Eh, uh Data eh Leadership Conference eh at eh eh the Baseball eh Stadium
Well, duh! Who doesn’t want to attend a leadership conference (as an entrepreneur/founder) in a baseball stadium! How fun? :) I love to learn and love a cool environment
Why Attend: I’m trying to build a billion-dollar business over time so you know, any chance i can have to study leadership, I’ll take it. I’m grateful to live in such a robust city. LOL robust is a very tech-leaning term, so i’ve learned lately. IDK if i used it propely here. But, let’s say I did.
Photo Collage + Commentary:
Notes from the Event:
RED ALERT, RED ALERT - hahhahahhahah. This event is NOT about leadership. Its about tech and data. OMG. Its liek about a super niche category of covidng and open source data storage??? So… actually it relates to the company I’m working to build. Competely. SO… here we go. I’m in WAY over my head, arrived late but just in time for lunch, and here we are. TEhy even spelled my name wrong on my badge. My lat name for tody is Kairk. Cool. Prounounced like park? Kelly Kairk. LOL. Love it.
Now a guy is talking who is a leader at Apple. This makes me laugh cause once again, I think of Ryan Cohen’s influence and how he’s a former top shareholder. haha homg. I’m such a super fan.
Okay so motivation for Valky is latency-sensitive use-cases, needs ocal reads
Multiple clusters need to be in data centers close to the application.
data centers different from where they need to be read
when data is dynamic and needs to e in sync for all data centers aroudn the world. Like a global counter of impressions on a post. Engagement across the world needs to update the numbers in all the data centers.
Not having any data in the new reason after switchover can cause a lot of cache misses and a lot of (idk)
its good to keep your data in sync so your backup data has warm data.
Apple, AWS, Snap Inc, Reddit, Nextdoor, Netflix (all going to talk soon -cool!!!)
Traffic over a wide area network doesn’t react like traffic in the same data center.
This guy has the world’s most simple ever ever ever PPT i’ve ever seen in my life. hahah omg.
But his voice/accent is so hard to follow along with. Strong indian accent.
And I think he’s reading off a script so that’s extra tough. I’m gonna wait for the next speaker. He’s running 10 minutes late with his speech, so it must send soon. Was only supposed to be 20 minutes
Yep. He’s finishing it. Oh ut he wants to take questions.
Okay these next two guys are giving a speech. These two guys. The’yre going to talk about something and internals and use cases.
lots of people haven’t been exposed to the search model of waht their search does in the Valkey context. These guys work at google/Amazon. Their new tool allows you to create an index model over exisitng data.
Your data lives where it always did. hashkeys, adjacent keys.
When you create an index over those keys, it automatically populates.
If you change the key, the contents of that key become immediately searchable. You don’t need to execute a new command. This guy talking, honestly, you’d not think he’s tuned into this stuff. He looks like a grandpa or a great-grandpa. Just makes you realize not to judge books by their covers and that elders are wise. ANd lots of peopel have been in this industry for a long time.
Their Query side has a sophisticated language that lets you extract keys out of it and take those keys you’ve located through the searching procedure. Now that you found the keys and the match criteria, give ack the key and the field and the key.
Aggregations can do that data and server-side processing.
The redisearch ecosystem has a module. The Valkey is pretty muc the same commands and query language.
Code base is separate. No code, licensing issues are nonexistent. Theyre are lots of internal differences on how this works. If you’ve been using the readisearch module, you can do it with almost no change back and forth.
You can have any number of indexes. There’s no limit on them. You could create hundreds or thousands and not notice any performance change.
- LOL suddenly i got a wrong email from some guys talkign about prompting AI governemance and it says stuff like:
Get to Yes — how do you get an AI system safely approved in the first place? Things like building an AI inventory, triaging new use cases, mapping policy to actual controls, getting data access approved, vetting vendors, designing human oversight checkpoints, red-teaming before launch.
Stay at Yes — once approved, how do you keep it compliant? Monitoring for drift, detecting when AI can suddenly see data it shouldn't, enforcing restricted topics, collecting continuous audit evidence, logging agent actions, managing exceptions when things bend the rules temporarily.
Recover to Yes — when something breaks, how do you get back to governed operation? Prompt injection attacks, sensitive data leaks, agents taking unauthorized actions, hallucinations causing real business errors, regulatory inquiries, audit failures, rebuilding stakeholder trust after a public failure.
LOL anyway. okay back to this. This next speaker has a strong accent, some sorta european and he says “ummmm” but it sounds liek “eeh” aaeh” and you can’t hardly understand him. What is it with these tech companies all having guys you can’t understand. For real. I’m not trying to be rude, but I’m being honest. Its wild. And I used to teach english overseas. I am experienced in interpreting nonnative speakers. But its serious.
Oaky so the scenario is an article library, users search content
omg lemme live-essay what he says. eeh eh moving on to ehh how ehhh we specific this search ehhh these esa eeh the first eeh the worst weeh wh articles is ehhh the detective appear in ehhh the typing.. you may eh the query ehh special eh eh is it the if it is not any attribute eh eh eh ze ze query criteria is considered eh indexed. against all ze eh eheh text fields which theory are assicated. The second object shows how you can combine multiple ze articles so we extend ehhh with an edition eh filtering, eh we specify we are interested in only eh eh the articles eh of the filler. eh and wil ze ze robot in app, apprising appears upriging eh. The last example is ze eh eh specificied eh query which eh eh eh combnies eh which filter eh eh genre and eh eh its published eh and date. eh em em its eh partic eh paritulary timestamp which is presented eh numerice eh fields eh eh require starting in eh eh so where hte stat is epscified with eh eh the users a special, eh indicaition, eh whch is uh which is eh positive number
moving ont the second use case eh eh is realated to the specific searh we are maintaining the catalogue which is eh eh the represetnative of the cproduct eh the first query sepcify if we are itnerested to eh only eh ze eh top 5 eh regular productes match eh any vector and ze if you, if you rare syntex its a little bit dffereint from what we’ve seen earlier eh one of which is eh eh left port which is a start, basically its eh eh a speicifc filter. whereas the vector if we operate the filter of hte dta set vector search we operate on
we are not constraining; we are looking for all ze ze ze recordes ze ze products ze ze ze input.
The second input is a bit more complex and dis ze ze cooperate ze eh filter. Instead of the start, we are actually constraining the latest eh eh eh to the product we see product while. The enterprise range is up to $100. betond eh searhcing we also support eh eh eh ft agguregate a powerful pipeline. service side pipeline. Comprised eh of several stages. The first stage is filtering, apply and specify what ed how do you. eh, the records you filter eh the group and production, and how do you profile and how do you eh sort and specify how do you them been done. in this particular example, eh eh it shows how you can craft a 2uery
The eh eh eh eh how many instances you’d like to to have meaning they’re having products in between the price range of 50, 50-100, which color is read. Following on th syntex titself eh eh the aggrigatoin it is ion eh is a filter and eh eh is in relation and function and in this project, grouping the reduction eh eh
omg now he has a new slide but they’re running like 20 minutes late to a literally 20-minute speech. Not many people in the audience are paying attention
Starting off with this capability. eh It was very important for us to eh to eh eheheh high performance in this scale and solution. eh and eh eh we are incorporating a coupel mechanism tin order to avoid eh eh remain optimized in teh sense that eh eh its eho eh doesn’t use any lock and processing commands and uh its we do scale literally pass and eh eh a locking port to data packs. but it is capable, and it performs and it also performs and provie very high an low latency
(okay now i’ll cut out his e'h’s and see if we can get more)
between the eh eh each eh span of (lol) okay
and the concurrency data support concurrent with the right and not read and write at the same time. this we’re implementing this mechanism and this ensures it is basically for the pass, which is very important for Valkey in general. and we can scale and the architecture. because of a consistency, this means that .
omg this guy is being told to hurry up cause yeah eh’s super over his time and the next guy was supposed to be done talkign by now. woild. and they told him to finish up and he laguhed and said okay. but omg you can’t undrestna dit all.
Then the other guy on stage is like laughin gand looks embarrassed to be up there 2with him wild ehahahah. I’m like jsut thinkign this whole time - this is a guy who is MUCh etter by email. thisis the kind of guy Ai can help him write a letter.
Encourage everyone to jump in and eh eh eh thank you. omg what hte heck who knows. but idk how they had that guy get up on stage. unreal. unreal.
the emcee says he’s excited to have folks use vathe speech
OMG the guy sititng next to me is the next speaker. That’s exciting. he goes up to the front stage.
And the last speaker literally left his laptop up there by mistake and then went up to go get it. He was editing his speh in hte back for with me right there one minute before. Okay let’s see if he’s a good speaker. hahah woo hoo let’s go hneighbor.
He starts off saying, “hey guys” and he works at snap. working in (idk what)
The net few minutes they’ll talk about small (idk) omg this guy youu can’t understand what he says at all hahaha ahhhhhhhh
snap builds ontop of envoy and build uh all the computer idk. omg his accent is so strong and he’s reading a speech.
hopes we can all walk away with how they build a snapcache and I've pillars that work really well.
A signle request fans out into millions of cache
100s-1000s of candidates evaluated per request
100’s of millions of feature-set lookups per seconds
10s-100s feature sets per candidate
At this scale (i’m just going to type what is oon his ppt) at this scale, every team that touches a cache hits the same walls. connection blow-up. Cancellation bugs. Painful migrations. Cache stampedes. Quiet six-figure egress bills. So we built the plumbing once.
All these are the things which we incorporated inside the snap cache library so that we (idk)
A stateless, handle-based cache library built on Svnyo’s Redis proxy components
stateles: no per-op state on the cahce. All state lives on operation handles, filters destroy the handle, the in-flight Redis request ancels itself
Move-only: zero copies on the hot path. Key and vlaues flow as sted: move from filter → cache → redis → callback
bults on envoy: reuse, don’t reinvent: conneciton pools, command splitter, cluster refresh, DNS casche - already battle-tested inside envoy’s redis proxy
the entire surface area auto op = cache → mget (ctx, std: move (keys) etc etc.
Some cache technology is OB rebuilds. Definitely since out itnerst is how its ranking, we didnt build on this is an open source. We leverage these componsenst and add our secret sauced that idk. IDK hahah
Snap Cash does something more.
So when you eep things longer than needed, it (idk)
When my neighbor was done speaking, he asked me if his speech was amazing or average. I told him he didnt’ have much time to talk cause everything is running late. And then I told him this is above anythin gI undrestand but he used nice colors on his powerpoint.
okay this is realyl a lot of niche talk for me . I mean, idk if i’m learning ANYTHING from this.
omg but this guy next to me, my neighor is sharing/posting pic sof himself all over the place showing his pic of him speakign even in his stock trading group and lots of things hahah omg.
Then this guy ends his talk and says thanks for this - wait, no, he keeps going. Now they want to talk about next steps:
roll out valkey to the entire Redis fleet
squeeze more from Valkey's new features:
faster replication from dual-channel replications
faster scaling from atomic slot migrations
multi-tenancy from numbered databases
migrating Valkey to Kubernetes
consolidated infrastructures, faster provisions and rotations, utilizing idle Kubernetes resources
ossnative valeky operations
GOTTA GO. :)
