feature(packaging): debian and rpm packaging

[CodeLieutenant]

Adding Debian and RPM Packaging

Tasks

Preview Give feedback

1. RPM
2. Debian
3. Github Action on release to publish .deb and .rpm

[mykaul]

What's the value of packaging it, if a container works great?

[CodeLieutenant]

What's the value of packaging it, if a container works great?

@mykaul We need it without a container, container works great, but we want to test it on bare metal.
It's bad for performance testing, just to avoid the overhead that we have with containers,
and also customers want to run it without docker. Cassandra stress was supplied with .deb and .rpm before, in scylla-tools-java so we are bringing the support back

[mykaul]

What's the value of packaging it, if a container works great?

@mykaul We need it without a container, container works great, but we want to test it on bare metal. It's bad for performance testing, just to avoid the overhead that we have with containers, and also customers want to run it without docker. Cassandra stress was supplied with .deb and .rpm before, in scylla-tools-java so we are bringing the support back

This isn't true. Containers work at 99-99.9% of the performance of bare metal.

I haven't heard of a customer asking to run non-containerized cassandra-stress. They may want to, they can use older versions that are still available. We have more important tasks (IMHO) to focus on that we can bring value to customers with.
(Random examples - move to newer Java - which will be faster most likely; move away from cassandra-stress to newer tools, using more modern drivers and architecture; extend c-s with more feature to provide more realistic workload simulations).

[fruch]

What's the value of packaging it, if a container works great?

@mykaul We need it without a container, container works great, but we want to test it on bare metal. It's bad for performance testing, just to avoid the overhead that we have with containers, and also customers want to run it without docker. Cassandra stress was supplied with .deb and .rpm before, in scylla-tools-java so we are bringing the support back

This isn't true. Containers work at 99-99.9% of the performance of bare metal.

It's close, if you use the correct options
Our throughput test so show a difference, and we still keeps the option of running without docker cause of it.

I haven't heard of a customer asking to run non-containerized cassandra-stress. They may want to, they can use older versions that are still available. We have more important tasks (IMHO) to focus on that we can bring value to customers with.

You didn't heard, cause there isn't any release without it yet.

(Random examples - move to newer Java - which will be faster most likely; move away from cassandra-stress to newer tools, using more modern drivers and architecture; extend c-s with more feature to provide more realistic workload simulations).

Not sure how those examples are counter to have other delivery option but docker ?

I can give one actual example, I know of a user (inside Scylla) that was given the docker image, he extracted the jar files from it, and replace them with installed c-s version he had.

I think it's a bit too soon to assume what users need in that regards.

We as users in SCT of c-s stress, need a debian package as well.

[mykaul]

Not sure how those examples are counter to have other delivery option but docker ?

Matter of priorities. If we do this, we don't do them.

[fruch]

Not sure how those examples are counter to have other delivery option but docker ?

Matter of priorities. If we do this, we don't do them.

there is not list of those other things you mentioned, hence they are not taken into any priority making

[fruch]

@CodeLieutenant

I've tried this branch, and I couldn't build a debian package out of it.

meanwhile we lower the priority of this one, since we don't need it as urgently in SCT as before.

[CodeLieutenant]

@CodeLieutenant

I've tried this branch, and I couldn't build a debian package out of it.

meanwhile we lower the priority of this one, since we don't need it as urgently in SCT as before.

I just didn't push the latest changes from my machine to the GH, sorry about that, it's working well.

• https://github.com/scylladb/cassandra-stress/releases/tag/v3.17.1 this is the build from GH Actions with uploaded .deb and .rpm packages.

I dont think there is need to pause this, cause it's finished, we can pause the other stuff, testing in SCT etc, but the core building is working

[fruch]

that's a bit limiting, i.e. if one choose different distro of java, it won't be able to install ?

[CodeLieutenant]

Yes it is, but this is what I consider the official supported version of JRE, with which I'm testing locally, there are 3 more which are good candidates: Azul, AWS Correto and Oracle distribution, we can add them if there is a need for it