Fix expel of the last vshard instance

tarantool · Jul 3, 2024 · ab31a0b · ab31a0b
1 parent 6f079b4
commit ab31a0b
Show file tree

Hide file tree

Showing 4 changed files with 68 additions and 9 deletions.
diff --git a/CHANGELOG.rst b/CHANGELOG.rst
@@ -12,6 +12,18 @@ and this project adheres to
 Unreleased
 -------------------------------------------------------------------------------
 
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+Added
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+- VShard alerts are displayed in issues list.
+
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+Fixed
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+- Last instance in vshard-storage replicaset can be expelled now.
+
 -------------------------------------------------------------------------------
 [2.12.2] - 2024-06-24
 -------------------------------------------------------------------------------

diff --git a/cartridge/roles/vshard-router.lua b/cartridge/roles/vshard-router.lua
@@ -291,6 +291,21 @@ local function bootstrap()
     return true
 end
 
+local function get_issues()
+    local issues = table.deepcopy(vars.issues)
+    if vshard.router.info == nil then
+        return issues
+    end
+    for _, alert in ipairs(vshard.router.info().alerts) do
+        table.insert(issues, {
+            level = 'warning',
+            topic = 'vshard',
+            message = alert[2],
+        })
+    end
+    return issues
+end
+
 return {
     role_name = 'vshard-router',
     implies_router = true,
@@ -299,8 +314,11 @@ return {
     validate_config = vshard_utils.validate_config,
     apply_config = apply_config,
     stop = stop,
-    get_issues = function() return vars.issues end,
+    get_issues = get_issues,
 
     get = get,
     bootstrap = bootstrap,
+    get_alerts = function()
+        return vshard and vshard.router and vshard.router.info and vshard.router.info().alerts or {}
+    end,
 }
diff --git a/cartridge/roles/vshard-storage.lua b/cartridge/roles/vshard-storage.lua
@@ -109,6 +109,21 @@ local function stop()
     _G_vshard_backup = nil
 end
 
+local function get_issues()
+    local issues = table.deepcopy(vars.issues)
+    if vshard.storage.info == nil then
+        return issues
+    end
+    for _, alert in ipairs(vshard.storage.info().alerts) do
+        table.insert(issues, {
+            level = 'warning',
+            topic = 'vshard',
+            message = alert[2],
+        })
+    end
+    return issues
+end
+
 return {
     role_name = 'vshard-storage',
     implies_storage = true,
@@ -117,5 +132,5 @@ return {
     on_apply_config = on_apply_config,
     init = init,
     stop = stop,
-    get_issues = function() return vars.issues end,
+    get_issues = get_issues,
 }
diff --git a/cartridge/vshard-utils.lua b/cartridge/vshard-utils.lua
@@ -13,6 +13,7 @@ local topology = require('cartridge.topology')
 local failover = require('cartridge.failover')
 local twophase = require('cartridge.twophase')
 local confapplier = require('cartridge.confapplier')
+local rpc = require('cartridge.rpc')
 
 local ok, vshard_consts = pcall(require, 'vshard-ee.consts')
 if not ok then
@@ -152,14 +153,27 @@ local function validate_group_upgrade(group_name, topology_new, topology_old)
             )
 
             if buckets_count == nil then
-                error(err)
-            end
 
-            ValidateConfigError:assert(
-                buckets_count == 0,
-                "replicasets[%s] rebalancing isn't finished yet",
-                replicaset_uuid
-            )
+                -- vshard storage is probably off
+                -- so we need to call a router to get the alerts
+                local alerts, call_err = rpc.call('vshard-router', 'get_alerts')
+                if alerts == nil then
+                    error(call_err)
+                end
+                for _, alert in ipairs(alerts) do
+                    if alert[1] == 'UNKNOWN_BUCKETS' then
+                        error(alert[2] ..
+                            '. Please make sure that all buckets are safe ' ..
+                            'before making any changes')
+                    end
+                end
+            else
+                ValidateConfigError:assert(
+                    buckets_count == 0,
+                    "replicasets[%s] rebalancing isn't finished yet",
+                    replicaset_uuid
+                )
+            end
         end
 
     end