Add PSRAM-based NVS caching

[0x-fran]

This PR introduces a RAM-based caching layer for configuration values stored in NVS, along with a batched flush mechanism to reduce latency and flash wear.

Summary of changes:

• All Config::get*() calls now return cached values if available, avoiding repeated flash reads.
• All Config::set*() calls only modify the RAM cache and mark entries as dirty.
• A central flush_nvs_changes() function writes only dirty entries to NVS in a single transaction.
• API endpoints now call flush_nvs_changes() after updates.
• A background timer (every 5 minutes) flushes dirty values if needed.
• System restart and OTA update paths also flush explicitly before rebooting.
• Debug logging has been added to track cache hits (ESP_LOGD).

Benefits:

• Significantly reduces flash read and write operations, especially during high-frequency polling (e.g. UI requests).
• Minimizes blocking flash access in GET requests like /api/system/info.
• Reduces flash wear and improves overall responsiveness.

No configuration values or behavior are affected unless they are written via the API or auto-flush is triggered.

[shufps]

thx a lot!

The idea is good and I (mostly) like the implementation.

A couple of things I noticed on first glance:

• it's using internal memory, maybe we could shift it to PSRAM and use the ALLOC/CALLOC macros for it. I'm always eagerly behind reducing internal memory where possible - this would work with PSRAM too imho
• I think we don't need a periodic flusher - we know exactly when values change. It only can happen in the HTTP handler to store new values
• It's not thread safe, imagine the cache being flushed but accessed through another thread concurrently

The config is a difficult topic at a whole I know. Complete mess 😅

I also tried - once at least - to refactor everything into a class with load and save and a member variable per field but it was a mess too.

I like the map approach though 🤔

What also would be possible is to cache the "rendered" json response from the HTTP API for a short period of time, it would reduce CPU load and NVS access too and would be maybe a little bit less "invasive".

Not sure though

[0x-fran]

Thanks for the quick feedback 🙌

• Config cache now uses PSRAM via heap_caps_calloc() to save internal memory
• All access is mutex-protected (simple FreeRTOS semaphore)
• Removed periodic flusher; changes are flushed explicitly only when needed
• Added optional short-lived JSON cache for /api/system/info (default: 15s)

PS: I added some ESP_LOGI lines (commented with TEMP: Debug) in system_handler.cpp for debugging the JSON caching – feel free to remove them if you'd prefer it clean.

[0x-fran]

I realized the last commit missed a few key points 🙈
This commit improves upon the initial version by adding stricter NULL handling, correct PSRAM usage with heap_caps_strdup(), and splitting NVS flushing into a non-blocking working copy.

Summary:

• handles potential NULL values more robustly (fallbacks always return valid strings)
• all strings are now consistently allocated in PSRAM (heap_caps_strdup)
• flush_nvs_changes() copies dirty entries into a temporary working buffer to avoid holding the mutex during I/O
• dump_cache_status() added for optional diagnostics (currently unused)

[0x-fran]

Pushed too quickly earlier 🙈 Sorry 🙌
This commit fixes broken return types, ensures proper heap_caps_strdup() use for PSRAM allocation, and improves NULL handling in get_string. compiles now cleanly and behave as expected.

[0x-fran]

🙌

@shufps : Pls. let me know if you’d prefer a new PR instead.

Changes:

• removed JSON Caching due to weird behavior -> 1c65567
• Enhanced thread-safety and added type mismatch checks in setters.
• Improved error handling across all getters/setters (e.g., for PSRAM allocation).
• Added detailed per-entry logging in dump_cache_status (incl. type information).
• Optimized flush_nvs_changes with proper memory management for dirty entries

Sample logoutput:

₿ I (16671) nvs_config: Cache miss for string key: influx_org. Reading from NVS.
₿ I (16681) nvs_config: Cache miss for string key: influx_prefix. Reading from NVS.
₿ I (34121) nvs_config: Scanning for dirty entries to flush.
₿ I (34121) nvs_config: Found 25 dirty entries. Flushing to NVS...
₿ I (34141) nvs_config: Configuration changes flushed to NVS.
₿ I (34141) http_system: Configuration changes flushed to NVS

PS: Performance improvement using PSRAM cache about ~400x faster than NVS 🚀🥳

[kukulle-hood]

Hi, is this subject still in the loop of attention?
Happy Sunday to all of you.

[0x-fran]

Rebased the branch onto latest develop and resolved all conflicts manually (including nvs_config.cpp).
Final adjustments were made to the NVS caching logic after testing.

The branch should now be ready for the next release @shufps 😉