Fixed - check synced slaves in RTransaction commit #6292 - #7392

Thanks for the review and detailed feedback @mrniko!

Addressed all items in commit 8d9e69a:

1. Multi-Shard Cluster Verification:

   • In CommandBatchService (near L339 and L775), tracked minSyncedSlaves across all WAIT/WAITAOF commands in the batch and exposed getMinSyncedSlaves() on BatchResult while preserving getSyncedSlaves() as the sum for backwards compatibility.
   • The transaction check verifies that every shard individually reached the quorum (result.getMinSyncedSlaves() >= options.getSyncSlaves()).

2. Default Options Compatibility:

   • Switched the check to evaluate options.getSyncSlaves() > 0 rather than batchOptions.getSyncSlaves() > 0.
   • Default transactions (options.getSyncSlaves() == 0) retain their original best-effort WAIT behavior and will never fail if a replica is slow or unacknowledged.

3. Dedicated Exception Subclass & Documentation:

   • Introduced TransactionSyncException extends TransactionException so application code and transaction managers can distinguish between an uncommitted/rolled-back transaction failure and a master-committed transaction where replica acknowledgment timed out.
   • Formatted the exception message to indicate exact counts: Transaction committed, but only %d of %d replicas acknowledged within %d ms.
   • Updated syncSlaves Javadoc on TransactionOptions and documented TransactionSyncException behavior in docs/transactions.md.

4. Integration Test Suite:

   • Extended RedissonTransactionSyncSlavesTest with RedisDockerTest verifying:
     • commit() throws TransactionSyncException with the exact count message when replica sync cannot be satisfied.
     • The master node still retains the written data after the exception.
     • commitAsync() similarly propagates TransactionSyncException.
     • Default transactions with no replica sync configured commit cleanly without throwing.

5. Code Hygiene:

   • Restored createOptions() to private.
   • Consolidated the verification logic into a single private checkSyncSlaves(BatchResult<?>) helper used by both sync and async commit paths.

Thanks @mrniko! Addressed all three items in commit 3c1f409:

1. Cluster tests (testInCluster): Added testSyncSlavesInCluster and testDefaultOptionsPassInCluster in RedissonTransactionSyncSlavesTest. testSyncSlavesInCluster writes keys across multiple master shards (test1 and test2) with syncSlaves(2, ...). Since each cluster master has 1 replica, the sum is 2 but the minimum is 1, verifying that TransactionSyncException is thrown and guarding against regressions that might switch back to getSyncedSlaves(). Also tested that syncSlaves(1, ...) and default options pass cleanly.
2. Javadoc location: Moved the @throws TransactionSyncException tag from TransactionOptions.syncSlaves() to RTransaction.commit() and commitAsync().
3. Docs example: Updated the example in docs/transactions.md to .syncSlaves(0, 5, TimeUnit.SECONDS).

Thanks @mrniko! Fixed both items in commit fbd8b01:

1. Wait Command Completion: In CommandBatchService.executeRedisBasedQueue(), updated the check from RedisCommands.WAIT.getName().equals(lastCommand.getCommand().getName()) to lastCommand != null && isWaitCommand(lastCommand), ensuring that both WAIT and WAITAOF future promises are awaited before batch completion.
2. Cluster Default Options Sync Check: In RedissonTransaction.createOptions(), initialized syncSlaves = 0 by default and set to availableSlaves only when options.getSyncSlaves() == 0 (ensuring that non-syncing commits never trigger false failures when replicas acknowledge asynchronously).
3. Cluster Test Assertion: Refined testSyncSlavesInCluster exception assertion to match [01] of 2 replicas acknowledged within 1000 ms accommodating varying cluster slave sync delays.

Thanks @mrniko!

The failures shown in the test runner were caused by:

1. testDefaultOptionsPassInCluster timeout (~15s): in createOptions(), options.getSyncSlaves() == 0 previously read entry.getAvailableSlaves(), which is -1 before an explicit info probe, causing BatchOptions to send invalid WAIT parameters and wait for the full response timeout. Fixed in commit 9223040 by setting syncSlaves = 0 whenever options.getSyncSlaves() <= 0, ensuring default transactions never send unnecessary WAIT commands or encounter sync timeouts.
2. testSyncSlavesInCluster: Wrapped cluster transaction operations in try...finally blocks to ensure keys are always cleanly unpinned and cleaned up across test assertions even if a previous commit throws an expected TransactionSyncException.