LakshSingla opened a new pull request, #15420:
URL: https://github.com/apache/druid/pull/15420

   ### Description
   
   When trying to merge the higher order group by results on broker 
(primarily), and perhaps on historicals (under special conditions when we push 
the subquery down to the historical), a deadlock can occur depending on the 
structure of the query.
   The primary cause of this deadlock is that currently, we are trying to 
acquire the merge buffer resources at two places - `GroupByQueryQueryToolchest` 
(which is the place where we merge the higher-level results from the runners), 
and at `GroupByMergingQueryRunnerV2` (to merge the results from the runners of 
the multiple segments). These two were incorrectly considered to be mutually 
exclusive till now:
   1. `GroupByQueryQueryToolchest` - Acquires the merge buffers on the brokers 
only, as the historicals get the query wit the subqueries and the grouping sets 
to null, therefore this doesn't acquire any merge buffers. However, this might 
not be true, because when we push down the subquery itself, the historicals 
will also acquire the merge buffers at this place.
   2. `GroupByMergingQueryRunnerV2` - Acquires merge buffers on the historicals 
only, as it primarily merges the results from the query runners that run on the 
segments. This is also incorrect because the broker can act as a data server 
itself when the data source is an inline data source, and we attempt to run 
nested group bys on that. 
   
   Therefore, we have conditions when the `GroupByQueryQueryToolchest` is 
holding some merge buffers to merge the results of the returned runners, 
however, the runner itself is `GroupByMergingQueryRunnerV2` (or a decorated 
runner on top of it), and it also requires merge buffers to merge the 
"segment-level" runners to provide to the mergeResults of the toolchest. As we 
donot acquire the resources in a single go, following situations happen:
   
   Total merge buffers on the broker: 2
   QueryA = QueryB = A query that needs 1 merge buffer to `merge results` (in 
toolchest), and 1 mergeBuffer for `GroupByMergingQueryRunnerV2` on the broker
   QueryA & QueryB are running simultaneously on the cluster
   
   | Time | QueryA actions | QueryB actions | Buffers in system
   |--------|--------|--------|--------|
   | 0 |  | | 2 |
   | 1 | toolchest acquires 1 buffer | | 1 |
   | 2 | | toolchest acquires 1 buffer | 0 |
   | 3 | mergingRunner acquires 1 buffer (blocked) | | 0 |
   | 4 | | mergingRunner acquires 1 buffer (blocked) | 0 |
   
   The queries could have passed in isolation, however now they are waiting on 
the other to release the single merge buffer they hold to proceed. 
   
   This PR prevents such a deadlock from happening by acquiring the merge 
buffers in a single place and passing it down to the runner that might need it. 
   
   #### Release note
   <!-- Give your best effort to summarize your changes in a couple of 
sentences aimed toward Druid users. 
   
   If your change doesn't have end user impact, you can skip this section.
   
   For tips about how to write a good release note, see [Release 
notes](https://github.com/apache/druid/blob/master/CONTRIBUTING.md#release-notes).
   
   -->
   
   
   <hr>
   
   ##### Key changed/added classes in this PR
    * `MyFoo`
    * `OurBar`
    * `TheirBaz`
   
   <hr>
   
   <!-- Check the items by putting "x" in the brackets for the done things. Not 
all of these items apply to every PR. Remove the items which are not done or 
not relevant to the PR. None of the items from the checklist below are strictly 
necessary, but it would be very helpful if you at least self-review the PR. -->
   
   This PR has:
   
   - [ ] been self-reviewed.
      - [ ] using the [concurrency 
checklist](https://github.com/apache/druid/blob/master/dev/code-review/concurrency.md)
 (Remove this item if the PR doesn't have any relation to concurrency.)
   - [ ] added documentation for new or modified features or behaviors.
   - [ ] a release note entry in the PR description.
   - [ ] added Javadocs for most classes and all non-trivial methods. Linked 
related entities via Javadoc links.
   - [ ] added or updated version, license, or notice information in 
[licenses.yaml](https://github.com/apache/druid/blob/master/dev/license.md)
   - [ ] added comments explaining the "why" and the intent of the code 
wherever would not be obvious for an unfamiliar reader.
   - [ ] added unit tests or modified existing tests to cover new code paths, 
ensuring the threshold for [code 
coverage](https://github.com/apache/druid/blob/master/dev/code-review/code-coverage.md)
 is met.
   - [ ] added integration tests.
   - [ ] been tested in a test Druid cluster.
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to