Drain the transaction overlay

[dvdplm]

When restoring a chunk during warp sync, drain the transaction overlay.
Fixes memory blowup seen after https://github.com/openethereum/openethereum/pull/11589 landed.

There's a discussion to be had about the semantics of the implementations of inject() and its docs. It can have significant side-effects beyond injecting independent extra database ops into a batch of writes.

[ordian]

Nice catch! Sorry I totally missed that in https://github.com/openethereum/openethereum/pull/11589#discussion_r408067341. Submitting an empty transaction to drain the transaction_overlay does def sound hacky, create an issue for it?

[dvdplm]

Submitting an empty transaction to drain the transaction_overlay does def sound hacky

It's pretty odd indeed. Is there no other way to do it?

[adria0]

When restoring a chunk during warp sync, drain the transaction overlay.
Fixes memory blowup seen after #11589 landed.

There's a discussion to be had about the semantics of the implementations of inject() and its docs. It can have significant side-effects beyond injecting independent extra database ops into a batch of writes.

so maybe inject deserves an unsafe tag?

[denisgranha]

based on the execution of this PR synchronizing mainnet, I can see that the memory leak is gone