scylla | NoSQL data store using the seastar framework

After research and consultation with product advisor from scylla. i has resolved the problems :

Scylla arsitecture if use node in virtual must be a public cloud source https://www.scylladb.com/product/technology/

So docker-compose.yaml i change map port to public ip like this

Yes, it is guaranteed (I am the author of that article).

Since you are running docker (no K8s), all parameters that you use when spinning up your container are saved under /etc/scylla.d/docker.conf.

You can edit this file and then use supervisorctl restart scylla to update the seed list as well as other parameters you wish.

The problem you're running into is inherent in any distributed row store (Cassandra or Scylla). In order for that to work, a coordinator node needs to contact all other nodes, query them, and assemble the result set. That causes a lot of contention which may prevent some replicas from reporting properly.

I recommend (downloading) using DSBulk for this type of operation. It has a count feature designed just for this purpose.

This is mentioned in the actix_web Changes.md for v4.0:

The client mod was removed. Clients should now use awc directly.

The actix_web::client module has long since largely been a wrapper around the awc crate ever since the 1.0 release, but it seems they now want to separate them entirely.

The types from awc should be near identical to those exposed in prior actix_web releases, however if you're using it with actix_web:4.0 (currently in beta) then you'll want to use awc:3.0 (currently in beta) for compatibility.

There is nothing special with Session, as seen here https://github.com/scylladb/scylla-rust-driver/blob/main/examples/parallel-prepared.rs

The solution is to wrap it in an Arc. We don't necessary need to do so ourselves as web::Data does this itself. Here is what I got.

Option 1, seems to be a good match: I assume each logical object have an unique id (or better uuid) In such a case, you can create something like

Since maps are a type of collection, I think this GH issue pretty much sums it and what's supported (or not). https://github.com/scylladb/scylla/issues/3654#issuecomment-479639450

Validations req mentioned in the GH issue:

• disallow FULL indexes on non-frozen collections
• disallow non-FULL indexes on frozen collections
• disallow repeated creation of FULL indexes on the same collection
• disallow referencing frozen-map entries in the WHERE clause (only allow complete map values)

What we have:

Non-frozen collections are currently not supported at all with proper error message, non-full indexes on frozen collections are also disallowed, repeated creation is by default forbidden for any index too. Also, I think that referencing frozen-map entries in WHERE clause should be allowed, at least with filtering.

If you find any different behavior, please let us know and file a GH issue.

The Client session timeout indicates that the driver is timing out before the server does or - should it be overloaded - that Scylla hasn't replied back the timeout to the driver. There are a couple of ways to figure this out:

1 - Ensure that your default_timeout is higher than Scylla enforced timeouts in /etc/scylla/scylla.yaml

2 - Check the Scylla logs for any sign of overload. If there is, consider throttling your requests to find a balanced sweet spot to ensure they no longer fail. If it continues, consider resizing your instances.

In addition to these, it is worth to mention that your sample code is not using PreparedStatements, TokenAwareness and other best practices as mentioned under https://docs.datastax.com/en/developer/python-driver/3.19/api/cassandra/policies/ that will certainly improve your overall throughput down the road.

You can find further information on Scylla docs: https://docs.scylladb.com/using-scylla/drivers/cql-drivers/scylla-python-driver/ and Scylla University https://university.scylladb.com/courses/using-scylla-drivers/lessons/coding-with-python/