ruby-on-rails - LogEntries 记录的 Heroku 路由器错误 H13

Question

我试图弄清楚我经常从 LogEntries 中得到的这些错误是否值得担心。我有一个与 heroku 上的 Rails API 对话的 iPhone 应用程序。我使用 HireFire 自动增加和减少应用程序所需的 dynos 和 worker 的数量。

我认为这些错误是由测功机按比例缩小然后调用该测功机并因此导致 H13 错误引起的。如果是这样，有没有办法阻止这种情况？用户请求肯定不会进入不再存在的dyno吗？

以下是我在上下文中获得的最常见的日志:

1.

2012-10-18 10:01:15.170 209 <13>1 2012-10-18T10:01:14+00:00 app web.1 - - Started GET "/hirefireapp/xxxxxxxxxxxxxxxx/info" for 23.22.73.241 at Thu Oct 18 03:01:14 -0700 2012
2012-10-18 10:01:15.170 162 <13>1 2012-10-18T10:01:14+00:00 app web.1 - - cache: [GET /hirefireapp/xxxxxxxxxxxxxxxx/info] miss
2012-10-18 10:01:22.147 146 <13>1 2012-10-18T10:01:22+00:00 app web.2 - - Connected to NewRelic Service at collector-2.newrelic.com:80
2012-10-18 10:01:22.308 217 <13>1 2012-10-18T10:01:22+00:00 app web.2 - - ** [NewRelic][10/18/12 03:01:22 -0700 xxxxxxxxxxxxxxxx (2)] INFO : Reporting performance data every 60 seconds.
2012-10-18 10:01:38.118 86 <13>1 2012-10-18T10:01:38+00:00 app web.2 - -
2012-10-18 10:01:38.212 86 <13>1 2012-10-18T10:01:38+00:00 app web.2 - -
2012-10-18 10:01:38.212 154 <13>1 2012-10-18T10:01:38+00:00 app web.2 - - Started GET "/" for 204.93.223.151 at Thu Oct 18 03:01:38 -0700 2012
2012-10-18 10:01:38.212 128 <13>1 2012-10-18T10:01:38+00:00 app web.2 - -   Processing by CmsController#index as */*
2012-10-18 10:01:38.212 148 <13>1 2012-10-18T10:01:38+00:00 app web.2 - - Rendered cms/index.html.erb within layouts/application (4.8ms)
2012-10-18 10:01:41.291 124 <40>1 2012-10-18T10:01:41+00:00 heroku web.2 - - Stopping all processes with SIGTERM
2012-10-18 10:01:42.243 217 <158>1 2012-10-18T10:01:42+00:00 heroku router - - Error H13 (Connection closed without response) -> GET my-api.heroku.com/ dyno=web.2 queue= wait= service= status=503 bytes=

2.

2012-10-16 17:31:25.184 189 <13>1 2012-10-16T17:31:25+00:00 app web.2 - - ** [NewRelic][10/16/12 10:31:25 -0700 (2)] INFO : Dispatcher: thin
2012-10-16 17:31:25.184 197 <13>1 2012-10-16T17:31:25+00:00 app web.2 - - ** [NewRelic][10/16/12 10:31:25 -0700 (2)] INFO : Application: my-api
2012-10-16 17:31:25.184 220 <13>1 2012-10-16T17:31:25+00:00 app web.2 - - ** [NewRelic][10/16/12 10:31:25 -0700 (2)] INFO : New Relic Ruby Agent 3.3.1 Initialized: pid = 2
2012-10-16 17:31:25.585 265 <13>1 2012-10-16T17:31:25+00:00 app web.2 - - ** [NewRelic][10/16/12 10:31:25 -0700 (2)] INFO : NewRelic::Agent::Samplers::DelayedJobLockSampler sampler not available: No DJ worker present
2012-10-16 17:31:37.068 217 <13>1 2012-10-16T17:31:37+00:00 app web.2 - - ** [NewRelic][10/16/12 10:31:36 -0700 (2)] INFO : Reporting performance data every 60 seconds.
2012-10-16 17:31:39.786 86 <13>1 2012-10-16T17:31:39+00:00 app web.2 - -
2012-10-16 17:31:39.884 86 <13>1 2012-10-16T17:31:39+00:00 app web.2 - -
2012-10-16 17:31:39.884 154 <13>1 2012-10-16T17:31:39+00:00 app web.2 - - Started GET "/" for xxx.xx.xxx.xxx at Tue Oct 16 10:31:39 -0700 2012
2012-10-16 17:31:39.884 128 <13>1 2012-10-16T17:31:39+00:00 app web.2 - -   Processing by CmsController#index as */*
2012-10-16 17:31:39.981 149 <13>1 2012-10-16T17:31:39+00:00 app web.2 - - Rendered cms/index.html.erb within layouts/application (18.1ms)
2012-10-16 17:31:46.082 217 <158>1 2012-10-16T17:31:46+00:00 heroku router - - Error H13 (Connection closed without response) -> GET my-api.heroku.com/ dyno=web.2 queue= wait= service= status=503 bytes=

任何建议将不胜感激!

谢谢
皮特

Answer 1

H13 问题的一个潜在原因是机架键空间。 Rack 通过使用 HTTP 请求发送非常大的参数 key 来增加对潜在恶意攻击的保护。

Rack Utils 将 key 空间限制为 65536，如果您发送带有大参数集的 HTTP 请求，该限制可能不够大。

您可以通过将以下内容添加到 Rails 初始值设定项文件来更改此限制:

if Rack::Utils.respond_to?("key_space_limit=")
  Rack::Utils.key_space_limit = 262144
end

我发现这个数字(4 倍标准)足够大，可以为我解决 H13 问题，但您可能需要根据自己的个人要求测试不同的设置。

这可能不是您的特定 H13 问题的原因，但作为可能的解决方案值得检查。