5.0.0-rc3: review if exiting cleanly when running out of RAM is best course of action

[matthewdarwin]

My nodeos ran out of RAM..

Dec 29 03:02:15 wax-snap71c nodeos[2594144]: std::bad_alloc: Error unpacking field transactions
Dec 29 03:02:15 wax-snap71c nodeos[2594144]:     {"field":"transactions","what":"std::bad_alloc"}
Dec 29 03:02:15 wax-snap71c nodeos[2594144]:     net-8  raw.hpp:363 operator()
Dec 29 03:02:15 wax-snap71c nodeos[2594144]: error unpacking eosio::chain::signed_block
Dec 29 03:02:15 wax-snap71c nodeos[2594144]:     {"type":"eosio::chain::signed_block"}
Dec 29 03:02:15 wax-snap71c nodeos[2594144]:     net-8  raw.hpp:668 unpack

and then exited cleanly:

Dec 29 03:05:12 wax-snap71c nodeos[2594144]: info  2023-12-29T03:05:12.914 nodeos    http_plugin.cpp:515           plugin_shutdown      ] exit shutdown
Dec 29 03:05:12 wax-snap71c nodeos[2594144]: info  2023-12-29T03:05:12.916 nodeos    main.cpp:155                  operator()           ] nodeos version v5.0.0wax01-rc3 v5.0.0wax01-rc3-10c04d06485c8a1fdb52f6036dfa90b93813cb3a-dirty
Dec 29 03:05:12 wax-snap71c nodeos[2594144]: info  2023-12-29T03:05:12.916 nodeos    main.cpp:62                   log_non_default_opti ] Non-default options: [.........]
Dec 29 03:05:52 wax-snap71c systemd[1]: nodeos.service: Main process exited, code=killed, status=11/SEGV

I am not convinced that the state of the world is 100% accurate in this situation. Would it be better to not exit cleanly instead?(force user to revert to well known state)

For me, I have rolled back to earlier snapshot to avoid any doubt.

[bhazzard]

Thank you for the suggestion, we will discuss. We won't consider this a blocker for 5.0.0 stable.

[heifner]

In general the strategy is to just exit (non-clean shutdown):

leap/plugins/chain_plugin/chain_plugin.cpp

Lines 1273 to 1277 in 6d52486

	void chain_apis::api_base::handle_bad_alloc() {
	elog("std::bad_alloc - memory exhausted");
	//return -2 -- it's what programs/nodeos/main.cpp reports for std::exception
	std::_Exit(-2);
	}

This must be a case where we are not catching and handling std::bad_alloc appropriately.