[
https://issues.apache.org/jira/browse/LENS-1243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15401780#comment-15401780
]
Rajat Khandelwal commented on LENS-1243:
----------------------------------------
Alright, so the design I'm moving to is the following:
QueryContext is an Observable in terms of its DriverStatus. It can have a bunch
of observers observing it.
Driver state update happens in two ways:
* A driver can call this:
{noformat}
context.getDriverStatus().setState(...).
context.getDriverStatus().setProgress(...)
context.getDriverStatus().setStatusMessage(...)
{noformat}
Which has no side-effects and sets the state quietly.
* A driver can call this: context.setDriverStatus(DriverQueryState.SUCCESSFUL),
which is basically asking the context to update its status. This will fit into
the observer pattern and the context will notify all the observers.
Slightly confusing, but it makes sense when you think about the observer
pattern. When you fetch driver status and modify the state variable there, it
has no side-effects, as the status fetched wasn't an observable. On the other
hand, when you ask the context to update itself to some status, you end up
invoking the Observers.
Now, the drivers which only support PULL method of status update, only use the
first method of status update. The drivers that use only PUSH can use only the
second method of status update. For this purpose, Query Service Impl has an
object that is registered as an Observer to queries running on a PUSH based
driver. This object receives callbacks of the status update from the driver and
updates the query's status accordingly.
At this point, we can do away with the first status update method and always
ask the query context to update its status. But that isn't possible since to be
safe, we'd want to implement a PUSH+PULL based status update for the drivers
that use PUSH updates since PULL is more reliable and already works. This makes
the presence of the two kinds of status update methods required, otherwise, the
code will revolve in an infinite loop for drivers using PUSH+PULL. So to escape
that, the PUSH updates will use the second method of status update and the PULL
updates will use the first method of status update.
>From the coding side, the code is complete till this point and all test cases
>passing. Now, For additional cleanup, I'd like to take the already existing
>interface QueryCompletionListener, and merge it with the new interface
>QueryDriverStatusUpdateListener, because the former is theoretically a less
>powerful version of the latter. I'm currently trying to achieve this. It was
>getting a bit confusing in my mind, so just thought I'd write down the entire
>thought process somewhere. Comments are welcome. Will be posting more thoughts
>when I hit a wall.
> Support Asynchronous status updates from drivers
> ------------------------------------------------
>
> Key: LENS-1243
> URL: https://issues.apache.org/jira/browse/LENS-1243
> Project: Apache Lens
> Issue Type: Improvement
> Reporter: Rajat Khandelwal
> Assignee: Rajat Khandelwal
>
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
