[1.0-beta4] Add max-reversible-blocks option

[heifner]

New chain_plugin option max-reversible-blocks. Specifies the maximum number of reversible blocks beyond current LIB to allow before the node is gracefully shutdown. The graceful shutdown is provided as a way to prevent out of memory kill by OS. Once consensus is working again the node can be started with a larger max-reversible-blocks value to continue.

  --max-reversible-blocks arg (=3600)   Approximate maximum allowed reversible
                                        blocks before shutdown. Will shut down
                                        if limit reached. Specify 0 to disable.

Resolves #447

[greg7mdp]

I think it might be a good idea to allow the user to disable the check with --max-reversible-blocks 0.

[greg7mdp]

The name is a bit misleading. It is not really max_reversible_blocks, but longest_reversible_branch_allowed.

[heifner]

The name is a bit misleading. It is not really max_reversible_blocks, but longuest_reversible_branch_allowed.

It is not that either. It is max-reversible-blocks-allowed-on-best-branch or max-reversible-blocks for short.

[greg7mdp]

The name is a bit misleading. It is not really max_reversible_blocks, but longuest_reversible_branch_allowed.

It is not that either. It is max-reversible-blocks-allowed-on-best-branch or max-reversible-blocks for short.

Not quite, it is used in create_block_state_i on a block pushed on a branch which may not be the best branch, right?

[heifner]

Not quite, it is used in create_block_state_i on a block pushed on a branch which may not be the best branch, right?

Correct, in that case it could be any branch.

[greg7mdp]

Not quite, it is used in create_block_state_i on a block pushed on a branch which may not be the best branch, right?

Correct, in that case it could be any branch.

So really we cannot create any branch longer than max_reversible_blocks, right? If so, it really is longuest_reversible_branch_allowed.

I feel that max_reversible_blocks implies the total number of blocks in fork_db, which this definitely isn't.

[heifner]

So really we cannot create any branch longer than max_reversible_blocks, right? If so, it really is longuest_reversible_branch_allowed.

I feel that max_reversible_blocks implies the total number of blocks in fork_db, which this definitely isn't.

The description is Approximate maximum allowed reversible blocks. I didn't want to tie down the implementation in case we have reason to change it. Would like to allow this config option to work even if we change to forkdb.size() or fork_db.root()->block_num() + {head or reversible_block_num}

[greg7mdp]

I didn't want to tie down the implementation in case we have reason to change it...

Sounds good, thanks!

[heifner]

I didn't want to tie down the implementation in case we have reason to change it...

Discussed with team and decided to tie the implementation to forkdb.size().

[linh2931]

I don't see where you set it to an internal big number to disable if the user configures it to 0.

[heifner]

https://github.com/AntelopeIO/spring/pull/493/files#diff-42e9f97eed543dd784b1b77a536c50ff8e2403d5bbd3b1fb7f5dd90201391061R4717

[heifner]

Now that I'm not adding the value to reversible_block_num, might be nicer to just set it to std::numeric_limits<uint32_t>::max() if 0.

[heifner]

Ah, but the if > 0 does avoid the mutex if that has any worth. I guess I'll leave it as is.

[ericpassmore]

Note:start
group: PROTOCOL
category: HTTP
summary: Adds new chain_plugin option max-reversible-blocks. Specifies the maximum number of reversible blocks beyond current LIB to allow before the node is gracefully shutdown.
Note:end