fix failed login workflow with Redis::Distributed

need to run the pipeline against a singular node

Change-Id: I33befbf4e11daaf5942c37af9233d8f90adde732
Reviewed-on: https://gerrit.instructure.com/c/canvas-lms/+/331124
Reviewed-by: August Thornton <august@instructure.com>
QA-Review: Cody Cutrer <cody@instructure.com>
Product-Review: Cody Cutrer <cody@instructure.com>
Tested-by: Service Cloud Jenkins <svc.cloudjenkins@instructure.com>

gems/canvas_cache/lib/canvas_cache/redis.rb

@@ -135,6 +135,11 @@ module CanvasCache
 
           def_failsafe_method(Distributed, m)
         end
+        # not command methods, so skipped by the above loop
+        # pipelined isn't technically allowed against Distributed,
+        # but we define this anyway so that we get a proper error
+        # about it not being supported, instead of an argument error
+        def_failsafe_method(Distributed, :pipelined)
 
         ::Redis::Scripting::Module.prepend(Scripting::Module) if defined?(::Redis::Scripting::Module)
         ::Redis.prepend(Redis)

lib/canvas/security/login_registry.rb

@@ -92,7 +92,9 @@ module Canvas::Security
 
       key = login_attempts_key(pseudonym)
       exptime = Setting.get("login_attempts_ttl", 5.minutes.to_s).to_i
-      Canvas.redis.pipelined(failsafe: nil) do |pipeline|
+      node = Canvas.redis
+      node = node.node_for(key) if node.respond_to?(:node_for)
+      node.pipelined(failsafe: nil) do |pipeline|
         pipeline.hset(key, "unique_id", pseudonym.unique_id)
         pipeline.hincrby(key, "total", 1)
         pipeline.hincrby(key, ip, 1) if ip.present?