storage: Remove revision from GetMerkleNodes #2391
Conversation
Log trees always read from the revision corresponding to the root which is read in the beginning of the transaction. There is no need for it to make a roundtrip to the application layer.
pav-kv
force-pushed
the
refactoring
branch
3 times, most recently
from
16023a7
to
7c06486
Compare
Codecov Report
@@ Coverage Diff @@
## master #2391 +/- ##
==========================================
- Coverage 65.76% 65.68% -0.09%
==========================================
Files 107 107
Lines 7759 7758 -1
==========================================
- Hits 5103 5096 -7
- Misses 2120 2127 +7
+ Partials 536 535 -1
Continue to review full report at Codecov.
|
| @@ -372,6 +373,14 @@ func (t *logTreeTX) WriteRevision(ctx context.Context) (int64, error) { | |||
| return t.treeTX.writeRevision, nil | |||
| } | |||
|
|
|||
| // GetMerkleNodes returns the requested nodes at the read revision. | |||
| func (t *logTreeTX) GetMerkleNodes(ctx context.Context, nodeIDs []tree.NodeID) ([]tree.Node, error) { | |||
There was a problem hiding this comment.
This bit might look confusing: The GetMerkleNodes method just moved from treeTX to logTreeTX.
Reason: treeTX (for this particular implementation) doesn't know root.revision, whereas the logTreeTX does. In contrast, in other implementations (e.g. cloudspanner and spanner) tree storage TX does know it.
This indicates that tree vs log storage abstractions are sometimes violated / not defined. Given that we only have logs now, the bigger plan is to combine the two into one interface [#2378].
Log trees always read from the revision corresponding to the root which
is read in the beginning of the transaction. There is no need for it to
make a roundtrip to the application layer.
Part of a larger refactoring [#2378]. More follow-up changes are WIP.
Checklist