feat(sdk): fail over to a surviving node when the current one dies

[numinnex]

No description provided.

[numinnex]

Merge after #3934

[codecov]

Codecov Report

❌ Patch coverage is 89.85598% with 162 lines in your changes missing coverage. Please review.
✅ Project coverage is 82.98%. Comparing base (37d7eee) to head (8570d17).

Files with missing lines	Patch %	Lines
core/sdk/src/tcp/tcp_client.rs	89.03%	54 Missing and 23 partials ⚠️
...ache/iggy/client/async/tcp/AsyncIggyTcpClient.java	78.65%	11 Missing and 8 partials ⚠️
...SDK/IggyClient/Implementations/TcpMessageStream.cs	84.31%	11 Missing and 5 partials ⚠️
core/sdk/src/leader_aware.rs	89.25%	13 Missing ⚠️
foreign/go/client/tcp/tcp_core.go	88.18%	10 Missing and 3 partials ⚠️
foreign/node/src/client/client.socket.ts	93.06%	12 Missing ⚠️
foreign/node/src/client/client.connection.ts	98.22%	2 Missing and 1 partial ⚠️
foreign/go/client/tcp/tcp_session_management.go	93.33%	1 Missing and 1 partial ⚠️
...figuration/tcp_config/tcp_client_config_builder.rs	95.65%	0 Missing and 1 partial ⚠️
core/sdk/src/clients/mod.rs	92.85%	1 Missing ⚠️
... and 5 more

Additional details and impacted files

@@             Coverage Diff              @@
##             master    #3944      +/-   ##
============================================
- Coverage     84.03%   82.98%   -1.06%     
- Complexity     1358     1395      +37     
============================================
  Files          1215     1217       +2     
  Lines        170421   166509    -3912     
  Branches     138177   133794    -4383     
============================================
- Hits         143218   138173    -5045     
- Misses        23351    24228     +877     
- Partials       3852     4108     +256     

Components	Coverage Δ
Rust Core	83.55% <89.29%> (-1.38%)	⬇️
Java SDK	67.09% <82.11%> (+0.41%)	⬆️
C# SDK	75.32% <85.96%> (+0.35%)	⬆️
Python SDK	90.10% <ø> (ø)
PHP SDK	84.68% <ø> (-0.98%)	⬇️
Node SDK	95.92% <95.61%> (+0.11%)	⬆️
Go SDK	68.99% <89.28%> (+0.69%)	⬆️

Files with missing lines	Coverage Δ
.../src/traits/binary_impls/personal_access_tokens.rs	100.00% <ø> (ø)
core/common/src/traits/binary_impls/users.rs	100.00% <ø> (ø)
core/common/src/traits/binary_transport.rs	100.00% <100.00%> (ø)
...ypes/configuration/tcp_config/tcp_client_config.rs	100.00% <100.00%> (ø)
core/sdk/src/client_provider.rs	43.62% <100.00%> (-3.48%)	⬇️
...e/sdk/src/clients/binary_personal_access_tokens.rs	100.00% <ø> (ø)
core/sdk/src/clients/binary_users.rs	100.00% <ø> (ø)
core/sdk/src/vsr.rs	93.90% <100.00%> (ø)
...IggyClient/Implementations/TcpMessageStream.Vsr.cs	74.40% <100.00%> (+1.67%)	⬆️
...rg/apache/iggy/client/async/tcp/ReconnectPlan.java	100.00% <ø> (ø)
... and 17 more

... and 165 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[hubcio]

cross-sdk:

• configured seeds (failover_addresses) exist only in rust; go/java/c#/node keep a single seed, so a first connect on a dead configured node has nowhere to go. in scope or follow-up?
• dial order: rust = current, roster, seeds; the four ports = current, configured seed, roster. every port dials a possibly dead configured node ahead of live survivors on every loss. pick one (rust's). same for reconnection disabled: rust bails on the first failed dial, go still sweeps everything.
• StaleClient forget: rust/go/c# forget the remembered credentials, java never does, c# misses the outcome-unknown wrapper, go forgets only after the reconnection-disabled return. same contract in all five or in none.
• empty description for a behaviour change across five sdks (remembered credentials, skipped reestablish_after, the 2s dial bound, replay semantics). write the contract down. the c# readme (InitialDelay, "without credentials a reconnect cannot restore the session") and the python readme (reestablish_after) are now wrong too.

outside the diff:

• rust: IggyClientBuilder::with_tcp() has no with_failover_addresses passthrough, so seeds are only reachable via TcpClient::create(config). max_retries now counts full sweeps but the builder doc still says retries.
• go: currentServerAddress is read without c.mtx in exchange (line 533) while dialCandidate writes it under the lock; time.Sleep(remaining) in the reestablish pause (line 926) ignores ctx. both pre-existing.
• node: the maxRetries doc in client.type.ts says reconnection attempts, it now counts full passes; same for the DefaultReconnectOption comment. the readme has no failover mention.

[numinnex]

/ready

[hubcio]

second pass at 4433f9a. most threads are fixed; what's below and the inline comments are still open.

• staleclient rule: replay the remembered login everywhere, like rust does now. an eviction is not caller intent. go, c# and java (with and without builder credentials) should match.
• PR description is still empty. write the contract down: remembered credentials, the 2s dial bound, the reestablish_after rotation, what is replayed after a reconnect and what is not.
• seeds still exist only in rust, and there only via TcpClient::create(config): no with_failover_addresses on IggyClientBuilder::with_tcp(), no connection string option, cli and python can't set them. in scope or follow-up?
• dial order still differs: rust = current, roster, seeds; go/java/c#/node = current, configured seed, roster, and the ports now pin their order with tests. pick one.
• reconnection disabled: rust, go and c# sweep every candidate once; java and node don't dial at all (maxRetries 0 / enabled = false skips the loop). same rule everywhere.
• docs: c# README lines 104, 111-112, 204-206 still say a hand-run login can't be restored; rust max_retries docs (client_builder.rs:168, tcp_client_config_builder.rs:61) and the python config doc still say retries, one retry is a full sweep now.
• go: currentServerAddress read without c.mtx in exchange (tcp_core.go:542) and clientAddress in tcp_session_management.go:79 - both reported by -race.
• go: concurrent requests failing during a reconnect with credentials hang the whole client. skipAutoLoginOnce is client-global: a second failing request's disconnect() flips the state so two Connects run, one eats the flag and the login replay's own Connect auto-logins under its own registerMtx. 8 concurrent pings on a dropped socket with WithAutoLogin hung until the test timeout. pre-existing for WithAutoLogin, remembered logins reach it now. the second Connect also overwrites c.conn and the first conn is never closed.
• java has no failover dial bound: netty CONNECT_TIMEOUT_MILLIS = connectionTimeout or 3s, always on, tcp only, so a long connectionTimeout stretches every rotation.

[numinnex]

/ready