cburgmer / greenyet Goto Github PK

hosts.yaml vs. status_url.yaml

E.g.

{
  "first_component": {
    "color": "green"
  },
  "more_components": [
    {
      "name": "another_component",
      "status": true
    }
  ]
}

should be supported, and two components "first_component" and "another_component" should be listed.

We should show in the UI/and log:

• If a json path expression couldn't be applied
• A variable could not be extracted

Rack::Protection::JsonCsrf seems to kick in if a refer header is present, resulting in a 403 response. This affects links from greenyet to JSON status pages.

Possible solution: <meta name="referrer" content="no-referrer" /> from http://stackoverflow.com/questions/6817595/remove-http-referer#answer-32014225

• Make aggregation of machines of an environment easier to understand (green vs. error states)
• Trying to click a link to the status page is annoying when the patch expands on long entries
• Hover roll out is lost every 5 secs on refresh

Spring Boot will expose child health indicators keyed by name on the top level response (as hash, e.g. http://www.christianmenz.ch/programmieren/spring-boot-health-checks/).
While greenyet can technically pick up the children via JSONPath, the name is currently expected to be exposed inside each entry's value.

We should probably support his usecase in some form.

If hostname: is misspelled the server crashes on startup rather than stating what's wrong. Other keys will probably have the same issue.

I might be wrong, but we're getting warnings in the logs of green yet relating to Apache responseprocesscookies , and it says the cookie has an invalid expires attribute. It looks like this might be due to the http client... Same can be seen here:
gradle/gradle#1596
https://issues.apache.org/jira/browse/HTTPCLIENT-1077

I can't tell which version is in use from the project file though (looks like it doesn't even use Apache somehow) but you please help take a quick look?

Thanks

While a socket timeout is currently used, this is not effective at enforcing a final request timeout.

Currently slow responses mean the dashboard will show stale results. Also, the response time adds up to the polling duration (polling duration starts counting after the request finishes), so refresh will take longer than the polling config suggests.

Both keys are required

• Fall back to a default
• or complain?

Currently the internal thread pool size is estimate by Clojure's core.async at #processors * 2 + 42 (https://github.com/clojure/core.async/blob/master/src/main/clojure/clojure/core/async/impl/exec/threadpool.clj). We possibly want to limit this number so that the request burst is somewhat limited.

Currently the example Python server seems to complain at 6 parallel requests.

Found with SSLEngine params provided from the outside.

Hovering on a status from the second environment (= second row) will on refresh unfold the equivalent status in the first environment instead.

Spring Boot's built-in /health endpoint seems to return the status JSON response via a service unavailable status (http://docs.spring.io/spring-boot/docs/current/reference/html/production-ready-endpoints.html#_writing_custom_healthindicators).

So far greenyet will only honour a JSON response if delivered via status 200, for every other status a red color is shown and the payload is ignored for the purpose of inspecting the application's status.

We might want to support the behaviour of Spring Boot out of the box.

When a page loads through the LB it tries to load resources (ccc / iso) from the site root.

This behavior is not regulated in any way. Need an option to set base path for load or remove '/' from index.html.template

It looks like sticking a proxy in front of greenyet will cover for when the actual instance is down. That means the user is seeing stale data 👎