strong +1 for adding a UI to Apex. This is actually something that users expect and are accustomed to with similar projects. Thanks for taking the initiative.
There are nuances to be discussed, but the overall plan LGTM. To your question, I think this should use the existing web port with a different path, this will make it easy to expose through the RM proxy. The question whether the port is static or dynamic is a different concern, there could be a separate feature that optionally allows the user to assign a static port. On YARN, there is already a static address, the RM proxy port. I would propose to start this feature on a branch. The requirements for build integration and source code structure should be fleshed out as you go. It is conceivable that the UI build, which requires separate tooling, runs as part of the release build, but not the usual CI. Thomas On Wed, Jun 6, 2018 at 7:43 AM, Chinmay Kolhatkar <chin...@apache.org> wrote: > Hi Community, > > I want to propose a feature in apex-core to have a UI component in stram. > This is influenced by how basic runtime information is shown on UI by > spark. > > This includes following features: > 1. Webpage will be hosted by stram and will be available for user to view > in one of the two ways (I'll prefer approach b): > a. Hosted on a static port in which case we'll need to add a new servlet > to stram > b. The webpage will be hosted on the same port as that of stram > webservice but at a different path. Say, http:// > <stram_host>:<webservice_port>/ui > > 2. The webpage will be a static page and depending on the framework we > chose, it can show the realtime metric data from stram. > > 3. There will be a categories of readonly information (maybe dynamically > changing) that will be shown on the UI: > a. Application level information > i. Status > ii. Number of tuples processed/emitted > iii. Start time/ Running span > iv. Stram events > v. Total memory used/available > vi. Number of container allocated/deployed > v. Other information which is available in stram > b. Logical Operator level information > i. Status > ii. Number of tuples processed/emitted > iii. Important events related to logical operator > iv. Container list in which the operator is deployed > v. Any other information available in stram > vi. Logical DAG View > c. Container level information > i. Status > ii. Number of tuples processed/emitted > iii. Important events related to logical operator > iv. Any other information available in stram > v. Physical DAG View > > 4. In second phase, there will be control related operations allowed from > the UI, as follows: > a. Stop/Kill application > b. Stop/Kill containers > c. Stack trace dump of containers > d. Logs related? > d. etc...?? > > The above implementation can be done in phases as follows: > 1. Have webpage display application level information only (read-only). > Static display first (i.e. page needs to refresh to see latest data), next > dynamically changing data. > 2. Have jenkins build tools updated as needed by UI framework. > 3. Update the backend (stram) to serve the UI pages > 3. Extend the webage to display logical operator information. > 4. Extend the webage to display physical operator information > 5. Implement control operations on UI. > > > To implement above functionality, I propose ReactJS as UI framework and > MaterialUI (based on Google's material design) for theme/controls/CSS > support. > > Please let me know what you think about the same. > > Also, please let know if anyone is interested in contributing to this > feature. > > Thanks, > Chinmay. >
