Abe Ratnofsky

Timex x Jacquie Aiche Guiding Light 36mm Stainless Steel Bracelet Watc Designed to be a talisman of clarity, protection, and trust in the journey ahead, the Timex x Jacquie Aiche Guiding Light timepiece features a mosaic dial pieced together from shimmering mother-of-pea...

Not exactly my style, but I’m a fan of the Apache Cassandra watch

timex.com/products/tim...

Any ideas @me.bernardo.dev?

Six-decade math puzzle solved by Korean mathematician A Korean mathematician has won international recognition for solving a geometry puzzle that had resisted proof for nearly six decades. US magazine Scientific Am

“You keep holding on to hope, then breaking it […] for me mathematical research is a repetition of dreaming and waking up.”

m.koreaherald.com/article/1064...

here's the weird data structure arxiv.org/abs/2505.049... fwiw

Lots of perspectives to have on this incident, but mine is that TLS is too important to have such opaque & sharp edges. We should expect better error messages than “PKIX path building failed”, or “OPENSSL_internal:TLSV1_ALERT_INTERNAL_ERROR”.

Expose the cert chain and handshake state in errors!

Depends on the workload, but for OLTP something like zone-aware LSMs on NVMe Zoned Namespaces: snia.org/sites/defaul...

Can you provide any info so we can triage the other issue (symptoms, impact, etc)?

Have you filed GH issues for those limitations?

We mainly use bytecode injection for “in-JVM distributed tests”, which create single-process virtual clusters with multiple Cassandra instances. The bytecode injections are specified for each virtual cluster, which is typically just used for a single test.

Cassandra uses Java bytecode injection to express failpoints, so an individual test can add arbitrary execution, like failures or delays, around any existing function. This technique requires neither code generation nor extra syntax in comments.

github.com/apache/cassa...

The mailing list archives have lots of context:

www.mail-archive.com/pgsql-hacker...

Looks like Merkle Mountain Ranges: docs.grin.mw/wiki/chain-s...

Making JFR Quack: Importing JFR files into DuckDB - Mostly nerdless Learn how to import JFR data into DuckDB, to easily query and analyze your profiling data in this weeks blog post.

Learn how to import JFR data into DuckDB, to easily query and analyze your profiling data in this week's blog post: mostlynerdless.de/blog/2025/10...

Actually, I’m wrong - same process can execute out of order, gets into prefix-consistent SI at that point

Depends on the processes for A and B. If they’re the same process and B follows A, then 2. Otherwise if they’re different processes and we’re in non-strict SI, then 3, due to the real-time requirement.

Heading back home from Community over Code. So much new stuff this year in Cassandra: Mutation Tracking, Witness Replicas, Auto Repair, a bunch of new Analytics related work, and, of course, Accord!

intro slide

slide on background entropy

Abe Ratnofsky presenting "What's New in CEP-45: Mutation Tracking and Reconciliation"

More info about this CEP here: cwiki.apache.org/confluence/d...

I’ll be doing two things at this conference:
- Presenting a new approach to anti-entropy in Cassandra
- Wearing an orange cap

Come say hi!

For extensibility: bsky.app/profile/andy...

Thumbnail: Vive la Différence: Practical Diff Testing of Stateful Applications

Vol:18 No:7 → Vive la Différence: Practical Diff Testing of Stateful Applications
👥 Authors: Kexin Zhu, Michael Whittaker, Srdjan Petrovic, Robert Grandl, Sanjay Ghemawat
📄 PDF: https://www.vldb.org/pvldb/vol18/p2018-zhu.pdf

Some shared state, some message passing in FDB: github.com/apple/founda...

for your consideration: www.nts.live/shows/sounds...

Excerpt of linked paper describing how even slightly skewed statistics can produce significantly worse query plans

Here’s my favorite paper about the impact of statistics collection on query planners. At scale, Postgres can produce wildly different plans for values in and out of the MCV lists, so setting statistics_target becomes increasingly important.

db.in.tum.de/~leis/papers...

Excerpt of linked paper describing how even slightly skewed statistics can produce significantly worse query plans

Here’s my favorite paper about the impact of statistics collection on query planners. At scale, Postgres can produce wildly different plans for values in and out of the MCV lists, so setting statistics_target becomes increasingly important.

db.in.tum.de/~leis/papers...

The first line references the Jepsen report, I’m referring to this part: “Coincidentally, our own internal testing recently found this same behavior”.

Any plans to share more about the tools you’re using for transaction visibility tests?

Automated Repair Inside Cassandra for CEP-37 · apache/cassandra@6753fb4 patch by Jaydeepkumar Chovatia; reviewed by Andy Tolbert, Chris Lohfink, Francisco Guerrero, Kristijonas Zalys for CASSANDRA-19918

More big news.

CEP-37: Automated Repair Inside Cassandra; has been merged to trunk.

No more worries about managing repair, the DB will do it for you.

This long standing operational headache is now a thing of the past.

github.com/apache/cassa...

issues.apache.org/jira/browse/...

Event-looped?

accord transaction example

Accord Transactions has been merged to trunk.

This is years of work and research, trillions of test runs.

Strict Serializability isolation across a globally distributed database, for transactions spanning multiple tables and keys, using commodity clocks.

github.com/apache/cassa...

Book: Understanding Checksums and Cyclic Redundancy Checks Amazon e-book:   https://www.amazon.com/dp/B0CVXWDZ99    (Also available from country-specific Amazon web sites for:  UK, DE, FR, ES, IT...

checksumcrc.blogspot.com/p/book.html