Login
You're viewing the mastodonapp.uk public feed.
  • Daphne Preston-Kendaldpk@swhack.com
    Oct 31, 2026, 7:52 PM

    The Jepsen analysis of ParticleDB is so funny

    DB companies usually commission Jepsen so they can go ‘hey, this expert analysis found some bugs which could happen under very rare conditions with multiple servers and multiple clients, and we fixed them’ – it’s overall good publicity because it makes the company look diligent about data integrity, notwithstanding the discovery of bugs

    This one is ‘the expert analysis found that almost no functionality worked as advertised even with a single server and client, basic SQL semantics were horrendously broken, some new versions created even more bizarre bugs, and the analysis also reveals that nobody at the company was able to explain the bugs or how they were fixed, and even the test suite they claimed they’d already successfully run was actually failing to begin with’

    (Also: Damn, it must have taken the patience of a saint to work with these people)

    💬 7🔄 33⭐ 0

Replies

  • 💬 2🔄 4⭐ 0
  • Daphne Preston-Kendaldpk@swhack.com
    Oct 31, 2026, 8:02 PM

    I assume the ParticleDB company is going to be doing some damage control PR very soon.

    Your bingo cards might include squares such as

    • we understood the bugs but chose not to reveal the causes to Jepsen
    • our LLMs understand the code fully
    • some bugs were present only in pre-release versions, so they’re fine
    • we were running Jepsen’s published test suite against a different version than the one they ran it on
    • ditto, omitting mention that this was also not a pre-release version
    • ditto, also omitting mention that the version Jepsen tested was a later version, so this was a regression they could and should have caught

    I hope no other DB company is put off hiring Jepsen by this, because if you’re actually competent at database implementation, a database reviewed by Jepsen actually makes you look more competent in the end.

    💬 2🔄 9⭐ 24
  • 💬 0🔄 0⭐ 0
  • equinoxequinox@chaos.social
    Oct 31, 2026, 10:13 PM

    @dpk "WHERE TRUE matched nothing (#35)" 😂

    "Joins … confused similarly-named columns (#77)" 🤯 how the F did they even do that?!? is the database itself an LLM?

    "Inserts ignored provided columns, creating rows best described as “deeply cursed” (#16).¹²"
    "¹²: Jepsen encourages the consistency model research community to come up with more official terms for this.↩︎"

    🤣 :awesome: :thisisfine:

    💬 1🔄 3⭐ 2
  • equinoxequinox@chaos.social
    Oct 31, 2026, 10:18 PM

    @dpk "Strings sometimes turned into eight-bit integers (#37)"

    OH! so THAT'S what happened with #eMudhra|'s keys!

    💬 0🔄 0⭐ 2
  • 💬 0🔄 0⭐ 0
  • 💬 1🔄 2⭐ 0
  • 💬 0🔄 1⭐ 1
  • 💬 0🔄 1⭐ 0
  • Dan Sugalskiwordshaper@weatherishappening.network
    Oct 31, 2026, 9:54 PM

    @dpk I’m impressed the ParticleDB folks actually thought their stuff was ready for Jepsen’s analysis. A little worried about their disconnect from reality given the number and obviousness of the bugs, but still impressed.

    💬 0🔄 0⭐ 2
  • Lucky ✨lucky@mastodon.nz
    Oct 31, 2026, 10:04 PM

    @dpk oh, now this is a WONDERFUL treat to read! The line "perplexing nondeterministic failures" really jumped out to me as a very polite summary of everything happening in vibecoded software right now. (I cannot actually prove that ParticleDB did their refactor entirely with LLMs but this analysis is full of clues that make me fairly convinced of that.)

    💬 0🔄 0⭐ 0
  • Bj�rnbjorndown@mastodon.social
    Nov 1, 2026, 9:29 AM

    @dpk "Conductor was initially built to store data in Redis and Postgres, but to improve performance, Orkes opted to write their own database."
    Ah yes. Whenever I find myself thinking "My application is slow it must be postgres or redis" I let myself consider it, then I shake my head, chuckle to myself and move on. But of course a system like Claude or whatever has an incentive to let you burn a gazillion tokens on a brainfart.

    💬 0🔄 0⭐ 0
  • ashapisashla@www.librepunk.club
    Nov 2, 2026, 12:44 AM

    @dpk
    "the company's claims of 'full ACID' were somewhat premature" to start out with is how you know this is gonna be great

    💬 0🔄 0⭐ 0