+1 (non-binding)
— Sent from my iPhone Pardon the dumb thumb typos :) On May 30, 2025, at 12:39 PM, Mark Hamstra <markhams...@gmail.com> wrote:
A soft real-time system still defines an interval or frame within which results should be available, and often provides explicit warning or error-handling mechanisms when frame rates are missed. I see nothing like that in the SPIP. Instead, the length of the underlying microbatches is specified in the Trigger, but result reporting is just as quickly as possible with no reporting interval or frame rate specified and nothing that I can see happening if results take longer than the user is guessing or expecting. That's a low-latency, "we'll do it as fast as we can, but no promises or guarantees" system, not real-time. Mark, For real-time systems there is a concept of "soft" real-time and "hard" real-time systems. These concepts exist in textbooks. Here is a document by intel that explains it:
https://www.intel.com/content/www/us/en/learn/what-is-a-real-time-system.html"In a soft real-time system, computers or equipment will continue to function after a missed deadline but may produce a lower-quality output. For example, latency in online video games can impact player interactions, but otherwise present no serious consequences." "Hard real-time systems have zero delay tolerance, and delayed signals can result in total failure or present immediate danger to users. Flight control systems and pacemakers are both examples where timeliness is not only essential but the lack of it can result in a life-or-death situation."
I don't think it is inaccurate or misleading to call this mode real-time. It is soft real-time.
Clarifying what is meant by "real-time" and explicitly differentiating it from actual real-time computing should be a bare minimum. I still don't like the use of marketing-speak "real-time" that isn't really real-time in engineering documents or API namespaces.
Mark,
I thought we are simply discussing the naming of the mode? Like I mentioned, if you think simply calling this mode "real-time" mode may cause confusion because "real-time" can mean other things in other fields, I can clarify what we mean by "real-time" explicitly in the SPIP document and any future documentation. That is not a problem and thank you for your feedback.
Referencing other misuse of "real-time" is not persuasive. A SPIP is an engineering document, not a marketing document. Technical clarity and accuracy should be non-negotiable.
Mark,
As an example of my point if you go the the Apache Storm (another stream processing engine) website:
It describes Storm as:
"Apache Storm is a free and open source distributed realtime computation system"
If you can to apache Flink:
"Apache Flink 2.0.0: A new Era of Real-Time Data Processing"
Thus, what the term "rea-time" implies in this should not be confusing for folks in this area.
Mich,
If I understood your last email correctly, I think you also wanted to have a discussion about naming? Why are we calling this new execution mode described in the SPIP "Real-time Mode"? Here are my two cents. Firstly, "continuous mode" is taken and we want another name to describe an execution mode that provides ultra low latency processing. We could have called it "low latency mode", though I don't really like that naming since it implies the other execution modes are not low latency which I don't believe is true. This new proposed mode can simply deliver even lower latency. Thus, we came up with the name "Real-time Mode". Of course, we are talking about "soft" real-time here. I think when we are talking about distributed stream processing systems in the space of big data analytics, it is reasonable to assume anything described in this space as "real-time" implies "soft" real-time. Though if this is confusing or misleading, we can provide clear documentation on what "real-time" in real-time mode means and what it guarantees. Just my thoughts. I would love to hear other perspectives.
I think from what I have seen there are a good number of +1 responses as opposed to quantitative discussions (based on my observations only). Given the objectives of the thread, we ought to focus on what is meant by real time compared to continuous modes.To be fair, it is a common point of confusion, and the terms are often used interchangeably in general conversation, but in technical contexts, especially with streaming data platforms, they have specific and important differences.
"Continuous Mode" refers to a processing strategy that aims for true, uninterrupted, sub-millisecond latency processing. Chiefly - Event-at-a-Time (or very small batch groups): The system processes individual events or extremely small groups of events -> microbatches as they flow through the pipeline.
- Minimal Latency: The primary goal is to achieve the absolute lowest possible end-to-end latency, often in the order of milliseconds or even below
- Most business use cases (say financial markets) can live with this as they do not rely on rdges
Now what is meant by "Real-time Mode" This is where the nuance comes in. "Real-time" is a broader and sometimes more subjective term. When the text introduces "Real-time Mode" as distinct from "Continuous Mode," it suggests a specific implementation that achieves real-time characteristics but might do so differently or more robustly than a "continuous" mode attempt. Going back to my earlier mention, in real time application , there is nothing as an answer which is supposed to be late and correct. The timeliness is part of the application. if I get the right answer too slowly it becomes useless or wrong. What I call the "Late and Correct is Useless" Principle In summary, "Real-time Mode" seems to describe an approach that delivers low-latency processing with high reliability and ease of use, leveraging established, battle-tested components.I invite the audience to have a discussion on this.
HTH
+1
On 2025/05/29 16:25:19 Xiao Li wrote:
> +1
>
> Yuming Wang <yumw...@apache.org> 于2025年5月29日周四 02:22写道：
>
> > +1.
> >
> > On Thu, May 29, 2025 at 3:36 PM DB Tsai <dbt...@dbtsai.com> wrote:
> >
> >> +1
> >> Sent from my iPhone
> >>
> >> On May 29, 2025, at 12:15 AM, John Zhuge <jzh...@apache.org> wrote:
> >>
> >>
> >> +1 Nice feature
> >>
> >> On Wed, May 28, 2025 at 9:53 PM Yuanjian Li <xyliyuanj...@gmail.com>
> >> wrote:
> >>
> >>> +1
> >>>
> >>> Kent Yao <y...@apache.org> 于2025年5月28日周三 19:31写道：
> >>>
> >>>> +1, LGTM.
> >>>>
> >>>> Kent
> >>>>
> >>>> 在 2025年5月29日星期四，Chao Sun <sunc...@apache.org> 写道：
> >>>>
> >>>>> +1. Super excited by this initiative!
> >>>>>
> >>>>> On Wed, May 28, 2025 at 1:54 PM Yanbo Liang <yblia...@gmail.com>
> >>>>> wrote:
> >>>>>
> >>>>>> +1
> >>>>>>
> >>>>>> On Wed, May 28, 2025 at 12:34 PM huaxin gao <huaxin.ga...@gmail.com>
> >>>>>> wrote:
> >>>>>>
> >>>>>>> +1
> >>>>>>> By unifying batch and low-latency streaming in Spark, we can
> >>>>>>> eliminate the need for separate streaming engines, reducing system
> >>>>>>> complexity and operational cost. Excited to see this direction!
> >>>>>>>
> >>>>>>> On Wed, May 28, 2025 at 9:08 AM Mich Talebzadeh <
> >>>>>>> mich.talebza...@gmail.com> wrote:
> >>>>>>>
> >>>>>>>> Hi,
> >>>>>>>>
> >>>>>>>> My point about "in real time application or data, there is nothing
> >>>>>>>> as an answer which is supposed to be late and correct. The timeliness is
> >>>>>>>> part of the application. if I get the right answer too slowly it becomes
> >>>>>>>> useless or wrong" is actually fundamental to *why* we need this
> >>>>>>>> Spark Structured Streaming proposal.
> >>>>>>>>
> >>>>>>>> The proposal is precisely about enabling Spark to power
> >>>>>>>> applications where, as I define it, the *timeliness* of the answer
> >>>>>>>> is as critical as its *correctness*. Spark's current streaming
> >>>>>>>> engine, primarily operating on micro-batches, often delivers results that
> >>>>>>>> are technically "correct" but arrive too late to be truly useful for
> >>>>>>>> certain high-stakes, real-time scenarios. This makes them "useless or
> >>>>>>>> wrong" in a practical, business-critical sense.
> >>>>>>>>
> >>>>>>>> For example *in real-time fraud detection* and In *high-frequency
> >>>>>>>> trading,* market data or trade execution commands must be
> >>>>>>>> delivered with minimal latency. Even a slight delay can mean missed
> >>>>>>>> opportunities or significant financial losses, making a "correct" price
> >>>>>>>> update useless if it's not instantaneous. able for these demanding
> >>>>>>>> use cases, where a "late but correct" answer is simply not good enough. As
> >>>>>>>> a colliery it is a fundamental concept, so it has to be treated as such not
> >>>>>>>> as a comment.in SPIP
> >>>>>>>>
> >>>>>>>> Hope this clarifies the connection in practical terms
> >>>>>>>> Dr Mich Talebzadeh,
> >>>>>>>> Architect | Data Science | Financial Crime | Forensic Analysis |
> >>>>>>>> GDPR
> >>>>>>>>
> >>>>>>>> view my Linkedin profile
> >>>>>>>> <https://www.linkedin.com/in/mich-talebzadeh-ph-d-5205b2/>
> >>>>>>>>
> >>>>>>>>
> >>>>>>>>
> >>>>>>>>
> >>>>>>>>
> >>>>>>>> On Wed, 28 May 2025 at 16:32, Denny Lee <denny.g....@gmail.com>
> >>>>>>>> wrote:
> >>>>>>>>
> >>>>>>>>> Hey Mich,
> >>>>>>>>>
> >>>>>>>>> Sorry, I may be missing something here but what does your
> >>>>>>>>> definition here have to do with the SPIP? Perhaps add comments directly
> >>>>>>>>> to the SPIP to provide context as the code snippet below is a direct copy
> >>>>>>>>> from the SPIP itself.
> >>>>>>>>>
> >>>>>>>>> Thanks,
> >>>>>>>>> Denny
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>> On Wed, May 28, 2025 at 06:48 Mich Talebzadeh <
> >>>>>>>>> mich.talebza...@gmail.com> wrote:
> >>>>>>>>>
> >>>>>>>>>> just to add
> >>>>>>>>>>
> >>>>>>>>>> A stronger definition of real time. The engineering definition of
> >>>>>>>>>> real time is roughly fast enough to be interactive
> >>>>>>>>>>
> >>>>>>>>>> However, I put a stronger definition. In real time application or
> >>>>>>>>>> data, there is nothing as an answer which is supposed to be late and
> >>>>>>>>>> correct. The timeliness is part of the application.if I get the right
> >>>>>>>>>> answer too slowly it becomes useless or wrong
> >>>>>>>>>>
> >>>>>>>>>>
> >>>>>>>>>>
> >>>>>>>>>> Dr Mich Talebzadeh,
> >>>>>>>>>> Architect | Data Science | Financial Crime | Forensic Analysis |
> >>>>>>>>>> GDPR
> >>>>>>>>>>
> >>>>>>>>>> view my Linkedin profile
> >>>>>>>>>> <https://www.linkedin.com/in/mich-talebzadeh-ph-d-5205b2/>
> >>>>>>>>>>
> >>>>>>>>>>
> >>>>>>>>>>
> >>>>>>>>>>
> >>>>>>>>>>
> >>>>>>>>>> On Wed, 28 May 2025 at 11:10, Mich Talebzadeh <
> >>>>>>>>>> mich.talebza...@gmail.com> wrote:
> >>>>>>>>>>
> >>>>>>>>>>> The current limitations in SSS come from micro-batching.If you
> >>>>>>>>>>> are going to reduce micro-batching, this reduction must be balanced against
> >>>>>>>>>>> the available processing capacity of the cluster to prevent back pressure
> >>>>>>>>>>> and instability. In the case of Continuous Processing mode, a
> >>>>>>>>>>> specific continuous trigger with a desired checkpoint interval quote
> >>>>>>>>>>>
> >>>>>>>>>>> "
> >>>>>>>>>>> df.writeStream
> >>>>>>>>>>> .format("...")
> >>>>>>>>>>> .option("...")
> >>>>>>>>>>> .trigger(Trigger.RealTime(“300 Seconds”)) // new trigger
> >>>>>>>>>>> type to enable real-time Mode
> >>>>>>>>>>> .start()
> >>>>>>>>>>> This Trigger.RealTime signals that the query should run in the
> >>>>>>>>>>> new ultra low-latency execution mode. A time interval can also be
> >>>>>>>>>>> specified, e.g. “300 Seconds”, to indicate how long each micro-batch should
> >>>>>>>>>>> run for.
> >>>>>>>>>>> "
> >>>>>>>>>>>
> >>>>>>>>>>> will inevitably depend on many factors. Not that simple
> >>>>>>>>>>> HTH
> >>>>>>>>>>>
> >>>>>>>>>>>
> >>>>>>>>>>> Dr Mich Talebzadeh,
> >>>>>>>>>>> Architect | Data Science | Financial Crime | Forensic Analysis |
> >>>>>>>>>>> GDPR
> >>>>>>>>>>>
> >>>>>>>>>>> view my Linkedin profile
> >>>>>>>>>>> <https://www.linkedin.com/in/mich-talebzadeh-ph-d-5205b2/>
> >>>>>>>>>>>
> >>>>>>>>>>>
> >>>>>>>>>>>
> >>>>>>>>>>>
> >>>>>>>>>>>
> >>>>>>>>>>> On Wed, 28 May 2025 at 05:13, Jerry Peng <
> >>>>>>>>>>> jerry.boyang.p...@gmail.com> wrote:
> >>>>>>>>>>>
> >>>>>>>>>>>> Hi all,
> >>>>>>>>>>>>
> >>>>>>>>>>>> I want to start a discussion thread for the SPIP titled
> >>>>>>>>>>>> “Real-Time Mode in Apache Spark Structured Streaming” that I've been
> >>>>>>>>>>>> working on with Siying Dong, Indrajit Roy, Chao Sun, Jungtaek Lim, and
> >>>>>>>>>>>> Michael Armbrust: [JIRA
> >>>>>>>>>>>> <https://issues.apache.org/jira/browse/SPARK-52330>] [Doc
> >>>>>>>>>>>> <https://docs.google.com/document/d/1CvJvtlTGP6TwQIT4kW6GFT1JbdziAYOBvt60ybb7Dw8/edit?usp=sharing>
> >>>>>>>>>>>> ].
> >>>>>>>>>>>>
> >>>>>>>>>>>> The SPIP proposes a new execution mode called “Real-time Mode”
> >>>>>>>>>>>> in Spark Structured Streaming that significantly lowers end-to-end latency
> >>>>>>>>>>>> for processing streams of data.
> >>>>>>>>>>>>
> >>>>>>>>>>>> A key principle of this proposal is compatibility. Our goal is
> >>>>>>>>>>>> to make Spark capable of handling streaming jobs that need results almost
> >>>>>>>>>>>> immediately (within O(100) milliseconds). We want to achieve this without
> >>>>>>>>>>>> changing the high-level DataFrame/Dataset API that users already use – so
> >>>>>>>>>>>> existing streaming queries can run in this new ultra-low-latency mode by
> >>>>>>>>>>>> simply turning it on, without rewriting their logic.
> >>>>>>>>>>>>
> >>>>>>>>>>>> In short, we’re trying to enable Spark to power real-time
> >>>>>>>>>>>> applications (like instant anomaly alerts or live personalization) that
> >>>>>>>>>>>> today cannot meet their latency requirements with Spark’s current streaming
> >>>>>>>>>>>> engine.
> >>>>>>>>>>>>
> >>>>>>>>>>>> We'd greatly appreciate your feedback, thoughts, and
> >>>>>>>>>>>> suggestions on this approach!
> >>>>>>>>>>>>
> >>>>>>>>>>>>
> >>>>>>
> >>>>>> --
> >>>>>> Best,
> >>>>>> Yanbo
> >>>>>>
> >>>>>
> >>
> >> --
> >> John Zhuge
> >>
> >>
>
---------------------------------------------------------------------
To unsubscribe e-mail: dev-unsubscr...@spark.apache.org
|
