Marti S
@8vicat.bsky.social
Code and data. Working on temporal graphs.
Ah, and I’m the author of the satirical card game PLAI: about AI ethics, startup culture and burnout.
Ah, and I’m the author of the satirical card game PLAI: about AI ethics, startup culture and burnout.
The game is ready for order! Now with an optional (free) extra of being blessed by a cat. See plai.cards for endless fun and more than 106 tech culture references
July 29, 2025 at 9:24 AM
The game is ready for order! Now with an optional (free) extra of being blessed by a cat. See plai.cards for endless fun and more than 106 tech culture references
The games just arrived, but she asserted dominance over them so it might be difficult to send them your way 🙀
June 24, 2025 at 11:47 AM
The games just arrived, but she asserted dominance over them so it might be difficult to send them your way 🙀
Reposted by Marti S
posted today!
BTW I peeked at the automerge Rust? Collaborative editing is an example where one probably *has* to resort to unsafe behavior (you're the expert there!) so I'm mostly advocating for more encapsulation/comments in that case.
jhellerstein.github.io/blog/crdt-do...
BTW I peeked at the automerge Rust? Collaborative editing is an example where one probably *has* to resort to unsafe behavior (you're the expert there!) so I'm mostly advocating for more encapsulation/comments in that case.
jhellerstein.github.io/blog/crdt-do...
CRDTs #3: Do Not Read!
Ever used a CRDT, thought you were safe, and—boom—you bought a Ferrari you didn't mean to? It could happen to you! The truth is that CRDTs are dangerous to…
jhellerstein.github.io
May 28, 2025 at 8:40 PM
posted today!
BTW I peeked at the automerge Rust? Collaborative editing is an example where one probably *has* to resort to unsafe behavior (you're the expert there!) so I'm mostly advocating for more encapsulation/comments in that case.
jhellerstein.github.io/blog/crdt-do...
BTW I peeked at the automerge Rust? Collaborative editing is an example where one probably *has* to resort to unsafe behavior (you're the expert there!) so I'm mostly advocating for more encapsulation/comments in that case.
jhellerstein.github.io/blog/crdt-do...
Reposted by Marti S
New post! I wrote about malaise in the streaming space. I'm not sure what's going on, but I'm thinking part of it has to do with Kafka.
Kafka: The End of the Beginning
A decade of focus on adoption has payed off. Now it's time to innovate.
materializedview.io
May 30, 2025 at 2:51 PM
New post! I wrote about malaise in the streaming space. I'm not sure what's going on, but I'm thinking part of it has to do with Kafka.
Reposted by Marti S
The main gotcha is that some of Rust's time logic (e.g. Instant::now()) doesn't exist in WASM, and you'll just get a runtime error. The fixes involve making these optional.
This imo goes deeper than WASM: it proves that timely doesn't rely on real time, and is purely event driven; no timeouts.
This imo goes deeper than WASM: it proves that timely doesn't rely on real time, and is purely event driven; no timeouts.
May 7, 2025 at 11:17 PM
The main gotcha is that some of Rust's time logic (e.g. Instant::now()) doesn't exist in WASM, and you'll just get a runtime error. The fixes involve making these optional.
This imo goes deeper than WASM: it proves that timely doesn't rely on real time, and is purely event driven; no timeouts.
This imo goes deeper than WASM: it proves that timely doesn't rely on real time, and is purely event driven; no timeouts.
Reposted by Marti S
Why is Remote ≠ Local++
Because distribution adds partial failure—and you can’t abstract away partial failure.
In the presence of partial failure, even the basic rules of reasoning break down.
— An Equational Theory for Transactions
bsky.app/profile/domi...
Because distribution adds partial failure—and you can’t abstract away partial failure.
In the presence of partial failure, even the basic rules of reasoning break down.
— An Equational Theory for Transactions
bsky.app/profile/domi...
May 6, 2025 at 9:05 AM
Why is Remote ≠ Local++
Because distribution adds partial failure—and you can’t abstract away partial failure.
In the presence of partial failure, even the basic rules of reasoning break down.
— An Equational Theory for Transactions
bsky.app/profile/domi...
Because distribution adds partial failure—and you can’t abstract away partial failure.
In the presence of partial failure, even the basic rules of reasoning break down.
— An Equational Theory for Transactions
bsky.app/profile/domi...
Huge electricity outage in Spain. A few places have internet, so everyone is mow congregated outside Lidl stores 🤣😅
April 28, 2025 at 2:03 PM
Huge electricity outage in Spain. A few places have internet, so everyone is mow congregated outside Lidl stores 🤣😅
Reposted by Marti S
Vim and its language is still one of the best things I learned as a data engineer, but even more so as a writer.
Why Vim Is More than Just an Editor: Vim Language, Motions, and Modes Explained.
Why Vim Is More than Just an Editor: Vim Language, Motions, and Modes Explained.
April 23, 2025 at 6:43 PM
Vim and its language is still one of the best things I learned as a data engineer, but even more so as a writer.
Why Vim Is More than Just an Editor: Vim Language, Motions, and Modes Explained.
Why Vim Is More than Just an Editor: Vim Language, Motions, and Modes Explained.
Reposted by Marti S
Chiming in with my analysis of Timely and Differential, as well as internal consistency concept: www.streamingdata.tech/p/attempting...
April 18, 2025 at 10:13 PM
Chiming in with my analysis of Timely and Differential, as well as internal consistency concept: www.streamingdata.tech/p/attempting...
Reposted by Marti S
Ok, y'all. This took me several weeks and a ton of help from @frankmcsherry.bsky.social and @lalithsuresh.bsky.social. I dug into timely dataflow, differential dataflow, and DBSP to get you up to speed on IVM engines and materialized views. Enjoy!
Everything You Need to Know About Incremental View Maintenance
An overview of incremental view maintenance, why it’s useful, and how you can implement it.
materializedview.io
April 18, 2025 at 6:30 PM
Ok, y'all. This took me several weeks and a ton of help from @frankmcsherry.bsky.social and @lalithsuresh.bsky.social. I dug into timely dataflow, differential dataflow, and DBSP to get you up to speed on IVM engines and materialized views. Enjoy!
Reposted by Marti S
Where we're at with the war on Rust compilation times.
Cutting Down Rust Compile Times With One Thousand Crates
We'll explore how we can leverage all cores on a machine with rustc using many crates to speed up compilation.
www.feldera.com
April 15, 2025 at 11:17 PM
Where we're at with the war on Rust compilation times.
It’s been in slow cooking mode for a couple years. Tech humor, sarcasm and AI ethics references in a box to play with your colleagues.
April 13, 2025 at 7:37 AM
It’s been in slow cooking mode for a couple years. Tech humor, sarcasm and AI ethics references in a box to play with your colleagues.
Reposted by Marti S
A few days ago I gave a keynote at the PaPoC workshop on Byzantine Eventual Consistency and Local-first Access Control. It wasn't recorded, but slides are here
speakerdeck.com/ept/byzantin...
speakerdeck.com/ept/byzantin...
Byzantine Eventual Consistency and Local-first Access Control
Slides from a talk given at the 12th Workshop on Principles and Practice of Consistency for Distributed Data (PaPoC), Rotterdam, Netherlands, 31 March 2…
speakerdeck.com
April 5, 2025 at 2:15 PM
A few days ago I gave a keynote at the PaPoC workshop on Byzantine Eventual Consistency and Local-first Access Control. It wasn't recorded, but slides are here
speakerdeck.com/ept/byzantin...
speakerdeck.com/ept/byzantin...
Reposted by Marti S
What if your app could combine the simplicity of physical replication with the efficiency of logical replication? Meet Graft: lazy, partial, strongly consistent edge replication.
#OpenSource #EdgeComputing #Replication
sqlsync.dev/posts/stop-s...
#OpenSource #EdgeComputing #Replication
sqlsync.dev/posts/stop-s...
Stop syncing everything
Discover Graft, an open-source transactional storage engine built to solve the challenges of syncing data at the edge. Inspired by lessons from SQLSync, Graft enables lazy, partial, and strongly consi...
sqlsync.dev
March 31, 2025 at 4:30 PM
What if your app could combine the simplicity of physical replication with the efficiency of logical replication? Meet Graft: lazy, partial, strongly consistent edge replication.
#OpenSource #EdgeComputing #Replication
sqlsync.dev/posts/stop-s...
#OpenSource #EdgeComputing #Replication
sqlsync.dev/posts/stop-s...
Reposted by Marti S
Opensource saving tons of money while boosting productivity 🥰
March 18, 2025 at 1:44 PM
Opensource saving tons of money while boosting productivity 🥰
Reposted by Marti S
Why not use analytical stacks for transactions?
- Lack transactionality/correctness: difficult to guarantee state changes happen only once
- Lack efficient state query mechanisms: optimised for analytics not range scans or key lookups/indexing
- Lack support for high availability and replication
- Lack transactionality/correctness: difficult to guarantee state changes happen only once
- Lack efficient state query mechanisms: optimised for analytics not range scans or key lookups/indexing
- Lack support for high availability and replication
Why aren’t software engineers using data engineering technology (like file formats, table formats, OLAP databases, object storage, query engines, pipelines etc) to build/operate event-driven architectures (incl event streaming, processing, sourcing)?
March 16, 2025 at 4:43 PM
Why not use analytical stacks for transactions?
- Lack transactionality/correctness: difficult to guarantee state changes happen only once
- Lack efficient state query mechanisms: optimised for analytics not range scans or key lookups/indexing
- Lack support for high availability and replication
- Lack transactionality/correctness: difficult to guarantee state changes happen only once
- Lack efficient state query mechanisms: optimised for analytics not range scans or key lookups/indexing
- Lack support for high availability and replication
Reposted by Marti S
One does not simply copy an Iceberg table
March 12, 2025 at 4:48 AM
One does not simply copy an Iceberg table
Reposted by Marti S
It seems to be the more widely used terminology. It's also a bit more specific: "partitioning" could be any split of data into subsets (not necessarily across nodes), whereas "sharding" is understood as meaning splitting the data across multiple nodes.
March 11, 2025 at 7:33 PM
It seems to be the more widely used terminology. It's also a bit more specific: "partitioning" could be any split of data into subsets (not necessarily across nodes), whereas "sharding" is understood as meaning splitting the data across multiple nodes.
Reposted by Marti S
Learn how hyperlight can create virtualization-based sandboxes for WASM applications:
Hyperlight: Achieving 0.0009-second micro-VM execution time - Microsoft Open Source Blog
In this post, we’ll take the demo application and show how it demonstrates one way you can use Hyperlight in your applications. Learn more.
opensource.microsoft.com
February 12, 2025 at 5:17 PM
Learn how hyperlight can create virtualization-based sandboxes for WASM applications:
Reposted by Marti S
Most of my promotions were based on my ability to attach my work to the metrics below. Working at a major cloud provider made this a lot easier because we tracked everything. It largely comes down to earning and sharing the credit across different teams including engineering, product, and sales.
1. Impact. How much revenue does my work protect or generate?
2. Quality. Does my work meet or exceed customer expectations?
3. Efficiency. Reward making the right buy versus build decision.
4. Reusability. How do others leverage my work?
5. Supportability. How much work do I create for others?
2. Quality. Does my work meet or exceed customer expectations?
3. Efficiency. Reward making the right buy versus build decision.
4. Reusability. How do others leverage my work?
5. Supportability. How much work do I create for others?
a question for the people who write code for money:
if you could wave a magic wand and have your performance/promotability measured on any 5 metrics of your choice, what would those metrics be?
if you could wave a magic wand and have your performance/promotability measured on any 5 metrics of your choice, what would those metrics be?
January 23, 2025 at 10:54 PM
Most of my promotions were based on my ability to attach my work to the metrics below. Working at a major cloud provider made this a lot easier because we tracked everything. It largely comes down to earning and sharing the credit across different teams including engineering, product, and sales.
Reposted by Marti S
The difference between Software Verification and Software Testing
Software Verification aims to proof that all traces are within the specification
Software Testing aims to find that some traces are outside the specification
bsky.app/profile/domi...
Software Verification aims to proof that all traces are within the specification
Software Testing aims to find that some traces are outside the specification
bsky.app/profile/domi...
January 15, 2025 at 10:55 AM
The difference between Software Verification and Software Testing
Software Verification aims to proof that all traces are within the specification
Software Testing aims to find that some traces are outside the specification
bsky.app/profile/domi...
Software Verification aims to proof that all traces are within the specification
Software Testing aims to find that some traces are outside the specification
bsky.app/profile/domi...
Reposted by Marti S
The MemoryDB paper shows the power of separating responsibilities through clever composition. I think this DB frontend/execution plus a distributed transaction log pattern can be promising for creating serverless variants of many popular databases. E.g., Aurora adopts a similar decoupling approach.
January 5, 2025 at 4:37 PM
The MemoryDB paper shows the power of separating responsibilities through clever composition. I think this DB frontend/execution plus a distributed transaction log pattern can be promising for creating serverless variants of many popular databases. E.g., Aurora adopts a similar decoupling approach.
