Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Remove vm accesses to stateManager trie and cache #376

Merged
merged 1 commit into from
Nov 6, 2018

Conversation

mattdean-digicatapult
Copy link
Contributor

This is part of the broader discussion in #268 and #309 so readers should probably start there.

This PR removes all direct accesses, from the VM functions, to the cache and trie properties of stateManager. This is essentially done by replacing calls to these properties to the equivalent stateManager functions.

There are potentially two more controversial change here. The first is the removal of the loadCompiled method on the VM. This method is unused internally and undocumented. It is problematic to keep this function as it interferes with checkpoint/commit structure used in stateManager. My assumption is that if someone desires this functionality they would interact with the stateManager directly prior to instantiating the vm. The second is the change of the step event which now exposes the stateManager instance rather than the cache. This is essentially a breaking change as the event object structure is documented.

@holgerd77
Copy link
Member

Hi Matthew, cool, I did a complex (or at least broadly scoped) API doc refactoring PR merge #377 in between and also merged the StateManager work of @danjm #375. Can you do a careful rebase on top of this?

Then I think we get your stuff merged relatively quickly.

Cheers, Holger (relaxing a bit at the hotel :-))

@mattdean-digicatapult mattdean-digicatapult force-pushed the remove-vm-accesses-to-statemanager-trie-cache branch from b08d5cb to 26e6541 Compare November 1, 2018 10:51
@mattdean-digicatapult
Copy link
Contributor Author

I've rebased and updated the docs to include the change to the step event format. Hopefully I've done this right 🤞

@coveralls
Copy link

coveralls commented Nov 1, 2018

Coverage Status

Coverage increased (+0.2%) to 90.62% when pulling d9cca33 on remove-vm-accesses-to-statemanager-trie-cache into 7390207 on master.

@mattdean-digicatapult
Copy link
Contributor Author

Hold off reviewing this for now as it's causing extra failing blockchain tests. Will investigate

@holgerd77 holgerd77 force-pushed the remove-vm-accesses-to-statemanager-trie-cache branch from 26e6541 to d759d35 Compare November 2, 2018 17:26
@holgerd77
Copy link
Member

Rebased this.

var trie = opts.state || new Trie()
if (opts.activatePrecompiles) {
trie = new Trie()
for (var i = 1; i <= 8; i++) {
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Totally irritated, was this one precompile forgotten before and this is an implicit fix (you are iterating up to 8 and not to 7)?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

looks that way...

@holgerd77
Copy link
Member

Additionally failing tests:

OOGStateCopyContainingDeletedContract, 542, 544
suicideCoinbase, 562, 564
suicideStorageCheck, 567, 568
suicideStorageCheckVCreate, 571, 572
RecallSuicidedContractInOneBlock, 775, 776

@@ -217,13 +227,11 @@ module.exports = function (opts, cb) {
if (ethUtil.bufferToInt(block.header.gasUsed) !== Number(gasUsed)) {
err = new Error((err || '') + 'invalid gasUsed ')
}
if (self.stateManager.trie.root.toString('hex') !== block.header.stateRoot.toString('hex')) {
if (stateRoot.toString('hex') !== block.header.stateRoot.toString('hex')) {
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Extra failures are happening here, not sure why yet, probably you'll get this faster than me.

When I run node ./tests/tester.js -b --file='suicideStorageCheck', looking at the first test and and add console output for the first and second state root value, I get:

211aca80b1d1084fd9a18fda2afbc5bdf3866d06c5321660e105546e79fded8d
334d8d12b70ab372703f37bf6c9fed76c7c61ba1f131517123d5d440776d825a

On the old version I get both:

334d8d12b70ab372703f37bf6c9fed76c7c61ba1f131517123d5d440776d825a
334d8d12b70ab372703f37bf6c9fed76c7c61ba1f131517123d5d440776d825a

@mattdean-digicatapult mattdean-digicatapult force-pushed the remove-vm-accesses-to-statemanager-trie-cache branch from d759d35 to 80c1c34 Compare November 5, 2018 09:09
@holgerd77 holgerd77 mentioned this pull request Nov 5, 2018
10 tasks
@mattdean-digicatapult
Copy link
Contributor Author

Right we're back to 50 failing blockchain tests. I'm not quite happy that I understand quite why what I've changed works, but I have an inkling. I'd recommend we hold off of merging until I've done a bit more digging

@holgerd77
Copy link
Member

Ok, no rush here.

@mattdean-digicatapult mattdean-digicatapult force-pushed the remove-vm-accesses-to-statemanager-trie-cache branch 2 times, most recently from f3eef12 to 86d2b05 Compare November 5, 2018 14:05
@mattdean-digicatapult
Copy link
Contributor Author

So after a bit of investigating, I found that the changes introduced in #375 were not particularly compatible with the work I've been doing on #309. Essentially the issue is that the cache doesn't handle checkpoints, in my mind, correctly when it comes to deletes. My expectation is that the cache should, in the case of a checkpoint revert, restore the state of deletes at the time of the corresponding checkpoint. This was not happening, but by assuming that deletes only occur at particular times things worked out. What broke is that the cache needs to flush at the end of the transaction in order to clear touched accounts correctly. These changes prevented that.

My additional changes make it so that the cache will, on a revert, restore the original delete set. This is done in much the same way that the modified accounts set is maintained, i.e. on the cache value. This removes the _getAccountPure method on stateManager as it no longer matters if the empty account is added to the cache during the accountIsEmpty call. I've also made clearer the logic in stateManager to flush the cache when the checkpoint stack reaches zero. If anyone doesn't like this logic my alternative would be to add a flush method to stateManager but I don't feel this is necessary as we only need to flush when leaving a user initiated vm method.

This has made this PR, once again, larger than I would like. The alternative would be to split this up to first change the cache behaviour and then complete this PR with it's original intention.

@holgerd77
Copy link
Member

I now merged the CREATE2 work from @rmeissner, if you want to do one last rebase I think we can get this merged fairly quickly.

@mattdean-digicatapult mattdean-digicatapult force-pushed the remove-vm-accesses-to-statemanager-trie-cache branch from c399832 to d9cca33 Compare November 6, 2018 16:34
@mattdean-digicatapult
Copy link
Contributor Author

Rebased as requested

Copy link
Member

@holgerd77 holgerd77 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ok, looks good, thanks again Matthew!

@@ -6,13 +6,12 @@ const async = require('async')
var Cache = module.exports = function (trie) {
this._cache = Tree()
this._checkpoints = []
this._deletes = []
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This refactoring seems to make very much sense, to remove this separate delete stack and put it in the values itself.

runState.stateManager.putAccount(runState.address, runState.contract, function (err) {
if (err) return cb(err)
runState._vm.runCall(callOptions, parseCallResults)
})
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ok.

// credit all block rewards
if (generateStateRoot) {
block.header.stateRoot = stateRoot
}
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ok.

* @property {Account} account the [`Account`](https://github.com/ethereum/ethereumjs-account) which owns the code running
* @property {Buffer} address the address of the `account`
* @property {Number} depth the current number of calls deep the contract is
* @property {Buffer} memory the memory of the VM as a `buffer`
* @property {FunctionalRedBlackTree} cache the account cache. Contains all the accounts loaded from the trie. It is an instance of [functional red black tree](https://www.npmjs.com/package/functional-red-black-tree)
* @property {StateManager} storageManager a state manager instance (EXPERIMENTAL - unstable API)
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Good that this aligned documentation is already used and further evolved and corrected, really glad that I once went through this hazzle, even if there might be still some glizzles every here and there to be corrected.

self.stateManager.commit(function (err) {
cb(err, results)
})
}
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ok.

self.stateManager.cache.put(block.header.coinbase, minerAccount)
self.stateManager.putAccount(block.header.coinbase, minerAccount, next)
} else {
next()
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ok.


function cleanTouched (done) {
self.stateManager.cleanupTouchedAccounts(done)
}
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ok.

})
})
}

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is evolving into a really clean and easy-to-grasp well-named API, nice! 👍

@holgerd77 holgerd77 merged commit 84db45e into master Nov 6, 2018
@holgerd77 holgerd77 deleted the remove-vm-accesses-to-statemanager-trie-cache branch November 6, 2018 22:46
@holgerd77
Copy link
Member

Have re-read your initial comments, think changes to loadCompiled and step event should be unproblematic.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

4 participants