folly::thread_cached

A synchronized value with a per‐thread cache for accelerated reads.

Synopsis

Declared in <folly/concurrency/ThreadCachedSynchronized.h>

template<
    typename T,
    typename Mutex = SharedMutex>
class thread_cached_synchronized;

Description

Roughly equivalent to Synchronized, but with a per‐thread cache for acceleration.

Use in hot code when Synchronized alone, with its shared lock and unlock, would be too costly.

Avoid when acceleration is marginal since per‐thread caches are expensive.

Example:

struct writer_and_readers { folly::thread_cached_synchronized<std::shared_ptr<data>> obj_; std::jthread background_writer_{std::bind(loop_update, this)};

std::shared_ptr<data> get_recent_data_fast() { return obj; }

data fetch_recent_data(); bool needs_data_and_not_signaled_done(); void loop_update() { while (needs_data_and_not_signaled_done()) { obj.exchange(folly::copy_to_shared_ptr(fetch_recent_data())); } } };

Note: A singleton variation of this with SingletonThreadLocal would remove one of the branches when looking up the per‐thread cache but would introduce a new branch when looking up the global version.