Having built a prototype of a system like Ursa myself, this isn't something that you need to use at all, especially because it seems like this is only available in S3 Express One Zone.

Ursa is available across all major cloud providers (GCP, Azure, AWS). It also supports pluggable write ahead log storage. For latency relaxed workloads, we use object storage to get the cost down. So it works with AWS S3, GCP GCS, Azure Blob Store. For latency sensitive workloads, we use Apache BookKeeper which is a low-latency replicated log storage. This allows us to support workloads ranging from milliseconds to sub-seconds. You can tune it based on latency and cost requirements.

If they're offering SSO using a 3rd party provider like Auth0 then they probably have to charge for it because of how expensive it is.

this algorithm is flawed. i created a test which can read failed write https://gist.github.com/pdeva/58854fa644d074cf07aa0ecca9a465...

Do you guys only offer SAML in your hosted SaaS?

Everything is open-source.

We implement providers when a paying customer requests them (Team plan for OIDC-compatible providers, Growth plan for everything else, including SAML). Once we've implemented them, though, everyone benefits.

To our surprise, as of right now we haven't received any requests for SAML from our customers.

Ah gotcha, that's a nice way of approaching it. Best of luck to you guys.

I always wondered about this, a while ago I saw a library similar to this that modeled FSMs in Postgres, and had a comment saying "todo: cycle detection". And made me wonder if it's even possible to do this.

Hmm… consider the following. Your FSM is acyclic iff you can assign each state an integer depth such that a state at depth d only ever transitions to states at depths strictly greater than d. So consider the following tables:

    CREATE TABLE states (
        state order_state PRIMARY KEY,
        depth int NOT NULL,
        UNIQUE(state, depth)
    );
    CREATE TABLE transitions (
        start_state order_state NOT NULL,
        event order_event NOT NULL,
        end_state order_state NOT NULL,
        start_depth int NOT NULL,
        end_depth int NOT NULL,
        CONSTRAINT transitions_start_depth_correct
            FOREIGN KEY(start_state, start_depth) REFERENCES states(state, depth)
            ON UPDATE CASCADE,
        CONSTRAINT transitions_end_depth_correct
            FOREIGN KEY(end_state, end_depth) REFERENCES states(state, depth)
            ON UPDATE CASCADE,
        CONSTRAINT transitions_depth_increases CHECK(end_depth > start_depth),
        PRIMARY KEY(start_state, event)
    );

Let's bang on it for a quick test. You can define a state machine; here's one that roughly matches the regex `^(AB|BA)$` (I know I'm being a bit sloppy):

    fsm=> CREATE DOMAIN order_state AS text;
    CREATE DOMAIN
    fsm=> CREATE DOMAIN order_event AS text;
    CREATE DOMAIN
    fsm=> -- "CREATE TABLE"s as above, then:
    fsm=> INSERT INTO states(state, depth) VALUES('start', 1), ('a', 2), ('b', 2), ('done', 3), ('error', 3);
    INSERT 0 5
    fsm=> INSERT INTO transitions(start_state, event, end_state, start_depth, end_depth) VALUES('start', 'A', 'a', 1, 2), ('start', 'B', 'b', 1, 2), ('a', 'B', 'done', 2, 3), ('b', 'A', 'done', 2, 3), ('a', 'A', 'error', 2, 3), ('b', 'B', 'error', 2, 3);
    INSERT 0 6

And, as you need to modify it, you can increase a node's depth to make room for intervening nodes:

    fsm=> UPDATE states SET depth = 4 WHERE state = 'error';
    UPDATE 1
    fsm=> TABLE transitions;
     start_state | event | end_state | start_depth | end_depth 
    -------------+-------+-----------+-------------+-----------
     start       | A     | a         |           1 |         2
     start       | B     | b         |           1 |         2
     a           | B     | done      |           2 |         3
     b           | A     | done      |           2 |         3
     a           | A     | error     |           2 |         4
     b           | B     | error     |           2 |         4
    (6 rows)

But you can't decrease a node's depth too far:

    fsm=> UPDATE states SET depth = 2 WHERE state = 'error';
    ERROR:  new row for relation "transitions" violates check constraint "transitions_depth_increases"
    DETAIL:  Failing row contains (a, A, error, 2, 2).
    CONTEXT:  SQL statement "UPDATE ONLY "public"."transitions" SET "end_state" = $1, "end_depth" = $2 WHERE $3::pg_catalog.text OPERATOR(pg_catalog.=) "end_state"::pg_catalog.text AND $4 OPERATOR(pg_catalog.=) "end_depth""

And you can't introduce transitions that don't increase depth:

    fsm=> INSERT INTO transitions(start_state, event, end_state, start_depth, end_depth) VALUES('done', 'AGAIN!', 'start', 3, 1);
    ERROR:  new row for relation "transitions" violates check constraint "transitions_depth_increases"
    DETAIL:  Failing row contains (done, AGAIN!, start, 3, 1).

Now, I don't know that I would immediately recommend this for high-throughput production use. You're storing "unnecessary" state not once but many times (each state's depth appears `1 + \deg(v)` times), plus additional indices and lookups. But I do think it meets the desired consistency goals!

Amazing! I also learned about domains with your comment.

I've seen a bunch of people recommending alternatives that are also Python based, makes me wonder if there's similar things for other languages like Go, Rust, etc

There's similar things in other languages like https://www.gpui.rs/ and https://github.com/linebender/xilem.

Mesop also drew inspiration from frameworks like LiveView (for Elixir/Phoenix) https://hexdocs.pm/phoenix_live_view/welcome.html which have demonstrated the viability of building server-driven web UIs for a number of years now.

I'd be curious to see a direct performance comparison between their python and JS workers. Based on my own experience with pyodide, I'd wager there might be up to a 2x performance penalty.

This tends to be my experience, but mainly when it comes to areas that neither model know about. Claude 3 Opus will happily make something up, but ChatGPT4 will point out where it's lacking.

Can you point to any resources to read up on this?

Thanks!

Which is interesting considering Freeman Dyson was the guy that made the connection between Schwinger's and Feynman's QED.