GitHub's reached the point where scaling out is not an option anymore probably about 3 years ago but no one noticed because it felt like they could. We're in a place were we need to start reshaping the underlying search indexes in order to have performant queries.

It has been 0 days since paginating through search data was a problem. When you are simply scrolling data MySQL is almost always going to give you better performance. Search engines just dont scroll text data well because of the orientation of the data.

Okay so on the vector math thing I am working on a sheet with Copilot to calculate the memory size needed... It produced this banger: SWITCH(B6,"Float16",2,"Int8",1,"Binary (1-bit)",0) Sure, technically a binary is 0 bytes but we need to determine how much memory is consumed with a gazillion bits

Who ever thought that my college math would help in life. I'm now working on a probabilistic formula to find the most number of items I can include in a query to ensure that some percent of queries only effects some percent of the cluster. 😭

Always fun visiting friends at the AWS booth 🤗 #AWS #OpenSource #ATO

I guess you could say I Dynatrace’d my way over to the RiOT demos. Always fun seeing what unique ways vendors engage conference viewers. #dynatrace4developers

Who’s going to All Things Open this year? Any evening events to check out? 👀

Woke up today and I am once again mad about the Bitnami rug pull. They were the only one I trusted publishing an OpenSearch image that could be deployed on a Raspberry Pi (CM4). And now they want me to pay for it. Rug-pulls are not a business strategy.

Few things are certain in the universe. Taxes, death, and people will misspell "Elasticsearch" I've seen: - ElasticSearch - Elastic search - Elastic-search - Elastisearch - Elastic's search - OpenSearch 🥵

Who all is using Apache Unomi? I have a feeling that it's probably like a handful of big players but I wonder if there's more than I know. unomi.apache.org/manual/2_7_x...

Elastic: please send us trace logs from these loggers... Me: you realize that is going to be literally billions of logs if we turn those on right? Are you sure that is what you want? Elastic: yes Me: okay here you go... Elastic: so we cant process all these logs can you send less 😐

"Where is the documentation telling us not to do this" - never thought I'd see that one IRL. If you cant find any documentation how to do something that typically means you can't do it. If you wonder why it's not documented you reach out and ask about it and then you'd find its unsupported.

So funny story I used IBM about 10 years ago. I created an account and after I was done with the experiment, I closed it. I have been receiving emails for the last 10 years about infrastructure changes happening at IBM. I cannot login anymore because my account doesn’t exist.

If you are in Germany and a search expert come work with me! We're looking for someone with strong experience in Elasticsearch particularly as it relates to both search engineering and infrastructure for large scale search. www.github.careers/careers-home...

I need to write a benchmarking databases guide. Too often I see people loading like 10k docs making 500 requests can then publishing results as facts. Don’t get me wrong I started out that way too. Good database benchmarks need to have a few things: 🧵

Ruh rogh, ChatGPT is leaking 💧

I totally forgot you can set backgrounds in iTerm until recently. I cant wait till the next time I have to share my terminal session with execs 😆

Hot take 🌶️ ChatGPT isn’t a LLM… it’s a controller. Controllers provide the: - Conversation context - Tool calling (search, plugins) - Ability to iterate I’ve seen a lot of people conflate the two and while I understand it’s important to know what value each provides.

I just realized ollama can have several models loaded into memory at once. Anyone hooked OpenSearch up to Ollama for document embeddings yet?

I am nervous for the next generation of ChatGPT engineers. I've just had several different models tell me to do things that would've for sure risked the integrity of our Elasticsearch infra. Nuking active scrolls during a reindex is a bad idea and Claude should feel bad for suggesting it.

Message of the day, run your own race. Some people start companies when they’re 22. Some people when they are 60. Meanwhile others work for companies their whole life. Don’t rush it just run the race you feel happiest running rather than chasing after someone else.

Im having a really hard time prioritizing which conferences to go to... On the potential docket: - Nvidia GTC DC - All Things Open - Literally any Elastic{ON}

My NVIDIA rep: Heads up the distributor should've gotten a bunch of inventory of the Spark so yours is probably coming soon... *checks tracking* - My date was pushed out another month 😭

Oh man there's so much tea in the OpenSearch community to catch up on. 🫖 I've not really paid too much attention since starting at GitHub (because you know... Elasticsearch). There've been a lot of people shuffling and foundation restructuring that I need to go through.

If you have an index created in version 8.7 of Elasticsearch, even if you upgrade the cluster to 8.18, that index will keep the defaults from the version it was created on. I get that for fields that I've used. For new things like `semantic_text` feels weird github.com/elastic/elas...

So I’ve put off developing more for github.com/downtime-ind... until I could get it hosted and things inventoried. Today I pulled out the Circuit and began naming containers so I can finally use it. 🔥 everything’s getting programming language names.

OpenAI just dropped new open source models 🔥 My colleague Logan spent his morning testing gpt-oss-20b - which runs on consumer hardware, is Apache 2.0 licensed, and the reasoning quality is solid. Also, full chain-of-thought access so you can see exactly how it's thinking.

Feels like a lot of my favorite content producers haven’t made the switch here yet which is kinda sad but it’s doing great things for my mental health. I’ve mostly completely detached from like 90% of my social media. Feels good 😊

not me like: why is the docker container saying it's out of memory? I allocated 20GB how much more do you need?! Not my docker settings: memory limit 6GB... 😐

It's been 0 days since I've had to describe why `_cluster/reroute` doesn't work the way we want. While you can move shards this API isn't intended to allow users to orchestrate the whole cluster. All the moves it makes are still subject to Elasticsearchs allocation deciders.

I’ve spent hours solutioning with a customer. Explained key explosion. Provided an architecture to work around it…. They came back with *key explosion v1.0*

Okay what on earth, this is too good: youtu.be/15_-hgsX2V0

There are a few old guys talking about 3d printers in front of me and it's taking everything in my body to resist going and telling them everything I know about printing. 😭 I need to do work but they're tempting me sooooooo

Oh great. It's available in 8.19... く(;_;) Well, guess all that practice testing 8.18 will help us roll 8.19 even sooner 😐

Search @elastic.co is literally changing so fast I cant keep up. 8.18 was just released a few months ago and already there are features in 9.0 that I need. I'm trying to get us setup to use `semantic_text`. I want to use BBQ, and reranking but the `index_options` for BBQ seems to not be in 8.18 😮‍💨

Credit has to be the dumbest system we’ve developed as a society. Congrats on paying off your loan!!! 🔴 Score: -29 pts Wut

I've been working on this one way too long. I pulled together a list of niche software engineering roles that I didn't know existed during college. tippybits.com/how-to-find-...

Idk if I should be happy or frustrated with the bugs I keep tripping on in GitHub's codebase 😆 So far all of them have been super niche and present in the codebase for like 15 years. Just fixed an issue with our crypto library so that's fun. I'm not really a security engineer but today I was

There comes a point with search where normal strategies just don't work anymore. Retrieval across 1-2 indexes at a time can be weighted by people. What about 5-10 though? Each index has features that makes that index special. How do you combine scores in a way that prioritizes the users?

Hats off to GitHub for letting this authentic community engagement roll. This is the hottest 💩 I’ve come across in a lot time 🔥

This is totally not alarming coming from critical infrastructure… VA just dispatched an amber alert and their site is down so no one can read the details.

What is document writing one of the best skills I picked up from AWS 😭 It’s so important though. Does anyone know of good engineering writing guides outside of AWS once since those are internal?

This is the type of stuff that’s made me leave Twitter. Never used to happen at the scale and frequency it happens now.

Here's a story about the time I broke Disney Animation's internet www.reddit.com/r/devops/com...

Inspired by @justingarrison.com here’s my “I broke prod” story from my like 3rd month at GitHub 🙈 😭 www.reddit.com/r/devops/com...

2 broken suitcases 6 flight delays 1 overnight stay 2 more flight delays We are finally on our way home. They literally called, “boarding all passengers except the Tippett’s” over the intercom because our tickets were the last to get rebooked 🫠 So glad to be going home 😭 Not pictured - 2 kids 😝

Why is it the one weekend that I need things to be chill ends up being the busiest on-call weekend 🙈 Thankfully nothings gone down hard but it’s been a fair bit of activity.

Not me trying to draw a physical decision tree to determine when we need to nuke the data on a particular server.... Paper is an under appreciated developer tool

The nvlink spine can handle 173TB/s greater than the peak data transfer of the entire internet. Shooketh

My favorite part of the Nvidia keynote is, “does that many any sense??”Jensen’s as amazed by some of this tech as we are 😆