Closed colmaengus closed 8 years ago
If you are using the OVA you will get the following error when connect to system / indices
and when you click on "update the configuration" this happens - nothing is loaded (no error give)
The Logfile looks like the following when you open system / indices
My Setup was updated from 2.0.3 (OVA) via Omnibus to 2.1.Alpha.2 (only for the records)
@jalogisch: I'm not sure how your comment is related. My query wasn't regarding an error but just wondering why polling was being used instead of websockets for notification. In our setup we are using graylog to monitor a kubernetes cluster which includes the reverse proxy fronting graylog. So a lot of the logs we are seeing are from graylog polling.
@colmaengus We used websockets at some point but feedback overwhelmingly was that it breaks in many of our users environments due to heavy use of proxies which simply do not support them.
Ideally I'd like us to use long polling, too, but from a systems point of view this should not make a difference.
Did you observe a performance problem or is this simply a question?
@colmaengus - wrong end from me but let the comment be present for reference
@kroepke: no particular performance issue (apart from filling up logs). it would be nice though to have an option to enable websocket as was done in your previous front end arch.
On a side note we've just started using graylog and its really nice to use. We use elasticsearch/kibana for some of our products and its nice to see an opinionated use of elasticsearch for logs compared to what your get with vanilla ELK.
Glad you like it :)
Regarding websockets, I don't think we are likely to look at it again any time soon. Previously we used sockjs which was supposed to take care of the incompatibilities, but in the end it didn't reliably for our users. Most ended up disabling it so we didn't use it when implementing 2.0.
If you feel strongly about it we are happy to look at pull requests to make it an option but quite honestly I feel it's mostly a waste of effort (even though the reloading annoys me greatly during development 😉 )
we use socket.io and haven't had issues with it yet but its early days in the product so maybe we've not hit enough proxy issues.
Closing this because we are not going to implement this in the near future. Websockets and proxy setups are still very hard to get right.
Can the time of the polling or every second or so be reduced to a more fine tuned time?
Problem description
XHR request to system/cluster/nodes, system/notifications, cluster/metrics/multiple are occurring every second or so. I would have assumed that these kind of notifications would be done via websocket.
Steps to reproduce the problem
Environment