[3.2] log integrity hash on start/stop with integrity-hash-on-start & integrity-hash-on-stop options #395
Conversation
libraries/chain/controller.cpp
Outdated
| @@ -723,6 +724,9 @@ struct controller_impl { | |||
| ~controller_impl() { | |||
| thread_pool.stop(); | |||
| pending.reset(); | |||
| //using the presence of a row in datebase_header index to indicate controller had a successful startup() | |||
There was a problem hiding this comment.
eh something better?
There was a problem hiding this comment.
Move the chain_plugin where you know if started successfully?
There was a problem hiding this comment.
I can't log the startup in chain_plugin because I need to to do the log before replay but after snapshot is loaded. So, it felt weird to do the startup log in controller but the shutdown log in chain_plugin.
I'm not sure doing it in chain_plugin really helps either since there are so many exceptions that can start unrolling the world at any time.
Actually I'm not really even sure this here works well enough: what if an exception is thrown half way through loading a snapshot. Then once everything unrolls the dtor is fired and if datebase_header made it then it does an integrity computation with a half initialized state? That's what I want to avoid.
But I also wanted to avoid adding a bool yep_controller_started_up. It's kinda wild there isn't a clear indicator anywhere 🤔
There was a problem hiding this comment.
In chain_plugin::plugin_startup you could add a catch(...) after the database_guard_exception catch and add a my->chain.reset();. Then in chain_plugin::plugin_shutdown() you could check my->chain to know if it is valid, if so then log the integrity.
There was a problem hiding this comment.
I capitulated and just added a bool 😕
|
Also for 3.1? |
Certainly a bit of a stretch by the typical playbook but maybe we can have that discussion since it's such a few line difference. |
Unequivocally logging the integrity hash after loading a snapshot increases start up time from a snapshot by nearly 33%. Most users probably don't even notice the log; let's not always do that.
Log integrity hash on start up and shut down via new
integrity-hash-on-start&integrity-hash-on-stopoptions.