hyperledger / fabric-x-committer / 34934205956

91%

Build Type

push

github

Committed by web-flow

Commit Message

[sidecar] Stop allocating per transaction in key validation and TX references (#821)

#### Type of change

- Improvement (improvement to code, performance, etc)
- Test update

#### Description

- `checkKeys` now takes the namespace and walks its three read/write
sets through `nsKey`,
instead of the caller copying every key into one slice for it. It finds
duplicates by
comparing pairs, falling back to a set above `maxKeysForPairwiseCheck`
keys where the
  quadratic comparison would cost more than the map does.
- `mapBlock` allocates the block's `TxRef` and `TxWithRef` values as two
slabs, one entry per
message, instead of one of each per transaction. Only their addresses
are handed on; nothing
  copies an element out, which `go vet`'s copylocks check enforces.
- The mapping benchmarks now generate their transactions with an EDDSA
policy
(`benchTxProfile`) instead of `workload.DefaultProfile`'s `NoScheme`.
With no scheme every
namespace carries a nil endorsement, so mapping rejected all of them
with
`MALFORMED_MISSING_SIGNATURE` before validating a single key or building
a `TxWithRef`:
the benchmarks measured the rejection path, not the path this change
touches.
- Adds `TestKeyFormValidation`, covering empty and duplicate keys on
both sides of the
pairwise/set threshold and duplicates that span two of a namespace's
read/write sets. Neither
`MALFORMED_EMPTY_KEY` nor `MALFORMED_DUPLICATE_KEY_IN_READ_WRITE_SET`
had any coverage before.

#### Additional details (Optional)

On the fixed workload described above, `BenchmarkMapBlockSize` drops
from 24 to 19 allocations
and 1,359 to 1,246 bytes per transaction:

go test ./service/sidecar/ -run '^$' -bench '^BenchmarkMapBlockSize$'
-benchmem

Of that, `checkKeys` accounted for three allocations for the two-key
namespace this workload
generates — the map, the slice the keys were copied into, and the string
conversion per key —
and now allocates none at any key count. The remaining two are the
per-transaction `TxR... (continued)

Coverage Stats

40 of 40 new or added lines in 1 file covered. (100.0%)

2 existing lines in 1 file now uncovered.

9509 of 10498 relevant lines covered (90.58%)

11047.62 hits per line