TRACY_ON_DEMAND consumes memory infinitely

[FedyuninV]

Hi, @wolfpld!

I'm developing an app which:

1. stays online for a long time (kind of a service)
2. when needed, can be profiled with Tracy on the fly

TRACY_ON_DEMAND was a lifesaver for me but when process lives for a long time the tracy::Profiler::m_deferredQueue grows infinitely and consumes too much memory. Do you have any ideas how to limit this queue? E.g. some smart clean-up which removes the oldest events and guarantees that remaining ones won't cause Discontinuous frame begin/end mismatch..

Right now it's a private field without any mechanism to get access to it or clean it up via Tracy.

P.S. Thanks for the cool tool!

[wolfpld]

The deferred queue should, by design, hold a small number of items needed to properly set up the initial state of things created in the past that newly arriving messages can reference. What kind of things do you see being put there in excess? You can check this by breaking in tracy::Profiler::DeferIteem().

[FedyuninV]

The idea was to launch a service (without connecting to Tracy) and connect to it only when something happens. There is a chance that Tracy server will connect to the profiled process after hours (or even days) of work.

[kekangshu99]

Maybe you can try to start Tracy manually using TRACY_MANUAL_LIFETIME, then shut down Tracy and relaunch it after a while?

[FedyuninV]

These 4 commits together allow us to not use DeferItem and avoid this infinite growth

• ad39a01de1376e0f7d8a5bd82dd0af94ee4fd5e7
• 216a4b7663db330c76210b2dccff6788a5d951c3
• afe5c8dfbed060ff0bfe584af028175df6fc4a03
• d3da7dcecdd631ee9cfa05d175ef96cd9f92e377