Description of problem: Browsing to the webconsole for pro-us-east-1 displays "An error occurred getting metrics". Looking in developer tools, the following error is displayed, "XMLHttpRequest cannot load https://metrics.pro-us-east-1.openshift.com/hawkular/metrics. No 'Access-Control-Allow-Origin' header is present on the requested resource. Origin 'https://console.pro-us-east-1.openshift.com' is therefore not allowed access." Version-Release number of selected component (if applicable): Hawkular: 0.27.4.Final-redhat-1 OpenShift Master: v3.6.173.0.21 (online version 3.5.1.76) Kubernetes Master: v1.6.1+5115d708d7 How reproducible: Most of the time. Oddly enough, I was able to get Metrics to work ONCE in incognito mode, but then a refresh caused them to disappear. Steps to Reproduce: 1. View a project in pro-us-east-1 2. See error 3. Open Chrome Dev Tools for the XML error Actual results: Errors Expected results: No errors Additional info:
I am seeing an error on the overview page: "An error occurred getting metrics. Open Metrics URL | Don't Show Me Again" And in the browser console it shows: "Failed to load https://metrics.pro-us-east-1.openshift.com/hawkular/metrics: No 'Access-Control-Allow-Origin' header is present on the requested resource. Origin 'https://console.pro-us-east-1.openshift.com' is therefore not allowed access." This only happens if I refresh the overview page with 'ctrl-shift-r', if I follow a link to the overview page I don't see this error. But either way I don't see metrics on the overview page as expected. If I open an incognito page, then I don't get any errors and I do see metrics on the overview page as expected. From the non-incognito page, when going through the network tab, I do see a call out to Hawkular Metrics and its returning a 200 result. But its also being loaded from disk cache. @spadgett: is there any reason why this is being loaded from cache and not freshly loaded each time? When I load through an incognito page, then it works without issue.
Matt, the /metrics page is not setting the right HTTP response headers to prevent caching. So we have a few options 1) Update /metrics to set the correct Cache-Control headers 2) Update the web console to hit a different endpoint to check if metrics are available 3) Add something unique to the URL like a random `preventCache` query parameter to prevent the browser from using the cached response. Let me know which approach you think works. If there is a better status endpoint, it might be better to switch to that.
The cache-control header should be set to private (curl -v https://metrics.pro-us-east-1.openshift.com/hawkular/metrics), so the browser can cache this in its own cache. What Cache-Control value should we be using here instead? I think I might be missing something here also, the cache result is returning a 200 response. What else should it be returning?
(In reply to Matt Wringe from comment #3) > The cache-control header should be set to private (curl -v > https://metrics.pro-us-east-1.openshift.com/hawkular/metrics), so the > browser can cache this in its own cache. Well, that's what's happening... I don't think it's what you want, though. The /metrics page shows current version and a status message ("Metrics Service: STARTED"). With that `Cache-Control` header and no `max-age`, you're telling the browser it can cache the page and not telling it to revalidate. It will continue to show that cached page even after Hawkular is upgraded or status changes. `Cache-Control: no-cache` will prevent the browser from caching. If you want to leave that page `Cache-Control: private`, we should probably switch the web console to use another endpoint for status. But I don't think `private` is best here anyway even if the web console wasn't using that URL. Let me know what you'd like to do.
The 200 response is misleading. The request is not event sent to the server since the browser has it in its cache.
(In reply to Samuel Padgett from comment #5) > The 200 response is misleading. The request is not event sent to the server > since the browser has it in its cache. Right, so if its not even going to the server, why is the console saying its in error?
Are the CORS headers missing? This would look like an error to the console. Not sure how CORS is handled with browser caching.
I wonder if you visited the /metrics page outside of the console, and the page got cached without the CORS headers. Then the requests from the console always fail because the cached version is missing `Access-Control-Allow-Origin`. This would also explain why it works in incognito mode since it wouldn't use the cached version.
@jtakori: Can you please take a look at this? It looks like we might not be using the correct caching headers here.
I don't see any Cache-Control header set to private in hawkular code or config, and when running the server locally (incognito mode or not), there's no such header set in response (trying on current release/0.27 branch, which doesn't differ from 0.27.4.Final in this aspect). So I'm wondering, could it be some intermediate layer or filter that manipulates the cache control / some networking configuration in openshift?
It's possible. But if Hawkular is not setting any Cache-Control header on that response, it means the page is cacheable and we have the same problem. So regardless, it should be setting something like `Cache-Control: no-cache`
PR opened in upstream hawkular-metrics
Actually, I am not totally convinced that hawkular server should ask to disable client caching on every path. It makes sense on path like "/metrics/status", but not on all pages, in particular not on "/metrics" which just serves static HTML (with xhr). To me the issue is more related to the combination CORS + browser caching. I'm trying to investigate more on that. But anyway the upstream PR is on its way if it's an acceptable workaround.
I wasn't expecting every page to disable caching. /metrics doesn't seem like a static page since it has a status message? If there is a real status endpoint, I would like to switch the console to use that.
It is actually static, but with javascript that fetches /metrics/status with xhr.
(In reply to Joel Takvorian from comment #13) > Actually, I am not totally convinced that hawkular server should ask to > disable client caching on every path. It makes sense on path like > "/metrics/status", but not on all pages, Other than /metrics, is there any other page you think we shouldn't disable caching? > in particular not on "/metrics" > which just serves static HTML (with xhr). Even static HTML can be updated
PR upstream: https://github.com/hawkular/hawkular-metrics/pull/887 Static HTML can be updated but I think it's fine to cache it and maybe set a max-age value. This is what I've done in the PR: "private, max-age=86400" for static HTML, and "no-cache" for all other. Note that I've found another problem that is probably what's originally caused the CORS issue: the CORS headers were not being set in response on "/metrics" endpoint that is serving HTML. It's also fixed in my patch.
[related JIRA: https://issues.jboss.org/browse/HWKMETRICS-736]
This should already be fixed in our 3.7 release.
Metrics route can be accessed and metrics diagrams are shown in web console. env: OpenShift Master:v3.6.173.0.21 (online version 3.6.0.45.1) Kubernetes Master:v1.6.1+5115d708d7 Images: metrics-cassandra:v3.6.173.0.7 metrics-hawkular-metrics:v3.6.173.0.7 metrics-heapster:v3.6.173.0.7