Thanks for all the clarifications and details Jerry, Jungtaek :-) This looks like an exciting improvement to Structured Streaming - looking forward to it becoming part of Apache Spark !
Regards, Mridul On Mon, Nov 28, 2022 at 8:40 PM Jerry Peng <jerry.boyang.p...@gmail.com> wrote: > Hi all, > > I will add my two cents. Improving the Microbatch execution engine does > not prevent us from working/improving on the continuous execution engine in > the future. These are orthogonal issues. This new mode I am proposing in > the microbatch execution engine intends to lower latency of this execution > engine that most people use today. We can view it as an incremental > improvement on the existing engine. I see the continuous execution engine > as a partially completed re-write of spark streaming and may serve as the > "future" engine powering Spark Streaming. Improving the "current" engine > does not mean we cannot work on a "future" engine. These two are not > mutually exclusive. I would like to focus the discussion on the merits of > this feature in regards to the current micro-batch execution engine and not > a discussion on the future of continuous execution engine. > > Best, > > Jerry > > > On Wed, Nov 23, 2022 at 3:17 AM Jungtaek Lim <kabhwan.opensou...@gmail.com> > wrote: > >> Hi Mridul, >> >> I'd like to make clear to avoid any misunderstanding - the decision was >> not led by me. (I'm just a one of engineers in the team. Not even TL.) As >> you see the direction, there was an internal consensus to not revisit the >> continuous mode. There are various reasons, which I think we know already. >> You seem to remember I have raised concerns about continuous mode, but have >> you indicated that it was even over 2 years ago? I still see no traction >> around the project. The main reason I abandoned the discussion was due to >> promising effort on integrating push based shuffle into continuous mode to >> achieve shuffle, but no effort has been made so far. >> >> The goal of this SPIP is to have an alternative approach dealing with >> same workload, given that we no longer have confidence of success of >> continuous mode. But I also want to make clear that deprecating and >> eventually retiring continuous mode is not a goal of this project. If that >> happens eventually, that would be a side-effect. Someone may have concerns >> that we have two different projects aiming for similar thing, but I'd >> rather see both projects having competition. If anyone willing to improve >> continuous mode can start making the effort right now. This SPIP does not >> block it. >> >> >> On Wed, Nov 23, 2022 at 5:29 PM Mridul Muralidharan <mri...@gmail.com> >> wrote: >> >>> >>> Hi Jungtaek, >>> >>> Given the goal of the SPIP is reducing latency for stateless apps, and >>> should reasonably fit continuous mode design goals, it feels odd to not >>> support it fin the proposal. >>> >>> I know you have raised concerns about continuous mode in past as well in >>> dev@ list, and we are further ignoring it in this proposal (and >>> possibly other enhancements in past few releases). >>> >>> Do you want to revisit the discussion to support it and propose a vote >>> on that ? And move it to deprecated ? >>> >>> I am much more comfortable not supporting this SPIP for CM if it was >>> deprecated. >>> >>> Thoughts ? >>> >>> Regards, >>> Mridul >>> >>> >>> >>> >>> On Wed, Nov 23, 2022 at 1:16 AM Jerry Peng <jerry.boyang.p...@gmail.com> >>> wrote: >>> >>>> Jungtaek, >>>> >>>> Thanks for taking up the role to shepard this SPIP! Thank you for also >>>> chiming in on your thoughts concerning the continuous mode! >>>> >>>> Best, >>>> >>>> Jerry >>>> >>>> On Tue, Nov 22, 2022 at 5:57 PM Jungtaek Lim < >>>> kabhwan.opensou...@gmail.com> wrote: >>>> >>>>> Just FYI, I'm shepherding this SPIP project. >>>>> >>>>> I think the major meta question would be, "why don't we spend >>>>> effort on continuous mode rather than initiating another feature aiming >>>>> for >>>>> the same workload?". Jerry already updated the doc to answer the question, >>>>> but I can also share my thoughts about it. >>>>> >>>>> I feel like the current "continuous mode" is a niche solution. (It's >>>>> not to blame. If you have to deal with such workload but can't rewrite the >>>>> underlying engine from scratch, then there are really few options.) >>>>> Since the implementation went with a workaround to implement which the >>>>> architecture does not support natively e.g. distributed snapshot, it gets >>>>> quite tricky on maintaining and expanding the project. It also requires >>>>> 3rd >>>>> parties to implement a separate source and sink implementation, which I'm >>>>> not sure how many 3rd parties actually followed so far. >>>>> >>>>> Eventually, "continuous mode" becomes an area no one in the active >>>>> community knows the details and has willingness to maintain. I wouldn't >>>>> say >>>>> we are confident to remove the tag on "experimental", although the feature >>>>> has been shipped for years. It was introduced in Spark 2.3, surprising >>>>> enough? >>>>> >>>>> We went back and thought about the approach from scratch. Jerry came >>>>> up with the idea which leverages existing microbatch execution, hence >>>>> relatively stable and no need to require 3rd parties to support another >>>>> mode. It adds complexity against microbatch execution but it's a lot less >>>>> complicated compared to the existing continuous mode. Definitely quite >>>>> less >>>>> than creating a new record-to-record engine from scratch. >>>>> >>>>> That said, we want to propose and move forward with the new approach. >>>>> >>>>> ps. Eventually we could probably discuss retiring continuous mode if >>>>> the new approach gets accepted and eventually considered as a stable one >>>>> after several minor releases. That's just me. >>>>> >>>>> On Wed, Nov 23, 2022 at 5:16 AM Jerry Peng < >>>>> jerry.boyang.p...@gmail.com> wrote: >>>>> >>>>>> Hi all, >>>>>> >>>>>> I would like to start the discussion for a SPIP, Asynchronous Offset >>>>>> Management in Structured Streaming. The high level summary of the SPIP >>>>>> is >>>>>> that currently in Structured Streaming we perform a couple of offset >>>>>> management operations for progress tracking purposes synchronously on the >>>>>> critical path which can contribute significantly to processing latency. >>>>>> If >>>>>> we were to make these operations asynchronous and less frequent we can >>>>>> dramatically improve latency for certain types of workloads. >>>>>> >>>>>> I have put together a SPIP to implement such a mechanism. Please >>>>>> take a look! >>>>>> >>>>>> SPIP Jira: https://issues.apache.org/jira/browse/SPARK-39591 >>>>>> >>>>>> SPIP doc: >>>>>> https://docs.google.com/document/d/1iPiI4YoGCM0i61pBjkxcggU57gHKf2jVwD7HWMHgH-Y/edit?usp=sharing >>>>>> >>>>>> >>>>>> Best, >>>>>> >>>>>> Jerry >>>>>> >>>>>
