Sunny Bains @TiDB

@Santy75 Well, this is my first experience and the email says otherwise.

@sunbains If it was paid by CC there is no chace they will refund to savings account. The refund will be processed to original modem of payment. I have used international CC and have received refund for cancelled ticket.

#airindia refund process is a borderline scam. Flight cancelled, they don't refund the charge back to your foreign CC only to a domestic savings account. As an NRI it's illegal to have a domestic savings account, their answer:

@lefred 😁

@sunbains Cancelled flight ?? Does that mean no new patch ??

@rene_cannao That’s why you need TiDB, you don’t have to worry about such dichotomies.

Read/write split needs humility . I am skeptical of "just send SELECTs to replicas." Sometimes it works. Sometimes it creates subtle correctness problems. `SELECT` does not always mean "safe for a replica." `SELECT ... FOR UPDATE` belongs on the writer. Reads inside transactions may need the writer. Read-after-write paths may need immediate consistency. Some workloads cannot tolerate replica lag. Some sessions depend on state that should not be moved casually. So the production question is not "can we split reads and writes?" The question is "which reads are we confident enough to move away from the writer?" My preferred policy is conservative: - writer by default - known-safe reads to replicas - explicit exceptions - measurement after every change ProxySQL is useful because hostgroups and query rules let this policy live in the traffic layer instead of being copied into every service. Automatic read/write split sounds attractive. Explicit read/write policy is usually safer.

@Santy75 The post explicitly states it was paid with a foreign CC.

@sunbains Did you pay with CC or bank account. If CC they will refund to the CC but if paid using NRE/NRO accounts the RBI guidelines does not allow money earned in India to be deposited into NRE account. You post is confusing, AI do issue charge back to foreign CC.

@airindia Dear #airindia , I have and I keep getting the same stock answer that I have posted above. I posted to X express my frustration with your process. Repeating the same line is not helping.

@sunbains Dear Sir, we understand that the booking belongs to Air India Express. Hence, we request reaching out to the respective carrier as they shall assist better in this regard.

@SatanAtWink Interesting. Why is that the case?

@sunbains Only drawback would be that you might not be able to use that exact card with airindia in the future.

@SatanAtWink I realize that but it’s just stupid.

@sunbains That’s okay. You can present your bank their exact email saying they can’t refund and that they have agreed the flight is not operational, & also that they refused to refund via the instrument paid. You will easily win this. No questions asked.

@SatanAtWink I did tell them that I will dispute the transaction. I should not have to do that.

@sunbains You can also do a chargeback? Isn’t that easier.

@SatanAtWink Hmmm … thanks, will try tomorrow. I have three days left. They give you 7 days to sort this out.

@sunbains You don’t have to tell them about type of account. Just give them the NRO account & say it’s saving. From a transfer pov locally - it’s the same. Most folks reverting to the email won’t know the nuance & club NRE with NRO.

@airindia Dear #airindia, I have and I posted the reply above. Who do I contact?

@sunbains Dear Sir, we request you to please connect with the Air India Express for assistance in this matter.

@dadua_daku @thegeeknarrator My last attempt was before the new models. It was on a c++ code base. It worked on very small pieces of code. Anything large was a complete mess. It compiled but that was about it.

@sunbains @thegeeknarrator How has your experience been with LLMs and legacy code? Multiple teams are working hard to make AI useful for many databases MSFT owns and it has been less than useful. It absolutely fails in producing even semi complicated features in Postgres code.

software engineering is DEAD. all you need to do is 👇 1) start a project. 2) give AI enough context, then ask it to brainstorm, challenge assumptions, and ask you a ton of questions. 3) keep refining until both of you actually understand the problem. 4) ask AI to create a living PRD. 5) review the PRD, cut the fluff, and finalize the scope. 6) define the APIs, key configurations, data models, edge cases, and testing strategy. add all of that to the PRD. 7) give the PRD to another AI agent and ask it to review, simplify, and find gaps. 8) once the PRD is solid, ask AI to create an implementation plan split into clean, incremental commits. 9) implement one step at a time: code, test, review, update the plan, repeat. 10) document key assumptions, trade-offs, and limitations as you go. 11) ask AI to generate scripts and test cases based on the testing strategy. 12) get approval, deploy, monitor carefully, and definitely do more than just cross your fingers. wait.. maybe it’s not?

@thegeeknarrator Too many Arc for my taste.

My experience is that it will create working code, mostly correct too. Great for bootstrapping but the performance in my experience is sub par. In my experiments the models seemed to be trained on code that have lots of queues/channels, unnecessary abstraction layering, lots of mutex/condvar , excessive use of async code where it’s not really necessary.

AI agents change the database business model. For years, SaaS apps were built around “one app, one database”. But agents are different. The emerging pattern is: one agent, one database. Why? Because each agent needs its own durable memory, context, tools, permissions, audit trail, and billing boundary. A shared database sounds cheaper at first, until you hit the real problems: - compliance gets messy - tenant isolation becomes fragile - per-agent billing is hard - debugging is painful - noisy neighbors appear everywhere - deleting one agent’s data cleanly is harder than it should be The clean abstraction is simple: Give every agent its own database. The hard part is cost. Can you imagine buying 1 million databases from Supabase, Aurora, or hosting 1 million MySQL/Postgres instances yourself? That model breaks immediately. This is where TiDB Cloud proves something important: “one agent, one database” is not just an architecture pattern. It is a new database business. With TiDB Cloud, we can make a database cheap enough to be provisioned per agent — around $2 or even cheaper per database per year. That changes the unit economics. Agents won’t share one giant memory forever. They will each get their own database. And the winning database platform will be the one that makes millions of tiny, isolated, compliant, billable databases feel boringly easy.

@AnirudhRowjee Sure, DM me.

@sunbains Hi Sunny! I Would love to catch up with you to discuss databases!

I’m going to be in Bangalore/Bengaluru this week, if you want to catch up for breakfast/coffee, discuss DBs, distributed systems etc. dm me.

@vaibhaw_vipul Would love to 🙂

@sunbains i would really like to catch up this time 🤞

@_mnpw Thanks :-)

@sunbains alas i am travelling out of Blr for next two weeks otherwise I would've loved to catch up with you. Hope you have a great trip!

@ckg404 Sure, I’ll be staying at the Marriott Whitefield, DM me and we can figure something out.

@sunbains Would love to catch up Sunny