Comments for Brave New Geek https://bravenewgeek.com Frozen comment archive; commenting is closed. By: desain kreatif https://bravenewgeek.com/benchmarking-message-queue-latency/#comment-67394 Wed, 10 Jun 2026 11:41:49 +0000 https://bravenewgeek.com/benchmarking-message-queue-latency/#comment-67394 see
The cash register analogy is perfect for explaining service time vs response time. It’s alarming how many production benchmarks still fall into the coordinated omission trap by only tracking synchronous t1-t0 durations. Thanks for emphasizing HdrHistogram!

]]>
By: Heth Gala https://bravenewgeek.com/you-cannot-have-exactly-once-delivery/#comment-66177 Thu, 15 Jan 2026 02:47:51 +0000 https://bravenewgeek.com/you-cannot-have-exactly-once-delivery/#comment-66177 This article was just a rant. Nothing useful came out of it. I was expecting how to achieve at-least once in notification systems here but its total rant.

]]>
By: GarrettabokS https://bravenewgeek.com/go-is-unapologetically-flawed-heres-why-we-use-it/#comment-61925 Sat, 24 May 2025 12:26:56 +0000 https://bravenewgeek.com/go-is-unapologetically-flawed-heres-why-we-use-it/#comment-61925 I realize this is sudden but this grabbed my attention today, there is a bunch of folks who assume that artificial intelligence is the future and will transform the method we exist but truthfully I am yet to see any alterations in my day to day life that aren’t on my mobile.

Why would people welcome some robot into their home as well let alone off some strange user online.

What do you all think? Would you rely on a bot android from someone you don’t know?

I’ve heard tech giants like Zuck, Bezos, and Standard Unions are building a marketplace for this, although I don’t know if I would trust these businesses in my personal space.

]]>
By: Paddoxant https://bravenewgeek.com/dissecting-message-queues/#comment-61350 Wed, 19 Mar 2025 20:00:57 +0000 https://bravenewgeek.com/dissecting-message-queues/#comment-61350 Abe Bet casino https://fourgreatinventions.com/ Ücretsiz deneme oyunları, yeni başlayanlar için casino dünyasıyla tanışmanın harika bir yoludur.

]]>
By: Miakoto https://bravenewgeek.com/dissecting-message-queues/#comment-61348 Wed, 19 Mar 2025 19:23:14 +0000 https://bravenewgeek.com/dissecting-message-queues/#comment-61348 7slots giriş twitter Büyük jackpot ödülleriyle şansınızı deneyin.

]]>
By: Gyula Csom https://bravenewgeek.com/you-cannot-have-exactly-once-delivery/#comment-60691 Sun, 09 Feb 2025 12:40:07 +0000 https://bravenewgeek.com/you-cannot-have-exactly-once-delivery/#comment-60691 So to clarify… the above protocol is still not “Exactly once” at the network level. That is impossible due to FLP. However a smart protocol may hide this from the Application and present itself as “Exactly once”.

Bottom line: If network partitions (resending lost messages) is an issue than “Exactly once” can be missleading, someone shall not forget about the physical level. However if partitions are not a big issue (network is robust enough, fast, etc.) then someone may forget about the physical level (network) and just think that at the Application level (i.e. logically) it is “Exactly once”. Question here: what is the added value here in comparison with “Idempotency” which sometimes can still be an issue at the Application level as well, such as when an impatient user resends the message that cannot be (easily) handled by infrastructure. What are your experiences in this regard?

]]>
By: Gyula Csom https://bravenewgeek.com/you-cannot-have-exactly-once-delivery/#comment-60669 Sat, 08 Feb 2025 01:38:41 +0000 https://bravenewgeek.com/you-cannot-have-exactly-once-delivery/#comment-60669 I think the reasoning here is not totally precise. From FLP it follows that “Exactly once” is impossible in network (where partitions can occur). However it does not seem to be impossible at the application level (if we assume the machine reliable where partition cannot occur). How? Through idempotency:

— Draft protocol

Here is a simple protocol using shared message log. Lets say the message broker is Kafka, which BTW also states it can do “Exactly once”:

Step 1. Kafka server->client: Server sends, client Receives new message
Step 2. Kafka client: Logs message as being processed
Step 3. Kafka client->Application: Client sends, Application receives new message
Step 4. Application: Process message
Step 5. Application: Logs message as processed

— Case of network failure:

Step 2. Check log before forwarding and only forward message to application iff it is not yet processed.

— Case of process failure of Kafka client (after the message is already processed):

Restart Kafka client: Consult log and only forward messages which are not marked as processed

Case of process failure of Application:

Restart Application: Consult log and process messages which are not marked as processed

— Case of process failure of Kafka client (while message is being processed by the application):

No need nothing (single threaded Application) or some intra-app coordination (multi threaded)

— Case of process failure of both Kafka client and Application:

No need nothing.

A thing which seems to be interesting here, in case of business applications (DB): what happens with the transaction? Then idempotency should be solved at that side as well, such as recording a version (such as e.g. optimistic lock) both at the local log and the DB.

Then the only problem remains when Application crashes during transaction, after transaction started but but before commited. To my understanding client-crash-survival DB sessions might be theoretically possible, but I am not aware of any RDBMS implementing it. Someone else? :-)

As always I can be wrong – these are just rough thoughts. Any feedback is more then welcome. :-)

]]>
By: nerved-dev https://bravenewgeek.com/platform-engineering-as-a-service/#comment-59111 Fri, 15 Nov 2024 04:06:03 +0000 https://bravenewgeek.com/platform-engineering-as-a-service/#comment-59111 > which provide a rigid, one-size-fits-all platform that abstracts away the underlying infrastructure, PEaaS is designed to be flexible and tailored to your unique requirements.

Isn’t Platform Engineering related to how team are organized contrary to DevOps? We normally talk of “streamlined teams” contrary to any other topology.

And even if, isn’t having “rigid, one-size-fits-all” a synonym of “DevOps Monolith” that should be avoided at all cost, to quote Ken Mugrage.

To end, DevOps isn’t about tooling; in the way of “tools as [a unique] solution”; it has a human side too.

]]>
By: haha https://bravenewgeek.com/you-cannot-have-exactly-once-delivery/#comment-58976 Thu, 07 Nov 2024 06:10:14 +0000 https://bravenewgeek.com/you-cannot-have-exactly-once-delivery/#comment-58976 So true… I was shocked to see so many blogs claiming exactly-once is impossible, citing the two-general problem and the FLP paper.

Those impossibility resutls only say that algorithms guaranteeing liveness (i.e., terminate within bounded number of steps) do not exist; there can still be algorithms that guarantee safety (i.e., produce correct consensus outcomes once terminated). Paxos and 2PC are such algorithms and they work just fine in practice.

]]>
By: iwa2no https://bravenewgeek.com/building-a-distributed-log-from-scratch-part-1-storage-mechanics/#comment-57369 Thu, 29 Aug 2024 12:38:19 +0000 https://bravenewgeek.com/building-a-distributed-log-from-scratch-part-1-storage-mechanics/#comment-57369 Hi, I want to ask why access ordered log in large file is not succifient. AFAIK, if we want to find a log in 10^6 logs, we just need log2(10^6) which approximates 20 times, which I believe is a small number. Have I oversimplified anything ?

]]>
By: Drew https://bravenewgeek.com/cloud-without-kubernetes/#comment-55190 Fri, 19 Apr 2024 06:20:52 +0000 https://bravenewgeek.com/cloud-without-kubernetes/#comment-55190 Cloud Run isn’t based on Knative anymore, if it ever was.
Google has two versions of Cloud Run, one based on internal infra, while the other based on Knative, which is called Cloud Run for Anthos now. Cloud Run refers to the former. The semantics to deploy workload are the same though.

Just found out Cloud Run for Anthos has been renamed to Knative Serving(https://cloud.google.com/anthos/run/docs/architecture-overview)

]]>
By: abc123 https://bravenewgeek.com/you-cannot-have-exactly-once-delivery/#comment-52785 Mon, 18 Sep 2023 15:48:57 +0000 https://bravenewgeek.com/you-cannot-have-exactly-once-delivery/#comment-52785 By: Noor Fatima https://bravenewgeek.com/building-a-distributed-log-from-scratch-part-5-sketching-a-new-system/#comment-52293 Sun, 09 Jul 2023 18:49:17 +0000 https://bravenewgeek.com/building-a-distributed-log-from-scratch-part-5-sketching-a-new-system/#comment-52293 Kudos for keeping it crystal clear!”

]]>
By: MarK A. https://bravenewgeek.com/building-a-distributed-log-from-scratch-part-5-sketching-a-new-system/#comment-52137 Fri, 16 Jun 2023 22:09:44 +0000 https://bravenewgeek.com/building-a-distributed-log-from-scratch-part-5-sketching-a-new-system/#comment-52137 Well done i am read your article i am very impressive.

]]>
By: Dan Keenan https://bravenewgeek.com/api-authentication-with-gcp-identity-aware-proxy/#comment-51905 Thu, 04 May 2023 07:59:57 +0000 https://bravenewgeek.com/api-authentication-with-gcp-identity-aware-proxy/#comment-51905 Dear Valued Customer,

We are excited to announce our latest service offerings from Web 20 Ranker! As a leading digital marketing agency, we provide top-notch SEO solutions that are designed to help businesses achieve their online goals.

Our team of experts is dedicated to delivering high-quality results, and our services are tailored to meet the needs of your business. Whether you need help with local SEO, link building, or content marketing, we’ve got you covered.

At Web 20 Ranker, we pride ourselves on our commitment to providing excellent customer service. Our team is available to answer any questions you may have and provide you with ongoing support throughout your marketing campaign.

So if you’re looking for a reliable and effective digital marketing partner, look no further than Web 20 Ranker. Visit our website today =>https://tinyurl.com/web2ranker<= to learn more about our services and how we can help you achieve your online goals.

Thank you for your time and consideration. We look forward to hearing from you soon.

Best regards,
The Web 20 Ranker Team

]]>
By: Rody https://bravenewgeek.com/you-cannot-have-exactly-once-delivery/#comment-51813 Wed, 19 Apr 2023 03:10:42 +0000 https://bravenewgeek.com/you-cannot-have-exactly-once-delivery/#comment-51813 What about modern day chat applications, they use sockets and messages are delivered once and they come under distributed systems right?
isnt exactly once not achieved, or is it with the sender and receiver being the same services giving an edge?

]]>
By: Muhammad Adnan https://bravenewgeek.com/building-a-distributed-log-from-scratch-part-5-sketching-a-new-system/#comment-51519 Thu, 09 Mar 2023 09:12:55 +0000 https://bravenewgeek.com/building-a-distributed-log-from-scratch-part-5-sketching-a-new-system/#comment-51519 very nice.

]]>
By: Simon Boddy https://bravenewgeek.com/you-cannot-have-exactly-once-delivery/#comment-51460 Thu, 02 Mar 2023 09:07:04 +0000 https://bravenewgeek.com/you-cannot-have-exactly-once-delivery/#comment-51460 Absolutely. Of course. And there’s a simple pattern for dealing with this… use a request/response protocol rather than messages, then make sure all unsafe requests have an application-level id. The receiving process, the server application, should store all responses. If it sees a request for the first time it does the work, then sends and stores the response. If it sees a request for which it has a stored response, it just replays the response. Reliability is an application level responsibility, and uniquely identified requests can be linked to uniqueness in the application context (1 shopping cart can have 1 payment request that the shopping cart app can repeat endlessly until it gets a response)

]]>
By: Alexey Stogny https://bravenewgeek.com/you-cannot-have-exactly-once-delivery/#comment-51403 Thu, 23 Feb 2023 17:15:21 +0000 https://bravenewgeek.com/you-cannot-have-exactly-once-delivery/#comment-51403 Nice post! Thanks! Wouldn’t post useless comment, but there’s no other way to subscribe to new posts ;)

]]>
By: M.Tanvir https://bravenewgeek.com/building-a-distributed-log-from-scratch-part-5-sketching-a-new-system/#comment-51242 Thu, 26 Jan 2023 14:36:16 +0000 https://bravenewgeek.com/building-a-distributed-log-from-scratch-part-5-sketching-a-new-system/#comment-51242 Thanks for your information . I am read your Replication Protocol all points its very helpful I am very impressive

]]>