Fix throttling using default limits instead of per-user limits is rare cases

There was a bug, let's simulate it step-by-step:
1. Chunk service is initialized and started. This means that default limits for throttlers are known. (Note that Hydra is not ready yet).
2. A request arrives, it becomes authenticated, proceeds to `HandleAuthenticatedRequest`. Here we choose which request queue it should be put in (`GetRequestQueue`).
3. The queue is chosen correctly and is attempting to be reconfigured with ReconfigureQueue. Long story short, a callback created by `CreateReconfigurationCallback` is called.
4. Here's the catch: the Hydra is still not active, but the callback is attempting to be invoked in the automaton invoker - `GetGuardedAutomatonInvoker(EAutomatonThreadQueue::ChunkService)`. Unfortunately before Hydra is fully active the invoker is not set up properly, which leads to a drop of a callback. (See `TDecoratedAutomaton::TGuardedUserInvoker`, `Invoke`).
5. Now the queue considers itself to be "reconfigured" while in reality it is default-initialized.

This PR should fix it.
commit_hash:5b9abfd2bfbce0fc668339d8981e66f77909b887

0 files changed, 0 insertions, 0 deletions