[core] Create event log files lazily to avoid stale/empty logs

[nuocbiendang12]

Why are these changes needed?

Fixes #64153
Ray currently creates several event log files eagerly at process startup,
even when they are never written to:

• logs/events/event_CORE_WORKER_<pid>.log — always empty
• logs/events/event_GCS.log — usually empty
• logs/events/event_RAYLET.log — usually empty
• logs/export_events/*.log — empty when new export framework is enabled
  This creates unnecessary file clutter and wastes I/O on every Ray startup.

What do these changes do?

Changes LogEventReporter in src/ray/util/event.cc to use lazy
initialization for the underlying log file. The log file is now only
created/opened on the first call to Report(), not in the constructor.
Key changes:

• Added EnsureSinkInitialized() private helper method
• Store log_sink_key_ as a member variable for deferred initialization
• Modified Flush() to safely handle uninitialized sink
• All existing behavior is preserved when events are actually written

Related issues/PRs

Closes #64153

Checks

• I've signed off all commits with git commit -s
• I've run scripts/format.sh to lint the changes in Python
• I've run clang-format on C++ changes
• All existing tests pass locally
• Added/modified tests to cover the changes (if applicable)

[gemini-code-assist]

Code Review

This pull request introduces lazy initialization for the LogEventReporter's spdlog logger (log_sink_) to prevent empty log files from being created at startup. The review feedback highlights a critical thread-safety issue: because EnsureSinkInitialized() can be called concurrently from multiple threads, the unsynchronized check and initialization of log_sink_ can lead to data races and crashes. The reviewer recommends using std::call_once with a std::once_flag to ensure thread-safe initialization.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

To support thread-safe lazy initialization of log_sink_ using std::call_once, we need to declare a std::once_flag member variable.

  // The underlying spdlog logger. nullptr until the first event is written.
  std::shared_ptr<spdlog::logger> log_sink_;

  // Ensure thread-safe lazy initialization of log_sink_.
  std::once_flag sink_init_once_;

[gemini-code-assist]

In a multi-threaded environment, EventManager::Publish can be called concurrently from multiple threads because it only acquires a reader lock (absl::ReaderMutexLock). This means LogEventReporter::Report (and consequently EnsureSinkInitialized) can be executed concurrently by multiple threads.

Without synchronization, the check if (log_sink_ != nullptr) and the subsequent initialization of log_sink_ (which is a std::shared_ptr) is a data race, leading to undefined behavior. Furthermore, concurrent calls to spdlog::rotating_logger_mt with the same key will throw a spdlog_ex exception, causing the process to crash.

To resolve this, we should use std::call_once with a std::once_flag to ensure that log_sink_ is initialized safely and exactly once.

void LogEventReporter::EnsureSinkInitialized() {
  std::call_once(sink_init_once_, [this]() {
    log_sink_ = spdlog::get(log_sink_key_);
    // If the file size is over {rotate_max_file_size_} MB, this file would be renamed
    // for example event_GCS.0.log, event_GCS.1.log, event_GCS.2.log ...
    // We allow to rotate for {rotate_max_file_num_} times.
    if (log_sink_ == nullptr) {
      log_sink_ = spdlog::rotating_logger_mt(log_sink_key_,
                                             log_dir_ + file_name_,
                                             1048576 * rotate_max_file_size_,
                                             rotate_max_file_num_);
    }
    log_sink_->set_pattern("%v");
  });
}

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 2 potential issues.

^{Reviewed by Cursor Bugbot for commit ea2243f. Configure here.}