VSR cluster support for iggy: status of SDKs, delivery plan

[hubcio]

hey all, quick update on the VSR (viewstamped replication) work since we're getting close to wrapping it up.

server side is largely done: server-ng runs a full vsr cluster with view changes, state transfer, leader forwarding and client session resume across restarts. what remains is mostly getting every SDK to speak the VSR wire protocol and pass its test suite against the new server.

where each sdk stands:

• rust: done, available behind the --features vsr flag. bdd suite already runs against iggy-server-ng with vsr (test(bdd): run Rust BDD against iggy-server-ng with VSR #3732)
• python: done, first sdk to fully switch - after feat(python): run Python SDK tests only against the VSR server #3838 all python tests in ci run only against iggy-server-ng with vsr hardcoded
• go: under review in feat(go): migrate SDK to the VSR wire protocol #3834, full migration to the vsr wire protocol
• node: done, framing landed in feat(js): add VSR framing to the Node SDK #3763 plus tls/connection fixes in fix(node): TLS + VSR connection support #3813
• java: wip by @mmodzelewski. cluster metadata and leader redirection already landed in feat(java): add cluster metadata and leader redirection to TCP client #3745
• c#: under review in feat(csharp): migrate to VSR wire protocol #3800 by @lukaszzborek , vsr framing
• c++ and php: prs coming very soon (maybe today). both call the rust bindings underneath, so it's mostly plumbing. they'll switch to iggy-server-ng the same way python did

the plan from here:

1. make all sdk test suites run against iggy-server-ng, one by one. python and golang are first, php and cpp follow, the rest as their PRs land
2. once ALL sdks are done, we remove core/server, rename core/server-ng to core/server and the binary goes back to being called iggy-server

from a contributor perspective not much changes: you build with cargo build --features vsr and start iggy-server-ng instead of iggy-server. that's it. and AFTER the rename even that difference goes away.

if you maintain or use one of these SDKs, reviews and testing are very welcome. my guesstimate is that it should all be merged within one or two weeks. apache release will follow afterwards.

[mmodzelewski]

Java SDK PR is currently under review: #3841

[hubcio]

quick follow-up, 3 days later: all remaining SDK PRs are merged.

• go: merged in feat(go): migrate SDK to the VSR wire protocol #3834
• java: TCP client speaks the VSR wire protocol, merged in feat(java): migrate the TCP client to the VSR wire protocol #3841 (thanks @mmodzelewski)
• c#: merged in feat(csharp): migrate to VSR wire protocol #3800 (thanks @lukaszzborek)
• node: classic framing removed, tests run only against the VSR server (feat(node): drop classic framing and test only against the VSR server #3843)
• c++ (feat(cpp): run C++ SDK tests only against the VSR server #3844) and php (feat(php): run PHP SDK tests only against the VSR server #3845): both suites switched to iggy-server-ng, same way python did in feat(python): run Python SDK tests only against the VSR server #3838

with that, step 1 of the plan is complete: every foreign SDK test suite in CI now runs exclusively against iggy-server-ng with vsr enabled (rust still covers both servers while the legacy one is in the tree).

a few server-side fixes landed along the way while chasing green suites: view changes no longer discard committed ops (#3824, thanks @krishvishal), redundant partition rebuilds no longer reset offsets (#3846), SDK-visible behavior gaps closed (#3828) and vsr test flakes fixed (#3833, #3854) by @numinnex, non-Linux builds fixed (#3850), plus a rust SDK heartbeat lifecycle fix (#3852).

next: step 2 - remove core/server, rename server-ng to server and the binary goes back to being iggy-server.

PR soon, apache release probably next week.