higeorge13

You need to run them in review loops, this is the only way to reduce or eliminate these issues.

It’s also funny because we seem to trust and select doctors by google reviews and assume they have a valid diploma and license to consult us for our lives, but we have to grill some poor software engineer previously working in faang with another 10 leetcode interviews to be ‘sure’ he’s a good fit. lol

I hope they won't chop gpt-4o-mini soon because it's fast and accurate for API usage.

Funny thing is that i had a negative experience helping someone who was laid off. I reached out, offered help, provided excellent reference for them, covid hit and hiring froze for that particular company, followed up a few times, ghosted, never talked again. It's you last phrase, that's the way the world and people works. It's people with their own troubles, insecurities and character. The most important bit is be yourself. If you are "built" to think and help others, keep doing it. If not, whatever.

I understand this for a dry candidate pool. I have been there as a hiring manager, you need some signal with take home or live interviewing. But on the rare opportunity that you find someone who is willing to talk in details about a similar project you want to do, you skip the pipeline and if he indeed built it, you hire him.

Why put the blame only on candidates? Interviewers are equally bad to interviewees. I have been to both sides of the table and can guarantee that 80% of interviewers would not be fit for my job or the process of hiring.

Companies keep ignoring the historical interview point. I have been to a few occasions when companies needed the exact thing i built in the past (e.g. migration to clickhouse), but chose to put me into a random take home related to a different technology (e.g. some bigquery assignment) and eventually reject me. Go off script and ask me details about the project which might solve your hands, why do i need to talk about something else?

I don’t trust vag for anything after the diesel scandal. I prefer any other brand than this.

That's a typical modern SaaS issue that all departments have to keep building just to appear busy and "meet goals". Product managers to find the new shiny features, designers to design them or redesign the whole app, engineers to build them or do some framework migration and so on.

I am not commenting on the interview effort-salary ratio, but the fact that credentials and experience mean nothing to the tech industry, also comparing to the rest of professions. I mean working in Google/Meta/Netflix, is like working on the best hospital if you are a doctor, in the best construction industry if you are an engineer, to best law firm if you are a lawyer, etc. Imagine having to pass a leetcode or iq test everytime you want to move to the next one. I definitely know that my cousin, who is an exceptional doctor in Greece with only 10 years of experience, laughs about it.

I trust they were really competent, but it's a bit depressing that these competent people will need to go through the leetcode rituals and 5-10 interviews to get a new job at Meta, Netflix, AWS or adjacent companies. That's actually the point of the original post; you are never judged by your (years of) experience or even your past companies, only by the results of a test from a random person/company.

Of course it's not right. Let's be honest, our profession is in the era where software engineer = factory worker, and the worst part is that we have been playing music chairs right for the last couple years. So yeah all these professions have some steady status/wealth/qol progression and upgrade while people gain years of experience, while in software development it doesn't matter how many years of experience we have, which companies we worked on, their sector, whether your company is using the saas we were working on, etc.; we are going to get judged by trivia questions and leetcode.

I checked pg_repack a while ago but some issues are a bit concerning to apply in production. Did you face any issues?

Let’s hope ai hacks mailboxes and google meet, and eventually replace c suite and managers as well. We might get more ‘reasoned’ and deterministic engineering roadmaps or financial strategies by claude ceo/cto/cfo/vp/director agents than current leaderships. lol

TBH the product team in whatsapp needs to find some way to be meaningful. What could justify being a head of product or product manager in a chat app? New emojis? Gif support? New backgrounds? No, let's just make whatsapp like instagram. She's going to get a promotion now or move on to some new business as chief product officer.

No offence to the product team, i know that this is how it works in tech. It's the same for engineering and design teams in every single b2b/b2c business. There is no concept of feature completeness anymore, every single service has to copy from others or be something instead of 10 other services.

Author here, happy to discuss any of the points mentioned in the blog post.

Do they have a status page?

Congratz to neon team (i like what they built), but i don’t see the value or relation to databricks. I hope neon will continue as a standalone product, otherwise we lose a solid postgres provider from the market.

This also matches my experience so far, but tbh top 10, group by, etc. queries are too easy. Llms produce terrible results on queries requiring multiple CTEs or complex tables. e.g. i still have not seen an llm able to calculate any stats on top of an intervals table.

Could you test with clickhouse-local? It always works better for me.

That’s an awesome change. Will that also work for limit offset queries?

I know, I guess my comment is driven by all the AI-resume generators.

Try a random resume and the first suggestion you get from them is "Add numbers". You add numbers and you still leave some bullet points without any, and the next suggestion is "Add missing numbers to these bullet points". So as a hiring manager I see resumes with numbers and my instant guess is that resume is generated by any of these services. :D

In general I feel the numbers are _more_ relevant if you (a) hold a senior/manager position and (b) had some actual business impact, e.g. "increased ARR from 1m to 100m as a revops manager", "grew the engineering team from 5 to 100 people as a head of engineering" and so on. Numbers like "made the api 10 times faster", "reduced ETL pipeline runtime from hours to minutes" feel almost arbitrary. You can justify them just fine, but still. But perhaps I am just biased by the AI-generated stuff, that's all.

It's extremely funny that this is accurate for souvlaki in Greece. :) On the contrary, I was amazed by the quality of food in stations in Japan. Some "shady" shops there had some of the best ramen/udon I had in Japan.

1.1.2. And use NUMBERS when describing these sections. I can't emphasize it enough. Quantify what you've done as much as possible.

I have a couple of those in my CV and they are easily justifiable, but I sometimes feel they are not valuable at all. And they also seem a bad side effect of the AI-generated resumes and AI-driven ATS filtering algorithms which almost demand you have bullets with numbers. Does anyone feel the same?

I guess you talk about hard failures or replication slot deletion, right? Debezium has a snapshot mode (1) to refetch everything from scratch, we have tested it but opted out of using it because it would put pressure on postgres with the queries on huge tables and on kafka and its storage. Imo we prefer to manually rebuild the clickhouse tables via postgres snapshots and latest cdc data, but there might be better options or ideas.

(1) https://debezium.io/documentation/reference/stable/connector...

I also agree with most of your comments and conclusions.

In our setup, we use app ingestion to send all the denormalised data into Clickhouse using async inserts and Debezium/Kafka/Kafka engine and materialized views to sync a few Postgres tables into Clickhouse. 2 of the replicated tables are in the order of billions of rows, and are used in 20% of the queries (usually directly and less frequently with no more than 1-2 joins). Everything else queries the denormalised tables (usually no joins there, only some dictionary usage). Overall query performance is great, although it would have been even better since we use replacing merge trees and final.

The 2 main issues that we are facing are:

- we need to periodically cleanup the deleted rows from the replacing merge trees, since the application does lots of upserts and deleted rows just stay there.

- there is not much flexibility in the ordering keys of the replicated Postgres tables, unless you enable full replica identity. We took that performance hit (although nothing really noticeable in Postgres side) in order to have some flexibility and better query performance in the replicated tables in Clickhouse.

I have seen people passing leetcodes and other types of intensive interviews with high ranks but then failed miserably at their jobs. I have seen hard working/learning people, having zero idea on specific programming language but know some basic fundamentals become the best engineers. I know lots of people who are great engineers and have become VPs of engineering, staff engineers, architects, etc. who didn't complete the technical assessment successfully. This means that they weren't perfect at the time of the interview but had other skills to succeed in their jobs in various companies. They were hard workers, great learners, versatile, innovative, etc. But that was a past era of interviews when people were not searching for perfection but mostly evaluating soft skills. Nowadays people just search for the perfect candidate. Any error in the technical assessment or leetcode means rejection, any imperfect/doesn't meet my own solution response also means rejection. Nobody checks for or evaluates soft skills, nobody checks whether I have solved similar problems in some past experience. I guess that's the issue with bad fits, not just the lack of technical skills (which I agree that needs to be evaluated as well but not in an extremely hard and strict yes/no format unless you are google and co).

I have seen quite a few times, VPs, EMs or lead/senior engineers being angry and resentful towards people switching companies and/or teams. A f2f meeting _could_ help, but not always.

Nice one, I am going to check it out. Just curious, is there anything with a forever (until further notice) free tier for toy projects?

Clickhouse on a cheap hetzner vm or server?