Thanks for your responses everyone. I seem to be missing some emails on this discussion and in general from the dev mailing list. I did see Kris¹s feedback in the archives and found it very useful. I have tried to capture the requirements everyone mentioned in the JIRA I just filed.
https://issues.apache.org/jira/browse/KNOX-643 We can use the same JIRA to hash through the requirements and the design/implementation. Sumit. On 12/11/15, 11:59 AM, "Balaji Ganesan" <[email protected]> wrote: >Apache Ambari has recently developed a metrics framework. Why not expose >the Knox metrics through the Ambari framework ? > >https://cwiki.apache.org/confluence/display/AMBARI/Metrics > >On Fri, Dec 11, 2015 at 7:38 AM, larry mccay <[email protected]> >wrote: > >> I like this idea. >> >> In addition to dropwizard, I think we need to rule out leveraging what >>is >> used in hadoop common. >> I don't know how generic it is but do know that there is a JMXServlet >>that >> can be used to expose the metrics >> in a REST-like way. >> >> Leveraging hadoop common where possible is something that I like to >> consider more frequently and I think that >> we should have probably done that more from the start. >> >> I would definitely say to create the JIRA and continue the proposal >>there >> with high level design in attachments or wiki. >> >> On Fri, Dec 11, 2015 at 10:28 AM, sumit gupta <[email protected]> wrote: >> >> > Hi everyone, >> > >> > I wanted to start a discussion about adding the capability of >> > capturing metrics in Knox. The initial step could be something simple >> > like request processing metrics and exposing the numbers via JMX >> > and/or REST APIs. There is a bigger discussion around metering and >> > throttling that this could play into but maybe wise to avoid initially >> > (maybe...). >> > >> > One of the projects that I came across while looking into this is >> > https://dropwizard.github.io/metrics/3.1.0/. It seems to be popular >> > amongst other open source projects and has a lot of useful >> > functionality and seems quite stable (also has Apache 2.0 license). >> > >> > Another project that comes up in this space is http://prometheus.io/, >> > but I am strongly leaning towards dropwizard based on some initial >> > playing around that I¹m doing. >> > >> > Anyhow, implementation aside, it would be good to discuss and agree >> > upon requirements. Things like: >> > >> > 1. What metrics are we interested in? >> > 2. Who are the potential consumers of this data? Dashboards? >> > 3. How do we want to expose the metrics? >> > 4. Do we want to capture metrics at a service level (e.g. All requests >> > made for WebHDFS)? >> > 5. Configuration (how do we want to turn this functionality on or off >> > or something more granular) >> > >> > >> > Of course the basic question is whether we would like to consider this >> > functionality at all. If we do, I¹d be happy to file a JIRA and work >> > on it. >> > >> > Thoughts? >> > >> > Sumit >> > >>
