending" then
+        pendingCount = pendingCount + 1
+    elseif status.phase == "failure" then
+        failureCount = failureCount + 1
+    end
+end
+if pendingCount > 0 or failureCount > 0 then
+    local envMessages = {}
+    for branch, counts in pairs(nonSuccessfulEnvironments) do
+        local msg = branch .. " (" .. counts.failure .. " failed, " .. counts.pending .. " pending)"
+        table.insert(envMessages, msg)
+    end
+    hs.status = "Healthy"
+    hs.message = "Environment is up-to-date, but there are non-successful commit statuses: " .. table.concat(envMessages, ", ") .. "."
+    return hs
+end
+
+hs.status = "Healthy"
+hs.message = "Environment is up-to-date on commit " .. getShortSha(obj.status.active.dry.sha) .. "."
+return hs