Refactor CDeltaWriter::addToQueue to use async writer thread for limit-triggered saves

[Copilot]

When transaction queue or memory limits are exceeded, CDeltaWriter::addToQueue was performing inline synchronous saves with complex lock orchestration. This refactoring delegates all saves to the async writer thread for consistency.

Changes

Added belowLimitsSem semaphore

• Signals when queue drops below limits after async writer completes save
• Enables waiting threads to resume when capacity is available

Modified addToQueue method

• Replaced inline save(pending) with requestAsyncWrite() + semaphore wait
• Removed blockedSaveCrit and inline lock management from limit-triggered path
• Added 10s timeout loop with warning for long waits

Updated threadmain method

• Signals belowLimitsSem after save when queue is below both transactionMaxMem and transactionQueueLimit

Updated stop() method

• Signals belowLimitsSem to prevent deadlock on shutdown

Before/After

// Before: inline synchronous save
if (needsSyncSave) {
    CHECKEDCRITICALBLOCK(blockedSaveCrit, fakeCritTimeout);
    CriticalBlock b(pendingCrit);
    if (items && save(pending)) {
        pendingSz = 0;
        // ...
    }
}

// After: trigger async and wait
if (needsSyncSave) {
    {
        CriticalBlock b(pendingCrit);
        if (!writeRequested)
            requestAsyncWrite();
    }
    while (!belowLimitsSem.wait(10000)) {
        if (aborted) return;
        WARNLOG("Waiting on CDeltaWriter to clear queue below limits");
    }
}

All delta writes now flow through the async writer thread, eliminating dual save paths and simplifying lock ordering requirements.

Original prompt

Look at dali/base/dasds.cpp, and the CDeltaWriter implementation, and in particular the addToQueue handling when needsSyncSave is triggered.
Instead of handling the save inline in this function, it should be refactored so it triggers a save (as it would if less than thresholds) and then wait on a Semaphore until writing thread (CDeltaWriter::threadmain) clears down below the limits and signals it.

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.