Added new affinity mode for maximum session stickyness. Fixes kubernetes/ingress-nginx#4475

rootfs/etc/nginx/lua/affinity/balanced.lua

@@ -0,0 +1,57 @@
+-- An affinity mode which makes sure connections are rebalanced when a deployment is scaled.
+-- The advantage of this mode is that the load on the pods will be redistributed.
+-- The drawback of this mode is that, when scaling up a deployment, roughly (n-c)/n users 
+-- will lose their session, where c is the current number of pods and n is the new number of 
+-- pods.
+--
+-- This class extends/implements the abstract class balancer.sticky.
+--
+local math = require("math")
+local resty_chash = require("resty.chash")
+local util = require("util")
+
+local _M = {}
+
+-- Consider the situation of N upstreams one of which is failing.
+-- Then the probability to obtain failing upstream after M iterations would be close to (1/N)**M.
+-- For the worst case (2 upstreams; 20 iterations) it would be ~10**(-6)
+-- which is much better then ~10**(-3) for 10 iterations.
+local MAX_UPSTREAM_CHECKS_COUNT = 20
+
+local function get_routing_key(self)
+  return self:get_cookie(), nil
+end
+
+local function set_routing_key(self, key)
+	self:set_cookie(key)
+end
+
+local function pick_new_upstream(self, failed_upstreams)
+  for i = 1, MAX_UPSTREAM_CHECKS_COUNT do
+    local key = string.format("%s.%s.%s", ngx.now() + i, ngx.worker.pid(), math.random(999999))
+
+    local new_upstream = self.instance:find(key)
+
+    if not failed_upstreams[new_upstream] then
+      return new_upstream, key
+    end
+  end
+
+  return nil, nil
+end
+
+function _M.new(self, sticky_balancer, backend)
+  local o = sticky_balancer or {}
+  
+  local nodes = util.get_nodes(backend.endpoints)
+
+  -- override sticky.balancer methods
+  o.instance = resty_chash:new(nodes)
+  o.get_routing_key = get_routing_key
+  o.set_routing_key = set_routing_key
+  o.pick_new_upstream = pick_new_upstream
+
+  return sticky_balancer
+end
+
+return _M

rootfs/etc/nginx/lua/affinity/persistent.lua

@@ -0,0 +1,53 @@
+-- An affinity mode which makes sure a session is always routed to the same endpoint.
+-- The advantage of this mode is that a user will never lose his session.
+-- The drawback of this mode is that when scaling up a deployment, sessions will not
+-- be rebalanced.
+--
+local util = require("util")
+local util_nodemap = require("util.nodemap")
+
+local _M = {}
+
+local function get_routing_key(self)
+  local cookie_value = self:get_cookie()
+
+  if cookie_value then
+    -- format <timestamp>.<workder-pid>.<routing-key>
+    local routing_key = string.match(cookie_value, '[^\\.]+$')
+
+    if routing_key == nil then
+      local err = string.format("Failed to extract routing key from cookie '%s'!", cookie_value)
+      return nil, err
+    end
+
+    return routing_key, nil
+  end
+
+  return nil, nil
+end
+
+local function set_routing_key(self, key)
+  local value = string.format("%s.%s.%s", ngx.now(), ngx.worker.pid(), key)
+  self:set_cookie(value);
+end
+
+local function pick_new_upstream(self, failed_upstreams)
+  return self.instance:random_except(failed_upstreams)
+end
+
+function _M.new(self, sticky_balancer, backend)
+  local o = sticky_balancer or {}
+  
+  local nodes = util.get_nodes(backend.endpoints)
+  local hash_salt = backend["name"]
+
+  -- override sticky.balancer methods
+  o.instance = util_nodemap:new(nodes, hash_salt)
+  o.get_routing_key = get_routing_key
+  o.set_routing_key = set_routing_key
+  o.pick_new_upstream = pick_new_upstream
+
+  return sticky_balancer
+end
+
+return _M

rootfs/etc/nginx/lua/balancer/sticky.lua

@@ -1,8 +1,8 @@
+local affinity_balanced = require("affinity.balanced")
+local affinity_persistent = require("affinity.persistent")
 local balancer_resty = require("balancer.resty")
-local resty_chash = require("resty.chash")
 local util = require("util")
 local ck = require("resty.cookie")
-local math = require("math")
 local ngx_balancer = require("ngx.balancer")
 local split = require("util.split")
 
@@ -10,34 +10,60 @@ local string_format = string.format
 local ngx_log = ngx.log
 local INFO = ngx.INFO
 
-local _M = balancer_resty:new({ factory = resty_chash, name = "sticky" })
+local _M = balancer_resty:new({ name = "sticky" })
 local DEFAULT_COOKIE_NAME = "route"
 
--- Consider the situation of N upstreams one of which is failing.
--- Then the probability to obtain failing upstream after M iterations would be close to (1/N)**M.
--- For the worst case (2 upstreams; 20 iterations) it would be ~10**(-6)
--- which is much better then ~10**(-3) for 10 iterations.
-local MAX_UPSTREAM_CHECKS_COUNT = 20
 
 function _M.cookie_name(self)
   return self.cookie_session_affinity.name or DEFAULT_COOKIE_NAME
 end
 
-function _M.new(self, backend)
-  local nodes = util.get_nodes(backend.endpoints)
+local function init_affinity_mode(self, backend)
+  local mode = backend["sessionAffinityConfig"]["mode"] or 'balanced'
 
+  -- set default mode to 'balanced' for backwards compatibility
+  if mode == nil or mode == '' then
+    mode = 'balanced'
+  end
+
+  self.affinity_mode = mode
+
+  if mode == 'persistent' then
+    return affinity_persistent:new(self, backend)
+  end
+
+  -- default is 'balanced' for backwards compatibility
+  if mode ~= 'balanced' then
+    ngx.log(ngx.WARN, string.format("Invalid affinity mode '%s'! Using 'balanced' as a default.", mode))
+  end
+
+  return affinity_balanced:new(self, backend)
+end
+
+function _M.new(self, backend)
   local o = {
-    instance = self.factory:new(nodes),
+    instance = nil,
+    affinity_mode = nil,
     traffic_shaping_policy = backend.trafficShapingPolicy,
     alternative_backends = backend.alternativeBackends,
     cookie_session_affinity = backend["sessionAffinityConfig"]["cookieSessionAffinity"]
   }
   setmetatable(o, self)
   self.__index = self
-  return o
+  
+  return init_affinity_mode(o, backend)
 end
 
-local function set_cookie(self, value)
+function _M.get_cookie(self)
+  local cookie, err = ck:new()
+  if not cookie then
+    ngx.log(ngx.ERR, err)
+  end
+
+  return cookie:get(self:cookie_name())
+end
+
+function _M.set_cookie(self, value)
   local cookie, err = ck:new()
   if not cookie then
     ngx.log(ngx.ERR, err)
@@ -86,19 +112,30 @@ local function get_failed_upstreams()
   return indexed_upstream_addrs
 end
 
-local function pick_new_upstream(self)
-  local failed_upstreams = get_failed_upstreams()
+--- get_routing_key gets the current routing key from the cookie
+-- @treturn string, string The routing key and an error message if an error occured.
+function _M.get_routing_key(self)
+  -- interface method to get the routing key from the cookie
+  -- has to be overridden by an affinity mode
+  ngx.log(ngx.ERR, "[BUG] Failed to get routing key as no implementation has been provided!")
+  return nil, nil
+end
 
-  for i = 1, MAX_UPSTREAM_CHECKS_COUNT do
-    local key = string.format("%s.%s.%s", ngx.now() + i, ngx.worker.pid(), math.random(999999))
-
-    local new_upstream = self.instance:find(key)
-
-    if not failed_upstreams[new_upstream] then
-      return new_upstream, key
-    end
-  end
+--- set_routing_key sets the current routing key on the cookie
+-- @tparam string key The routing key to set on the cookie.
+function _M.set_routing_key(self, key)
+  -- interface method to set the routing key on the cookie
+  -- has to be overridden by an affinity mode
+  ngx.log(ngx.ERR, "[BUG] Failed to set routing key as no implementation has been provided!")
+end
 
+--- pick_new_upstream picks a new upstream while ignoring the given failed upstreams.
+-- @tparam {[string]=boolean} A table of upstreams to ignore where the key is the endpoint and the value a boolean.
+-- @treturn string, string The endpoint and its key.
+function _M.pick_new_upstream(self, failed_upstreams)
+  -- interface method to get a new upstream
+  -- has to be overridden by an affinity mode
+  ngx.log(ngx.ERR, "[BUG] Failed to pick new upstream as no implementation has been provided!")
   return nil, nil
 end
 
@@ -128,15 +165,9 @@ local function should_set_cookie(self)
 end
 
 function _M.balance(self)
-  local cookie, err = ck:new()
-  if not cookie then
-    ngx.log(ngx.ERR, "error while initializing cookie: " .. tostring(err))
-    return
-  end
-
   local upstream_from_cookie
 
-  local key = cookie:get(self:cookie_name())
+  local key = self:get_routing_key()
   if key then
     upstream_from_cookie = self.instance:find(key)
   end
@@ -151,24 +182,34 @@ function _M.balance(self)
 
   local new_upstream
 
-  new_upstream, key = pick_new_upstream(self)
+  new_upstream, key = self:pick_new_upstream(get_failed_upstreams())
   if not new_upstream then
     ngx.log(ngx.WARN, string.format("failed to get new upstream; using upstream %s", new_upstream))
   elseif should_set_cookie(self) then
-    set_cookie(self, key)
+    self:set_routing_key(key)
   end
 
   return new_upstream
 end
 
 function _M.sync(self, backend)
+  local changed = false
+
+  -- check and reinit affinity mode before syncing the balancer which will reinit the nodes
+  if self.affinity_mode ~= backend.sessionAffinityConfig.mode then
+    changed = true
+    init_affinity_mode(self, backend)
+  end
+
+  -- reload balancer nodes
   balancer_resty.sync(self, backend)
 
   -- Reload the balancer if any of the annotations have changed.
-  local changed = not util.deep_compare(
+  changed = changed or not util.deep_compare(
     self.cookie_session_affinity,
     backend.sessionAffinityConfig.cookieSessionAffinity
   )
+
   if not changed then
     return
   end

rootfs/etc/nginx/lua/test/balancer/sticky_test.lua

@@ -15,11 +15,16 @@ local function reset_ngx()
 end
 
 function get_mocked_cookie_new()
+  local o = { value = nil }
+  local mock = {
+    get = function(self, n) return self.value end,
+    set = function(self, c) self.value = c.value ; return true, nil end
+  }
+  setmetatable(o, mock)
+  mock.__index = mock
+
   return function(self)
-    return {
-      get = function(self, n) return nil, "error" end,
-      set = function(self, n) return true, "" end
-    }
+    return o;
   end
 end
 
@@ -229,7 +234,7 @@ describe("Sticky", function()
     end)
   end)
 
-  local function get_several_test_backends(change_on_failure)
+  local function get_several_test_backends(option)
     return {
       name = "access-router-production-web-80",
       endpoints = {
@@ -238,7 +243,13 @@ describe("Sticky", function()
       },
       sessionAffinityConfig = {
         name = "cookie",
-        cookieSessionAffinity = { name = "test_name", hash = "sha1", change_on_failure = change_on_failure }
+        mode = option["mode"],
+        cookieSessionAffinity = {
+          name = "test_name",
+          hash = "sha1",
+          change_on_failure = option["change_on_failure"],
+          locations = { ['test.com'] = {'/'} }
+        }
       },
     }
   end
@@ -257,21 +268,20 @@ describe("Sticky", function()
 
     context("when request to upstream fails", function()
       it("changes upstream when change_on_failure option is true", function()
-        -- create sticky cookie
-        cookie.new = function(self)
-          local return_obj = {
-            set = function(v) return false, nil end,
-            get = function(k) return "" end,
-          }
-          return return_obj, false
-        end
-
-        local options = {false, true}
+        local options = {
+          {["change_on_failure"] = false, ["mode"] = nil},
+          {["change_on_failure"] = false, ["mode"] = 'balanced'},
+          {["change_on_failure"] = false, ["mode"] = 'persistent'},
+          {["change_on_failure"] = true, ["mode"] = nil},
+          {["change_on_failure"] = true, ["mode"] = 'balanced'},
+          {["change_on_failure"] = true, ["mode"] = 'persistent'}
+        }
 
         for _, option in ipairs(options) do
           local sticky_balancer_instance = sticky:new(get_several_test_backends(option))
 
           local old_upstream = sticky_balancer_instance:balance()
+          assert.is.Not.Nil(old_upstream)
           for _ = 1, 100 do
             -- make sure upstream doesn't change on subsequent calls of balance()
             assert.equal(old_upstream, sticky_balancer_instance:balance())
@@ -281,11 +291,11 @@ describe("Sticky", function()
           sticky_balancer_instance.get_last_failure = function()
             return "failed"
           end
-          _G.ngx.var = { upstream_addr = old_upstream }
+          _G.ngx.var.upstream_addr = old_upstream
 
           for _ = 1, 100 do
             local new_upstream = sticky_balancer_instance:balance()
-            if option == false then
+            if option["change_on_failure"] == false then
               -- upstream should be the same inspite of error, if change_on_failure option is false
               assert.equal(new_upstream, old_upstream)
             else

rootfs/etc/nginx/lua/test/util/nodemap_test.lua

@@ -0,0 +1,167 @@
+local util = require("util")
+local nodemap = require("util.nodemap")
+
+local function get_test_backend_single()
+  return {
+    name = "access-router-production-web-80",
+    endpoints = {
+      { address = "10.184.7.40", port = "8080", maxFails = 0, failTimeout = 0 }
+    }
+  }
+end
+
+local function get_test_backend_multi()
+  return {
+    name = "access-router-production-web-80",
+    endpoints = {
+      { address = "10.184.7.40", port = "8080", maxFails = 0, failTimeout = 0 },
+      { address = "10.184.7.41", port = "8080", maxFails = 0, failTimeout = 0 }
+    }
+  }
+end
+
+local function get_test_nodes_ignore(endpoint)
+  local ignore = {}
+  ignore[endpoint] = true
+  return ignore
+end
+
+describe("Node Map", function()
+
+  local test_backend_single = get_test_backend_single()
+  local test_backend_multi = get_test_backend_multi()
+  local test_salt = test_backend_single.name
+  local test_nodes_single = util.get_nodes(test_backend_single.endpoints)
+  local test_nodes_multi = util.get_nodes(test_backend_multi.endpoints)
+  local test_endpoint1 = test_backend_multi.endpoints[1].address .. ":" .. test_backend_multi.endpoints[1].port
+  local test_endpoint2 = test_backend_multi.endpoints[2].address .. ":" .. test_backend_multi.endpoints[2].port
+  local test_nodes_ignore = get_test_nodes_ignore(test_endpoint1)
+
+  describe("new()", function()
+    context("when no salt has been provided", function()
+      it("random() returns an unsalted key", function()
+        local nodemap_instance = nodemap:new(test_nodes_single, nil)
+        local expected_endpoint = test_endpoint1
+        local expected_hash_key = ngx.md5(expected_endpoint)
+        local actual_endpoint
+        local actual_hash_key
+
+        actual_endpoint, actual_hash_key = nodemap_instance:random()
+
+        assert.equal(actual_endpoint, expected_endpoint)
+        assert.equal(expected_hash_key, actual_hash_key)
+      end)
+    end)
+
+    context("when a salt has been provided", function()
+      it("random() returns a salted key", function()
+        local nodemap_instance = nodemap:new(test_nodes_single, test_salt)
+        local expected_endpoint = test_endpoint1
+        local expected_hash_key = ngx.md5(test_salt .. expected_endpoint)
+        local actual_endpoint
+        local actual_hash_key
+
+        actual_endpoint, actual_hash_key = nodemap_instance:random()
+
+        assert.equal(actual_endpoint, expected_endpoint)
+        assert.equal(expected_hash_key, actual_hash_key)
+      end)
+    end)
+
+    context("when no nodes have been provided", function()
+      it("random() returns nil", function()
+        local nodemap_instance = nodemap:new({}, test_salt)
+        local actual_endpoint
+        local actual_hash_key
+
+        actual_endpoint, actual_hash_key = nodemap_instance:random()
+
+        assert.equal(actual_endpoint, nil)
+        assert.equal(expected_hash_key, nil)
+      end)
+    end)
+  end)
+
+  describe("find()", function()
+    before_each(function()
+      package.loaded["util.nodemap"] = nil
+      nodemap = require("util.nodemap")
+    end)
+
+    context("when a hash key is valid", function()
+      it("find() returns the correct endpoint", function()
+        local nodemap_instance = nodemap:new(test_nodes_single, test_salt)
+        local test_hash_key
+        local expected_endpoint
+        local actual_endpoint
+
+        expected_endpoint, test_hash_key = nodemap_instance:random()
+        assert.not_equal(expected_endpoint, nil)
+        assert.not_equal(test_hash_key, nil)
+
+        actual_endpoint = nodemap_instance:find(test_hash_key)
+        assert.equal(actual_endpoint, expected_endpoint)
+      end)
+    end)
+
+    context("when a hash key is invalid", function()
+      it("find() returns nil", function()
+        local nodemap_instance = nodemap:new(test_nodes_single, test_salt)
+        local test_hash_key = "invalid or nonexistent hash key"
+        local actual_endpoint
+
+        actual_endpoint = nodemap_instance:find(test_hash_key)
+
+        assert.equal(actual_endpoint, nil)
+      end)
+    end)
+  end)
+
+
+  describe("random_except()", function()
+    before_each(function()
+      package.loaded["util.nodemap"] = nil
+      nodemap = require("util.nodemap")
+    end)
+
+    context("when nothing has been excluded", function()
+      it("random_except() returns the correct endpoint", function()
+        local nodemap_instance = nodemap:new(test_nodes_single, test_salt)
+        local expected_endpoint = test_endpoint1
+        local test_hash_key
+        local actual_endpoint
+
+        actual_endpoint, test_hash_key = nodemap_instance:random_except({})
+        assert.equal(expected_endpoint, actual_endpoint)
+        assert.not_equal(test_hash_key, nil)
+      end)
+    end)
+
+    context("when everything has been excluded", function()
+      it("random_except() returns nil", function()
+        local nodemap_instance = nodemap:new(test_nodes_single, test_salt)
+        local actual_hash_key
+        local actual_endpoint
+
+        actual_endpoint, actual_hash_key = nodemap_instance:random_except(test_nodes_ignore)
+
+        assert.equal(actual_endpoint, nil)
+        assert.equal(actual_hash_key, nil)
+      end)
+    end)
+
+    context("when an endpoint has been excluded", function()
+      it("random_except() does not return it", function()
+        local nodemap_instance = nodemap:new(test_nodes_multi, test_salt)
+        local expected_endpoint = test_endpoint2
+        local actual_endpoint
+        local test_hash_key
+
+        actual_endpoint, test_hash_key = nodemap_instance:random_except(test_nodes_ignore)
+
+        assert.equal(actual_endpoint, expected_endpoint)
+        assert.not_equal(test_hash_key, nil)
+      end)
+    end)
+  end)
+end)

rootfs/etc/nginx/lua/util/nodemap.lua

@@ -0,0 +1,120 @@
+local math = require("math")
+local util = require("util")
+
+local _M = {}
+
+--- create_map generates the node hash table
+-- @tparam {[string]=number} nodes A table with the node as a key and its weight as a value.
+-- @tparam string salt A salt that will be used to generate salted hash keys.
+local function create_map(nodes, salt)
+  local hash_map = {}
+
+  for endpoint, _ in pairs(nodes) do
+    -- obfuscate the endpoint with a shared key to prevent brute force
+    -- and rainbow table attacks which could reveal internal endpoints
+    local key = salt .. endpoint
+    local hash_key = ngx.md5(key)
+    hash_map[hash_key] = endpoint
+  end
+
+  return hash_map
+end
+
+--- get_random_node picks a random node from the given map.
+-- @tparam {[string], ...} map A key to node hash table.
+-- @treturn string,string The node and its key
+local function get_random_node(map)
+  local size = util.tablelength(map)
+
+  if size < 1 then
+    return nil, nil
+  end
+
+  local index = math.random(1, size)
+  local count = 1
+
+  for key, endpoint in pairs(map) do
+      if count == index then
+        return endpoint, key
+      end
+
+      count = count + 1
+  end
+
+  ngx.log(ngx.ERR, string.format("Failed to find node %d of %d! This is a bug, please report!", index, size))
+
+  return nil, nil
+end
+
+--- new constructs a new instance of the node map
+--
+-- The map uses MD5 to create hash keys for a given node. For security reasons it supports
+-- salted hash keys, to prevent attackers from using rainbow tables or brute forcing
+-- the node endpoints, which would reveal cluster internal network information.
+--
+-- To make sure hash keys are reproducible on different ingress controller instances the salt
+-- needs to be shared and therefore is not simply generated randomly.
+--
+-- @tparam {[string]=number} endpoints A table with the node endpoint as a key and its weight as a value.
+-- @tparam[opt] string hash_salt A optional hash salt that will be used to obfuscate the hash key.
+function _M.new(self, endpoints, hash_salt)
+
+  if hash_salt == nil then
+    hash_salt = ''
+  end
+
+  -- the endpoints have to be saved as 'nodes' to keep compatibility to balancer.resty
+  local o = {
+    salt = hash_salt,
+    nodes = endpoints,
+    map = create_map(endpoints, hash_salt)
+  }
+
+  setmetatable(o, self)
+  self.__index = self
+  return o
+end
+
+--- reinit reinitializes the node map reusing the original salt
+-- @tparam {[string]=number} nodes A table with the node as a key and its weight as a value.
+function _M.reinit(self, nodes)
+  self.nodes = nodes
+  self.map = create_map(nodes, self.salt)
+end
+
+--- find looks up a node by hash key.
+-- @tparam string key The hash key.
+-- @treturn string The node.
+function _M.find(self, key)
+  return self.map[key]
+end
+
+--- random picks a random node from the hashmap.
+-- @treturn string,string A random node and its key or both nil.
+function _M.random(self)
+  return get_random_node(self.map)
+end
+
+--- random_except picks a random node from the hashmap, ignoring the nodes in the given table
+-- @tparam {string, } ignore_nodes A table of nodes to ignore, the node needs to be the key,
+--                                 the value needs to be set to true
+-- @treturn string,string A random node and its key or both nil.
+function _M.random_except(self, ignore_nodes)
+  local valid_nodes = {}
+
+  -- avoid generating the map if no ignores where provided
+  if ignore_nodes == nil or util.tablelength(ignore_nodes) == 0 then
+    return get_random_node(self.map)
+  end
+
+  -- generate valid endpoints
+  for key, endpoint in pairs(self.map) do
+      if not ignore_nodes[endpoint] then
+        valid_nodes[key] = endpoint
+      end
+  end
+
+  return get_random_node(valid_nodes)
+end
+
+return _M