fix(ext-plugin): don't use stale key (apache#5782)

Co-authored-by: sunaowei <sunaowei@neild>

apisix/plugins/ext-plugin/init.lua

@@ -63,11 +63,18 @@ local type = type
 
 
 local events_list
-local lrucache = core.lrucache.new({
-    type = "plugin",
-    invalid_stale = true,
-    ttl = helper.get_conf_token_cache_time(),
-})
+
+local function new_lrucache()
+    return core.lrucache.new({
+        type = "plugin",
+        invalid_stale = true,
+        ttl = helper.get_conf_token_cache_time(),
+    })
+end
+local lrucache = new_lrucache()
+
+local shdict_name = "ext-plugin"
+local shdict = ngx.shared[shdict_name]
 
 local schema = {
     type = "object",
@@ -588,15 +595,14 @@ rpc_call = function (ty, conf, ctx, ...)
 end
 
 
-local function create_lrucache()
+local function recreate_lrucache()
+    flush_token()
+
     if lrucache then
         core.log.warn("flush conf token lrucache")
     end
 
-    lrucache = core.lrucache.new({
-        type = "plugin",
-        ttl = helper.get_conf_token_cache_time(),
-    })
+    lrucache = new_lrucache()
 end
 
 
@@ -620,7 +626,7 @@ function _M.communicate(conf, ctx, plugin_name)
         end
 
         core.log.warn("refresh cache and try again")
-        create_lrucache()
+        recreate_lrucache()
     end
 
     core.log.error(err)
@@ -717,7 +723,7 @@ function _M.init_worker()
     )
 
     -- flush cache when runner exited
-    events.register(create_lrucache, events_list._source, events_list.runner_exit)
+    events.register(recreate_lrucache, events_list._source, events_list.runner_exit)
 
     -- note that the runner is run under the same user as the Nginx master
     if process.type() == "privileged agent" then