-
Notifications
You must be signed in to change notification settings - Fork 3.7k
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
Add SMT store type #8507
Add SMT store type #8507
Conversation
SMT (Sparse Merkle Tree) is intended to replace IAVL. New type implements same interfaces as iavl.Store.
Sparse Merkle Tree does not support iteration over keys in order. To provide drop-in replacement for IAVL, Iterator and ReverseIterator has to be implemented. SMT Store implementation use the underlying KV store to: - maintain a list of keys (under a prefix) - iterate over a keys Values are stored only in SMT.
@robert-zaremba is it okay to merge this as is for now? |
Yes, all changes must go to |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
left few comments.
store/smt/iterator.go
Outdated
|
||
var ( | ||
indexPrefix = []byte("smt-ordering-idx-") | ||
afterIndex = []byte("smt-ordering-idx.") // '.' is next after '-' in ASCII |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
would it be more beneficial to have the prefix keys shorter?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I expect this to be compressed away by DB.
Single-byte, numerical value would be much cleaner.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
yes, let's use single byte values.
store/smt/iterator.go
Outdated
} | ||
|
||
func plainKey(key []byte) []byte { | ||
return key[len(indexPrefix):] |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
let's store cache len(indexPrefix)
in a module scope.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
If prefix is single-byte, we can even hardcode this ;)
assert.Equal(t, p.key, iter.Key()) | ||
assert.Equal(t, p.val, iter.Value()) | ||
iter.Next() | ||
} |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I thought that LL SMT is hashing the keys before inserting, so how possible it preserves a pre-image order?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Here, we're testing cosmos-sdk SMT store, not LL SMT data structure. I'm using ordering from key->value mapping (https://github.com/cosmos/cosmos-sdk/pull/8507/files/6fd860d514daa4c9caf55df383d81b1a1e671ede#diff-0d018f820578ed2dd08d3bd1cb6f854aead4e2b32333b15105743eecb21bba77R41).
store/smt/iterator_test.go
Outdated
for _, p := range pairs { | ||
require.True(t, iter.Valid()) | ||
assert.Equal(t, p.key, iter.Key()) | ||
assert.Equal(t, p.val, iter.Value()) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
we should use require
here. It will make lot of error messages if something is wrong and it will be messy. Assert doesn't stop the exectuion.
Tip: I'm usually declaring at the top of the function:
require := require.New(t) // works with assert as well
Then in the function I don't need to pass t
any more: require.Equal(a, b)
.
store/smt/store.go
Outdated
func (s *Store) Get(key []byte) []byte { | ||
defer telemetry.MeasureSince(time.Now(), "store", "smt", "get") | ||
s.mtx.RLock() | ||
defer s.mtx.RUnlock() |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
why do we need to lock? Isn't the underlying DB solving this?
store/smt/store.go
Outdated
defer telemetry.MeasureSince(time.Now(), "store", "smt", "get") | ||
s.mtx.RLock() | ||
defer s.mtx.RUnlock() | ||
val, err := s.tree.Get(key) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
why do we query data through a tree, rather than directly from the DB?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This is probably leftover from version with only SMT tree (no plain-kv).
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
DB is used only for iteration in this version. It was implemented before finalization of ADR-040 and deciding that values goes to plain-KV.
store/smt/store.go
Outdated
if err != nil { | ||
panic(err.Error()) | ||
} | ||
s.mtx.Unlock() |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
we should use defer s.mtx.Unlock()
right after the Lock.
Unfortunately SDK is using a bad pattern to panic to quickly wind up the stack, rather than passing errors. So, we don't want to leave the store locked in case of panic.
shall we update the PR and merge into the branch? |
lets merge and follow up |
Not sure we want to add more new features into this release. |
This would be merged into a feature branch |
@tzdybal , there are few small changes to do in this PR. Will you like to finish it and merge? |
Yes, I'll fix this today. |
Hi @tzdybal - if you need a help let me know - I can clean up this PR for you and merge it. |
can you take over it please. The LL team is busy at the moment, dont want to add more to their plate. |
Sorry for late reply and fix. I addressed all review comments. |
SMT stores: * key -> hash(key, value) KV store stores: * key->value in "bucket 1", * hash(key, value) -> key in "bucket 2".
OK, so I changed the implementation to use ADR-040 data layout for KV and SMT. Please re-review @robert-zaremba, @marbar3778. |
@tzdybal - thanks for handling this, let me merge it so we can move forward. |
* Initial SMT store type SMT (Sparse Merkle Tree) is intended to replace IAVL. New type implements same interfaces as iavl.Store. * Add iteration support to SMT Sparse Merkle Tree does not support iteration over keys in order. To provide drop-in replacement for IAVL, Iterator and ReverseIterator has to be implemented. SMT Store implementation use the underlying KV store to: - maintain a list of keys (under a prefix) - iterate over a keys Values are stored only in SMT. * Migrate to smt v0.1.1 * Extra test for SMT iterator * CommitStore implementation for SMT store * Use interface instead of concrete type * Add telemetry to SMT store * SMT: version->root mapping, cleanup * SMT proofs - initial code * Tests for SMT store ProofOp implementation * Fix linter errors * Use simple 1 byte KV-store prefixes * Improve assertions in tests * Use mutex properly * Store data in ADR-040-compatible way SMT stores: * key -> hash(key, value) KV store stores: * key->value in "bucket 1", * hash(key, value) -> key in "bucket 2".
* Initial SMT store type SMT (Sparse Merkle Tree) is intended to replace IAVL. New type implements same interfaces as iavl.Store. * Add iteration support to SMT Sparse Merkle Tree does not support iteration over keys in order. To provide drop-in replacement for IAVL, Iterator and ReverseIterator has to be implemented. SMT Store implementation use the underlying KV store to: - maintain a list of keys (under a prefix) - iterate over a keys Values are stored only in SMT. * Migrate to smt v0.1.1 * Extra test for SMT iterator * CommitStore implementation for SMT store * Use interface instead of concrete type * Add telemetry to SMT store * SMT: version->root mapping, cleanup * SMT proofs - initial code * Tests for SMT store ProofOp implementation * Fix linter errors * Use simple 1 byte KV-store prefixes * Improve assertions in tests * Use mutex properly * Store data in ADR-040-compatible way SMT stores: * key -> hash(key, value) KV store stores: * key->value in "bucket 1", * hash(key, value) -> key in "bucket 2".
This PR introduces SMT store. It doesn't support
Queryable
interface yet, and it's not used anywhere.This work is related to #8430 and #8297.