[jira] [Commented] (SPARK-34198) Add RocksDB StateStore as external module
[ https://issues.apache.org/jira/browse/SPARK-34198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17329276#comment-17329276 ] Yuanjian Li commented on SPARK-34198: - [~viirya] [~kabhwan] Make sense, I should bring it to the community to confirm and list the reasons. I thought it's not a decision but implementation details. > Add RocksDB StateStore as external module > - > > Key: SPARK-34198 > URL: https://issues.apache.org/jira/browse/SPARK-34198 > Project: Spark > Issue Type: New Feature > Components: Structured Streaming >Affects Versions: 3.2.0 >Reporter: L. C. Hsieh >Priority: Major > > Currently Spark SS only has one built-in StateStore implementation > HDFSBackedStateStore. Actually it uses in-memory map to store state rows. As > there are more and more streaming applications, some of them requires to use > large state in stateful operations such as streaming aggregation and join. > Several other major streaming frameworks already use RocksDB for state > management. So it is proven to be good choice for large state usage. But > Spark SS still lacks of a built-in state store for the requirement. > We would like to explore the possibility to add RocksDB-based StateStore into > Spark SS. For the concern about adding RocksDB as a direct dependency, our > plan is to add this StateStore as an external module first. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-34198) Add RocksDB StateStore as external module
[ https://issues.apache.org/jira/browse/SPARK-34198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17326716#comment-17326716 ] L. C. Hsieh commented on SPARK-34198: - The major issue is the additional rocksdb dependency. For me, I'm not against it. But maybe others have strong preferences not to include it by default. I agree with [~kabhwan] that we may need to get a consensus from the community. > Add RocksDB StateStore as external module > - > > Key: SPARK-34198 > URL: https://issues.apache.org/jira/browse/SPARK-34198 > Project: Spark > Issue Type: New Feature > Components: Structured Streaming >Affects Versions: 3.2.0 >Reporter: L. C. Hsieh >Priority: Major > > Currently Spark SS only has one built-in StateStore implementation > HDFSBackedStateStore. Actually it uses in-memory map to store state rows. As > there are more and more streaming applications, some of them requires to use > large state in stateful operations such as streaming aggregation and join. > Several other major streaming frameworks already use RocksDB for state > management. So it is proven to be good choice for large state usage. But > Spark SS still lacks of a built-in state store for the requirement. > We would like to explore the possibility to add RocksDB-based StateStore into > Spark SS. For the concern about adding RocksDB as a direct dependency, our > plan is to add this StateStore as an external module first. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-34198) Add RocksDB StateStore as external module
[ https://issues.apache.org/jira/browse/SPARK-34198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17326460#comment-17326460 ] Jungtaek Lim commented on SPARK-34198: -- Please note that the decision was taken from community discussion. If you need to change it, please bring it to the community with rationalization. > Add RocksDB StateStore as external module > - > > Key: SPARK-34198 > URL: https://issues.apache.org/jira/browse/SPARK-34198 > Project: Spark > Issue Type: New Feature > Components: Structured Streaming >Affects Versions: 3.2.0 >Reporter: L. C. Hsieh >Priority: Major > > Currently Spark SS only has one built-in StateStore implementation > HDFSBackedStateStore. Actually it uses in-memory map to store state rows. As > there are more and more streaming applications, some of them requires to use > large state in stateful operations such as streaming aggregation and join. > Several other major streaming frameworks already use RocksDB for state > management. So it is proven to be good choice for large state usage. But > Spark SS still lacks of a built-in state store for the requirement. > We would like to explore the possibility to add RocksDB-based StateStore into > Spark SS. For the concern about adding RocksDB as a direct dependency, our > plan is to add this StateStore as an external module first. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-34198) Add RocksDB StateStore as external module
[ https://issues.apache.org/jira/browse/SPARK-34198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17326427#comment-17326427 ] Yuanjian Li commented on SPARK-34198: - [~viirya] Since the RocksDBStateStore can solve the major drawbacks for the current HDFS based one. I think it's a better choice to directly add it as a build-in RocksDBStateStoreProvider. It's also convenient for the end-users to choose it directly. If you agree, I will change the description and the title for this ticket. > Add RocksDB StateStore as external module > - > > Key: SPARK-34198 > URL: https://issues.apache.org/jira/browse/SPARK-34198 > Project: Spark > Issue Type: New Feature > Components: Structured Streaming >Affects Versions: 3.2.0 >Reporter: L. C. Hsieh >Priority: Major > > Currently Spark SS only has one built-in StateStore implementation > HDFSBackedStateStore. Actually it uses in-memory map to store state rows. As > there are more and more streaming applications, some of them requires to use > large state in stateful operations such as streaming aggregation and join. > Several other major streaming frameworks already use RocksDB for state > management. So it is proven to be good choice for large state usage. But > Spark SS still lacks of a built-in state store for the requirement. > We would like to explore the possibility to add RocksDB-based StateStore into > Spark SS. For the concern about adding RocksDB as a direct dependency, our > plan is to add this StateStore as an external module first. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-34198) Add RocksDB StateStore as external module
[ https://issues.apache.org/jira/browse/SPARK-34198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17325162#comment-17325162 ] L. C. Hsieh commented on SPARK-34198: - Thanks [~XuanYuan] and Databricks for the decision. Looking forward to the implementation! > Add RocksDB StateStore as external module > - > > Key: SPARK-34198 > URL: https://issues.apache.org/jira/browse/SPARK-34198 > Project: Spark > Issue Type: New Feature > Components: Structured Streaming >Affects Versions: 3.2.0 >Reporter: L. C. Hsieh >Priority: Major > > Currently Spark SS only has one built-in StateStore implementation > HDFSBackedStateStore. Actually it uses in-memory map to store state rows. As > there are more and more streaming applications, some of them requires to use > large state in stateful operations such as streaming aggregation and join. > Several other major streaming frameworks already use RocksDB for state > management. So it is proven to be good choice for large state usage. But > Spark SS still lacks of a built-in state store for the requirement. > We would like to explore the possibility to add RocksDB-based StateStore into > Spark SS. For the concern about adding RocksDB as a direct dependency, our > plan is to add this StateStore as an external module first. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-34198) Add RocksDB StateStore as external module
[ https://issues.apache.org/jira/browse/SPARK-34198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17325031#comment-17325031 ] Yuanjian Li commented on SPARK-34198: - Thanks all for joining the discussion! Databricks decided to donate the commercial implementation of the RocksDBStateStore implementation. I just linked the design doc for this. The code work will start this week. Please help with reviewing the design and PRs. Hope we can have this great feature in the 3.2 release! > Add RocksDB StateStore as external module > - > > Key: SPARK-34198 > URL: https://issues.apache.org/jira/browse/SPARK-34198 > Project: Spark > Issue Type: New Feature > Components: Structured Streaming >Affects Versions: 3.2.0 >Reporter: L. C. Hsieh >Priority: Major > > Currently Spark SS only has one built-in StateStore implementation > HDFSBackedStateStore. Actually it uses in-memory map to store state rows. As > there are more and more streaming applications, some of them requires to use > large state in stateful operations such as streaming aggregation and join. > Several other major streaming frameworks already use RocksDB for state > management. So it is proven to be good choice for large state usage. But > Spark SS still lacks of a built-in state store for the requirement. > We would like to explore the possibility to add RocksDB-based StateStore into > Spark SS. For the concern about adding RocksDB as a direct dependency, our > plan is to add this StateStore as an external module first. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-34198) Add RocksDB StateStore as external module
[
https://issues.apache.org/jira/browse/SPARK-34198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17290771#comment-17290771
]
L. C. Hsieh commented on SPARK-34198:
-
FYI, I ran a benchmark against two open source implementations. The Chermenin's
implementation is consistently slower than Qubole's about 30%.
{code}
[info] Running benchmark: Put key value pairs
[info] Running case: Chermenin
[info] Stopped after 3 iterations, 103521 ms
[info] Running case: Qubole
[info] Stopped after 3 iterations, 76908 ms
[info] OpenJDK 64-Bit Server VM 1.8.0_265-b01 on Mac OS X 10.15.7
[info] Intel(R) Core(TM) i7-9750H CPU @ 2.60GHz
[info] Put key value pairs: Best Time(ms) Avg Time(ms)
Stdev(ms)Rate(M/s) Per Row(ns) Relative
[info]
[info] Chermenin 32725 34507
2030 0.0 49934.0 1.0X
[info] Qubole25493 25636
186 0.0 38899.0 1.3X
{code}
> Add RocksDB StateStore as external module
> -
>
> Key: SPARK-34198
> URL: https://issues.apache.org/jira/browse/SPARK-34198
> Project: Spark
> Issue Type: New Feature
> Components: Structured Streaming
>Affects Versions: 3.2.0
>Reporter: L. C. Hsieh
>Priority: Major
>
> Currently Spark SS only has one built-in StateStore implementation
> HDFSBackedStateStore. Actually it uses in-memory map to store state rows. As
> there are more and more streaming applications, some of them requires to use
> large state in stateful operations such as streaming aggregation and join.
> Several other major streaming frameworks already use RocksDB for state
> management. So it is proven to be good choice for large state usage. But
> Spark SS still lacks of a built-in state store for the requirement.
> We would like to explore the possibility to add RocksDB-based StateStore into
> Spark SS. For the concern about adding RocksDB as a direct dependency, our
> plan is to add this StateStore as an external module first.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
-
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-34198) Add RocksDB StateStore as external module
[ https://issues.apache.org/jira/browse/SPARK-34198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17284491#comment-17284491 ] Jungtaek Lim commented on SPARK-34198: -- Thanks for considering it. I think it would be the best option for Apache Spark among these if it makes sense to Databricks as well, just because it has been served for years with enterprise level of support. We can't expect the stability from other options and may struggle with it for some period - it'd be best if we can avoid it. (Worth noting that second one may also provide enterprise level of support, but less than an year, and I had 50+ of review comments on proposed PR and personally didn't feel the PR was super solid at that time. I mean, for me, the PR was not proposed with production quality at first.) > Add RocksDB StateStore as external module > - > > Key: SPARK-34198 > URL: https://issues.apache.org/jira/browse/SPARK-34198 > Project: Spark > Issue Type: New Feature > Components: Structured Streaming >Affects Versions: 3.2.0 >Reporter: L. C. Hsieh >Assignee: L. C. Hsieh >Priority: Major > > Currently Spark SS only has one built-in StateStore implementation > HDFSBackedStateStore. Actually it uses in-memory map to store state rows. As > there are more and more streaming applications, some of them requires to use > large state in stateful operations such as streaming aggregation and join. > Several other major streaming frameworks already use RocksDB for state > management. So it is proven to be good choice for large state usage. But > Spark SS still lacks of a built-in state store for the requirement. > We would like to explore the possibility to add RocksDB-based StateStore into > Spark SS. For the concern about adding RocksDB as a direct dependency, our > plan is to add this StateStore as an external module first. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-34198) Add RocksDB StateStore as external module
[ https://issues.apache.org/jira/browse/SPARK-34198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17284479#comment-17284479 ] Reynold Xin commented on SPARK-34198: - I don't know the intricate details of it but I suspect it's a different one because it existed long before those two. > Add RocksDB StateStore as external module > - > > Key: SPARK-34198 > URL: https://issues.apache.org/jira/browse/SPARK-34198 > Project: Spark > Issue Type: New Feature > Components: Structured Streaming >Affects Versions: 3.2.0 >Reporter: L. C. Hsieh >Assignee: L. C. Hsieh >Priority: Major > > Currently Spark SS only has one built-in StateStore implementation > HDFSBackedStateStore. Actually it uses in-memory map to store state rows. As > there are more and more streaming applications, some of them requires to use > large state in stateful operations such as streaming aggregation and join. > Several other major streaming frameworks already use RocksDB for state > management. So it is proven to be good choice for large state usage. But > Spark SS still lacks of a built-in state store for the requirement. > We would like to explore the possibility to add RocksDB-based StateStore into > Spark SS. For the concern about adding RocksDB as a direct dependency, our > plan is to add this StateStore as an external module first. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-34198) Add RocksDB StateStore as external module
[ https://issues.apache.org/jira/browse/SPARK-34198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17284478#comment-17284478 ] L. C. Hsieh commented on SPARK-34198: - Thanks [~rxin]. Is the implementation used in Databricks a completely different one than other two implementations? Or it is also based on any one of the two? > Add RocksDB StateStore as external module > - > > Key: SPARK-34198 > URL: https://issues.apache.org/jira/browse/SPARK-34198 > Project: Spark > Issue Type: New Feature > Components: Structured Streaming >Affects Versions: 3.2.0 >Reporter: L. C. Hsieh >Assignee: L. C. Hsieh >Priority: Major > > Currently Spark SS only has one built-in StateStore implementation > HDFSBackedStateStore. Actually it uses in-memory map to store state rows. As > there are more and more streaming applications, some of them requires to use > large state in stateful operations such as streaming aggregation and join. > Several other major streaming frameworks already use RocksDB for state > management. So it is proven to be good choice for large state usage. But > Spark SS still lacks of a built-in state store for the requirement. > We would like to explore the possibility to add RocksDB-based StateStore into > Spark SS. For the concern about adding RocksDB as a direct dependency, our > plan is to add this StateStore as an external module first. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-34198) Add RocksDB StateStore as external module
[ https://issues.apache.org/jira/browse/SPARK-34198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17284474#comment-17284474 ] Reynold Xin commented on SPARK-34198: - [~kabhwan] let me talk to the team that built our internal version of that on whether it'd make sense. > Add RocksDB StateStore as external module > - > > Key: SPARK-34198 > URL: https://issues.apache.org/jira/browse/SPARK-34198 > Project: Spark > Issue Type: New Feature > Components: Structured Streaming >Affects Versions: 3.2.0 >Reporter: L. C. Hsieh >Assignee: L. C. Hsieh >Priority: Major > > Currently Spark SS only has one built-in StateStore implementation > HDFSBackedStateStore. Actually it uses in-memory map to store state rows. As > there are more and more streaming applications, some of them requires to use > large state in stateful operations such as streaming aggregation and join. > Several other major streaming frameworks already use RocksDB for state > management. So it is proven to be good choice for large state usage. But > Spark SS still lacks of a built-in state store for the requirement. > We would like to explore the possibility to add RocksDB-based StateStore into > Spark SS. For the concern about adding RocksDB as a direct dependency, our > plan is to add this StateStore as an external module first. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-34198) Add RocksDB StateStore as external module
[ https://issues.apache.org/jira/browse/SPARK-34198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17284381#comment-17284381 ] L. C. Hsieh commented on SPARK-34198: - I'd tend to take as the baseline from [https://github.com/qubole/spark-state-store|https://github.com/qubole/spark-state-store,] as we are experimenting it internally and seems we are not only one using it based on previous comments, and yea I think it is basically from the previous PR [https://github.com/apache/spark/pull/24922]. It looks newer than the first one and has a better structure. > Add RocksDB StateStore as external module > - > > Key: SPARK-34198 > URL: https://issues.apache.org/jira/browse/SPARK-34198 > Project: Spark > Issue Type: New Feature > Components: Structured Streaming >Affects Versions: 3.2.0 >Reporter: L. C. Hsieh >Assignee: L. C. Hsieh >Priority: Major > > Currently Spark SS only has one built-in StateStore implementation > HDFSBackedStateStore. Actually it uses in-memory map to store state rows. As > there are more and more streaming applications, some of them requires to use > large state in stateful operations such as streaming aggregation and join. > Several other major streaming frameworks already use RocksDB for state > management. So it is proven to be good choice for large state usage. But > Spark SS still lacks of a built-in state store for the requirement. > We would like to explore the possibility to add RocksDB-based StateStore into > Spark SS. For the concern about adding RocksDB as a direct dependency, our > plan is to add this StateStore as an external module first. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-34198) Add RocksDB StateStore as external module
[ https://issues.apache.org/jira/browse/SPARK-34198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17284336#comment-17284336 ] Jungtaek Lim commented on SPARK-34198: -- Now we have consensus on having this, which is great. Thanks for raising the discussion and gathering consensus. Which implementation we should take as baseline remains the question. Two implementations what I'm aware of are: 1. https://github.com/chermenin/spark-states 2. https://github.com/apache/spark/pull/24922 Ideally it'd be nice to take the first one as it has been known to be the way to use RocksDB state store with SS for years. Probably we need to pursue the maintainer of the project, but in most cases they tend to be happy with donating the code (with simple PR, as it doesn't seem to be tied with specific employer) so worth trying out. Second case is probably easier to continue, as we can leverage existing PR without explicit approval from original author. What we should keep in mind is "retaining" main authorship, otherwise it wouldn't be a problem to take it over. Another crazy idea (but probably most stable among all implementations) is asking Databricks to donate the commercial implementation of RocksDB state store. I guess it should have been used by their customers for years, so proven to be stable relatively. [~rxin] Given you'd also give +1 on adding RocksDB state store to Spark codebase, any chance Databricks donate the existing implementation to Spark? Is it just me and would it be crazy idea asking this? > Add RocksDB StateStore as external module > - > > Key: SPARK-34198 > URL: https://issues.apache.org/jira/browse/SPARK-34198 > Project: Spark > Issue Type: New Feature > Components: Structured Streaming >Affects Versions: 3.2.0 >Reporter: L. C. Hsieh >Assignee: L. C. Hsieh >Priority: Major > > Currently Spark SS only has one built-in StateStore implementation > HDFSBackedStateStore. Actually it uses in-memory map to store state rows. As > there are more and more streaming applications, some of them requires to use > large state in stateful operations such as streaming aggregation and join. > Several other major streaming frameworks already use RocksDB for state > management. So it is proven to be good choice for large state usage. But > Spark SS still lacks of a built-in state store for the requirement. > We would like to explore the possibility to add RocksDB-based StateStore into > Spark SS. For the concern about adding RocksDB as a direct dependency, our > plan is to add this StateStore as an external module first. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-34198) Add RocksDB StateStore as external module
[ https://issues.apache.org/jira/browse/SPARK-34198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17276944#comment-17276944 ] L. C. Hsieh commented on SPARK-34198: - Raised the discussion: http://apache-spark-developers-list.1001551.n3.nabble.com/DISCUSS-Add-RocksDB-StateStore-td30645.html > Add RocksDB StateStore as external module > - > > Key: SPARK-34198 > URL: https://issues.apache.org/jira/browse/SPARK-34198 > Project: Spark > Issue Type: New Feature > Components: Structured Streaming >Affects Versions: 3.2.0 >Reporter: L. C. Hsieh >Assignee: L. C. Hsieh >Priority: Major > > Currently Spark SS only has one built-in StateStore implementation > HDFSBackedStateStore. Actually it uses in-memory map to store state rows. As > there are more and more streaming applications, some of them requires to use > large state in stateful operations such as streaming aggregation and join. > Several other major streaming frameworks already use RocksDB for state > management. So it is proven to be good choice for large state usage. But > Spark SS still lacks of a built-in state store for the requirement. > We would like to explore the possibility to add RocksDB-based StateStore into > Spark SS. For the concern about adding RocksDB as a direct dependency, our > plan is to add this StateStore as an external module first. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-34198) Add RocksDB StateStore as external module
[ https://issues.apache.org/jira/browse/SPARK-34198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17276893#comment-17276893 ] L. C. Hsieh commented on SPARK-34198: - Thanks [~kabhwan] for your point. Besides the maintenance cost of extra code, I remember one concern of adding it, is the rocksdb dependency. I think the concern is valid and so it actually does have some differences between putting in sql core module or as an external module. IIUC, that is why we have external modules. If raising a discussion in dev mailing list helps, I think I will do it. The RocksDB StateStore we are working with, is also based on the existing implementation with our bug fix. So I think the review cost should be as lower as possible even we submit the changed code. Of course if the original author can contribute the code, it will be great too. And sure, this depends on what the consensus we get eventually. > Add RocksDB StateStore as external module > - > > Key: SPARK-34198 > URL: https://issues.apache.org/jira/browse/SPARK-34198 > Project: Spark > Issue Type: New Feature > Components: Structured Streaming >Affects Versions: 3.2.0 >Reporter: L. C. Hsieh >Assignee: L. C. Hsieh >Priority: Major > > Currently Spark SS only has one built-in StateStore implementation > HDFSBackedStateStore. Actually it uses in-memory map to store state rows. As > there are more and more streaming applications, some of them requires to use > large state in stateful operations such as streaming aggregation and join. > Several other major streaming frameworks already use RocksDB for state > management. So it is proven to be good choice for large state usage. But > Spark SS still lacks of a built-in state store for the requirement. > We would like to explore the possibility to add RocksDB-based StateStore into > Spark SS. For the concern about adding RocksDB as a direct dependency, our > plan is to add this StateStore as an external module first. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-34198) Add RocksDB StateStore as external module
[ https://issues.apache.org/jira/browse/SPARK-34198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17276883#comment-17276883 ] Jungtaek Lim commented on SPARK-34198: -- The external modules means modules in external directory. Personally I don't think there's huge difference between adding it in spark-sql core module vs adding it via external module. The major point of this is whether we want to add the functionality to Spark codebase or not. As we already confirmed there're concerns on adding this in Spark codebase, unless you raise the discussion in dev@ mailing list and gather consensus, the effort can be easily wasted. Please make sure we don't have such case. And once we decide to add this, I'd rather say I'd like to see either we persuade repo owner to contribute well-known existing implementation (https://github.com/chermenin/spark-states) to ASF, or new PR based on #24922. I wouldn't like to review multiple PRs again and again for the same functionality. > Add RocksDB StateStore as external module > - > > Key: SPARK-34198 > URL: https://issues.apache.org/jira/browse/SPARK-34198 > Project: Spark > Issue Type: New Feature > Components: Structured Streaming >Affects Versions: 3.2.0 >Reporter: L. C. Hsieh >Assignee: L. C. Hsieh >Priority: Major > > Currently Spark SS only has one built-in StateStore implementation > HDFSBackedStateStore. Actually it uses in-memory map to store state rows. As > there are more and more streaming applications, some of them requires to use > large state in stateful operations such as streaming aggregation and join. > Several other major streaming frameworks already use RocksDB for state > management. So it is proven to be good choice for large state usage. But > Spark SS still lacks of a built-in state store for the requirement. > We would like to explore the possibility to add RocksDB-based StateStore into > Spark SS. For the concern about adding RocksDB as a direct dependency, our > plan is to add this StateStore as an external module first. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-34198) Add RocksDB StateStore as external module
[ https://issues.apache.org/jira/browse/SPARK-34198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17276882#comment-17276882 ] L. C. Hsieh commented on SPARK-34198: - For external module here, I mean to put the related code under external/ along with other external modules like avro, kafka-0-10-sql, etc. > Add RocksDB StateStore as external module > - > > Key: SPARK-34198 > URL: https://issues.apache.org/jira/browse/SPARK-34198 > Project: Spark > Issue Type: New Feature > Components: Structured Streaming >Affects Versions: 3.2.0 >Reporter: L. C. Hsieh >Assignee: L. C. Hsieh >Priority: Major > > Currently Spark SS only has one built-in StateStore implementation > HDFSBackedStateStore. Actually it uses in-memory map to store state rows. As > there are more and more streaming applications, some of them requires to use > large state in stateful operations such as streaming aggregation and join. > Several other major streaming frameworks already use RocksDB for state > management. So it is proven to be good choice for large state usage. But > Spark SS still lacks of a built-in state store for the requirement. > We would like to explore the possibility to add RocksDB-based StateStore into > Spark SS. For the concern about adding RocksDB as a direct dependency, our > plan is to add this StateStore as an external module first. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-34198) Add RocksDB StateStore as external module
[ https://issues.apache.org/jira/browse/SPARK-34198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17276875#comment-17276875 ] Cheng Su commented on SPARK-34198: -- Sorry I think my question is not very clear. I am aware of [https://github.com/apache/spark/pull/24922] and what was the concern back then. I think I am not sure what's external module here. Do you mind explaining a bit more of what's external module meaning here (or is there an existing example in current codebase I can refer to)? More context is we are working on RocksDB state store internally as well based on above PR. So would like to check if there's any good things here we need to watch out for backport, thanks. > Add RocksDB StateStore as external module > - > > Key: SPARK-34198 > URL: https://issues.apache.org/jira/browse/SPARK-34198 > Project: Spark > Issue Type: New Feature > Components: Structured Streaming >Affects Versions: 3.2.0 >Reporter: L. C. Hsieh >Assignee: L. C. Hsieh >Priority: Major > > Currently Spark SS only has one built-in StateStore implementation > HDFSBackedStateStore. Actually it uses in-memory map to store state rows. As > there are more and more streaming applications, some of them requires to use > large state in stateful operations such as streaming aggregation and join. > Several other major streaming frameworks already use RocksDB for state > management. So it is proven to be good choice for large state usage. But > Spark SS still lacks of a built-in state store for the requirement. > We would like to explore the possibility to add RocksDB-based StateStore into > Spark SS. For the concern about adding RocksDB as a direct dependency, our > plan is to add this StateStore as an external module first. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-34198) Add RocksDB StateStore as external module
[ https://issues.apache.org/jira/browse/SPARK-34198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17276863#comment-17276863 ] L. C. Hsieh commented on SPARK-34198: - If you are asking why adding it as an external module instead of directly into streaming codebase, one previous concern is that this introduces extra dependency of RocksDB. So to add it as external module, we hope to relieve the concern. We will add RocksDB StateStore code as an external module as the JIRA title describes. Spark SS already can use a config provider class to choose what StateStore provider to use. So I think there won't be too many tasks involved. > Add RocksDB StateStore as external module > - > > Key: SPARK-34198 > URL: https://issues.apache.org/jira/browse/SPARK-34198 > Project: Spark > Issue Type: New Feature > Components: Structured Streaming >Affects Versions: 3.2.0 >Reporter: L. C. Hsieh >Priority: Major > > Currently Spark SS only has one built-in StateStore implementation > HDFSBackedStateStore. Actually it uses in-memory map to store state rows. As > there are more and more streaming applications, some of them requires to use > large state in stateful operations such as streaming aggregation and join. > Several other major streaming frameworks already use RocksDB for state > management. So it is proven to be good choice for large state usage. But > Spark SS still lacks of a built-in state store for the requirement. > We would like to explore the possibility to add RocksDB-based StateStore into > Spark SS. For the concern about adding RocksDB as a direct dependency, our > plan is to add this StateStore as an external module first. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-34198) Add RocksDB StateStore as external module
[ https://issues.apache.org/jira/browse/SPARK-34198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17276857#comment-17276857 ] Cheng Su commented on SPARK-34198: -- [~viirya] - could you help elaborate what's the benefit of adding as an external module? Also do you mind sharing a list of potential things/sub-tasks need to be done to make it work? Thanks. > Add RocksDB StateStore as external module > - > > Key: SPARK-34198 > URL: https://issues.apache.org/jira/browse/SPARK-34198 > Project: Spark > Issue Type: New Feature > Components: Structured Streaming >Affects Versions: 3.2.0 >Reporter: L. C. Hsieh >Priority: Major > > Currently Spark SS only has one built-in StateStore implementation > HDFSBackedStateStore. Actually it uses in-memory map to store state rows. As > there are more and more streaming applications, some of them requires to use > large state in stateful operations such as streaming aggregation and join. > Several other major streaming frameworks already use RocksDB for state > management. So it is proven to be good choice for large state usage. But > Spark SS still lacks of a built-in state store for the requirement. > We would like to explore the possibility to add RocksDB-based StateStore into > Spark SS. For the concern about adding RocksDB as a direct dependency, our > plan is to add this StateStore as an external module first. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-34198) Add RocksDB StateStore as external module
[ https://issues.apache.org/jira/browse/SPARK-34198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17271308#comment-17271308 ] Gabor Somogyi commented on SPARK-34198: --- +1 on this, I've started to review it back in the days when a guy wanted to add it embedded. > Add RocksDB StateStore as external module > - > > Key: SPARK-34198 > URL: https://issues.apache.org/jira/browse/SPARK-34198 > Project: Spark > Issue Type: New Feature > Components: Structured Streaming >Affects Versions: 3.2.0 >Reporter: L. C. Hsieh >Priority: Major > > Currently Spark SS only has one built-in StateStore implementation > HDFSBackedStateStore. Actually it uses in-memory map to store state rows. As > there are more and more streaming applications, some of them requires to use > large state in stateful operations such as streaming aggregation and join. > Several other major streaming frameworks already use RocksDB for state > management. So it is proven to be good choice for large state usage. But > Spark SS still lacks of a built-in state store for the requirement. > We would like to explore the possibility to add RocksDB-based StateStore into > Spark SS. For the concern about adding RocksDB as a direct dependency, our > plan is to add this StateStore as an external module first. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-34198) Add RocksDB StateStore as external module
[ https://issues.apache.org/jira/browse/SPARK-34198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17269737#comment-17269737 ] L. C. Hsieh commented on SPARK-34198: - cc [~dbtsai][~dongjoon][~hyukjin.kwon] > Add RocksDB StateStore as external module > - > > Key: SPARK-34198 > URL: https://issues.apache.org/jira/browse/SPARK-34198 > Project: Spark > Issue Type: Bug > Components: Structured Streaming >Affects Versions: 3.2.0 >Reporter: L. C. Hsieh >Priority: Major > > Currently Spark SS only has one built-in StateStore implementation > HDFSBackedStateStore. Actually it uses in-memory map to store state rows. As > there are more and more streaming applications, some of them requires to use > large state in stateful operations such as streaming aggregation and join. > Several other major streaming frameworks already use RocksDB for state > management. So it is proven to be good choice for large state usage. But > Spark SS still lacks of a built-in state store for the requirement. > We would like to explore the possibility to add RocksDB-based StateStore into > Spark SS. For the concern about adding RocksDB as a direct dependency, our > plan is to add this StateStore as an external module first. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
