What

Instead of decoding the raw binary data directly into an xdr.LedgerCloseMeta, this modifies the code to only decode the xdr.LedgerHeaderHistoryEntry out of the data which comes first in the layout. The rest of the meta is passed around directly as a []byte until the final step, either converting it to base64 directly or to JSON.

Rudimentary benchmarks show a ~40% improvement over the baseline:

$ git switch main
$ go test -failfast -run=None -bench=BenchmarkBatchGetLedgers ./cmd/stellar-rpc/internal/db
BenchmarkBatchGetLedgers-16    	1788	    773248 ns/op
ok  	github.com/stellar/stellar-rpc/cmd/stellar-rpc/internal/db	7.652s

$ git switch getLedgers_depthCheck
$ go test -failfast -run=None -bench=BenchmarkBatchGetLedgers ./cmd/stellar-rpc/internal/db
BenchmarkBatchGetLedgers-16    	    1939	    530915 ns/op
ok  	github.com/stellar/stellar-rpc/cmd/stellar-rpc/internal/db	3.325s

Why

We have observed performance issues with getLedgers due to their density and would like to improve them.

Known limitations

This needs to be evaluated against real load to see the improvement.