GeezerVison

netdata_logger_with_limit() checked the rate-limit window under the limiter's spinlock, but wrote last_logged and reset count only after logging, outside the lock. Threads sharing one limiter (nd_log_limit_static_global_var) could all pass the check before any of them moved the window, so a flood emitted duplicates; an attempt counted while another thread was logging was also erased by the late reset.

The window is now reserved (and the count reset) before unlocking; the log call still runs outside the lock. Side effect: the next window starts when a message is admitted rather than when its logging finishes.

Tested on beast with a test-only harness (16 threads logging through one shared limiter at the same instant, 300 rounds, 3 repeats):

• master: 544 / 610 / 779 lines instead of 300, duplicates in 112-229 of 300 rounds
• fix: exactly 300 lines every time, no duplicates
• -W unittest passes

Note: draft #24006 splits this function into nd_log_limit_admit() / nd_log_limit_emitted(); on rebase, the window reservation belongs inside nd_log_limit_admit() (under the lock) and nd_log_limit_emitted() is no longer needed.

Summary by CodeRabbit

• Bug Fixes
  • Improved the consistency of rate-limited logging when multiple events are handled concurrently, helping keep log limits accurate.