ipld-eth-server/pkg/super_node/eth/cid_retriever.go

620 lines
22 KiB
Go

// VulcanizeDB
// Copyright © 2019 Vulcanize
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU Affero General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU Affero General Public License for more details.
// You should have received a copy of the GNU Affero General Public License
// along with this program. If not, see <http://www.gnu.org/licenses/>.
package eth
import (
"database/sql"
"fmt"
"math/big"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/crypto"
"github.com/jmoiron/sqlx"
"github.com/lib/pq"
log "github.com/sirupsen/logrus"
utils "github.com/vulcanize/vulcanizedb/libraries/shared/utilities"
"github.com/vulcanize/vulcanizedb/pkg/postgres"
"github.com/vulcanize/vulcanizedb/pkg/super_node/shared"
)
// CIDRetriever satisfies the CIDRetriever interface for ethereum
type CIDRetriever struct {
db *postgres.DB
}
// NewCIDRetriever returns a pointer to a new CIDRetriever which supports the CIDRetriever interface
func NewCIDRetriever(db *postgres.DB) *CIDRetriever {
return &CIDRetriever{
db: db,
}
}
// RetrieveFirstBlockNumber is used to retrieve the first block number in the db
func (ecr *CIDRetriever) RetrieveFirstBlockNumber() (int64, error) {
var blockNumber int64
err := ecr.db.Get(&blockNumber, "SELECT block_number FROM eth.header_cids ORDER BY block_number ASC LIMIT 1")
return blockNumber, err
}
// RetrieveLastBlockNumber is used to retrieve the latest block number in the db
func (ecr *CIDRetriever) RetrieveLastBlockNumber() (int64, error) {
var blockNumber int64
err := ecr.db.Get(&blockNumber, "SELECT block_number FROM eth.header_cids ORDER BY block_number DESC LIMIT 1 ")
return blockNumber, err
}
// Retrieve is used to retrieve all of the CIDs which conform to the passed StreamFilters
func (ecr *CIDRetriever) Retrieve(filter shared.SubscriptionSettings, blockNumber int64) ([]shared.CIDsForFetching, bool, error) {
streamFilter, ok := filter.(*SubscriptionSettings)
if !ok {
return nil, true, fmt.Errorf("eth retriever expected filter type %T got %T", &SubscriptionSettings{}, filter)
}
log.Debug("retrieving cids")
// Begin new db tx
tx, err := ecr.db.Beginx()
if err != nil {
return nil, true, err
}
defer func() {
if p := recover(); p != nil {
shared.Rollback(tx)
panic(p)
} else if err != nil {
shared.Rollback(tx)
} else {
err = tx.Commit()
}
}()
// Retrieve cached header CIDs at this block height
headers, err := ecr.RetrieveHeaderCIDs(tx, blockNumber)
if err != nil {
log.Error("header cid retrieval error")
return nil, true, err
}
cws := make([]shared.CIDsForFetching, len(headers))
empty := true
for i, header := range headers {
cw := new(CIDWrapper)
cw.BlockNumber = big.NewInt(blockNumber)
if !streamFilter.HeaderFilter.Off {
cw.Header = header
empty = false
if streamFilter.HeaderFilter.Uncles {
// Retrieve uncle cids for this header id
uncleCIDs, err := ecr.RetrieveUncleCIDsByHeaderID(tx, header.ID)
if err != nil {
log.Error("uncle cid retrieval error")
return nil, true, err
}
cw.Uncles = uncleCIDs
}
}
// Retrieve cached trx CIDs
if !streamFilter.TxFilter.Off {
cw.Transactions, err = ecr.RetrieveTxCIDs(tx, streamFilter.TxFilter, header.ID)
if err != nil {
log.Error("transaction cid retrieval error")
return nil, true, err
}
if len(cw.Transactions) > 0 {
empty = false
}
}
trxIds := make([]int64, len(cw.Transactions))
for j, tx := range cw.Transactions {
trxIds[j] = tx.ID
}
// Retrieve cached receipt CIDs
if !streamFilter.ReceiptFilter.Off {
cw.Receipts, err = ecr.RetrieveRctCIDsByHeaderID(tx, streamFilter.ReceiptFilter, header.ID, trxIds)
if err != nil {
log.Error("receipt cid retrieval error")
return nil, true, err
}
if len(cw.Receipts) > 0 {
empty = false
}
}
// Retrieve cached state CIDs
if !streamFilter.StateFilter.Off {
cw.StateNodes, err = ecr.RetrieveStateCIDs(tx, streamFilter.StateFilter, header.ID)
if err != nil {
log.Error("state cid retrieval error")
return nil, true, err
}
if len(cw.StateNodes) > 0 {
empty = false
}
}
// Retrieve cached storage CIDs
if !streamFilter.StorageFilter.Off {
cw.StorageNodes, err = ecr.RetrieveStorageCIDs(tx, streamFilter.StorageFilter, header.ID)
if err != nil {
log.Error("storage cid retrieval error")
return nil, true, err
}
if len(cw.StorageNodes) > 0 {
empty = false
}
}
cws[i] = cw
}
return cws, empty, err
}
// RetrieveHeaderCIDs retrieves and returns all of the header cids at the provided blockheight
func (ecr *CIDRetriever) RetrieveHeaderCIDs(tx *sqlx.Tx, blockNumber int64) ([]HeaderModel, error) {
log.Debug("retrieving header cids for block ", blockNumber)
headers := make([]HeaderModel, 0)
pgStr := `SELECT * FROM eth.header_cids
WHERE block_number = $1`
return headers, tx.Select(&headers, pgStr, blockNumber)
}
// RetrieveUncleCIDsByHeaderID retrieves and returns all of the uncle cids for the provided header
func (ecr *CIDRetriever) RetrieveUncleCIDsByHeaderID(tx *sqlx.Tx, headerID int64) ([]UncleModel, error) {
log.Debug("retrieving uncle cids for block id ", headerID)
headers := make([]UncleModel, 0)
pgStr := `SELECT * FROM eth.uncle_cids
WHERE header_id = $1`
return headers, tx.Select(&headers, pgStr, headerID)
}
// RetrieveTxCIDs retrieves and returns all of the trx cids at the provided blockheight that conform to the provided filter parameters
// also returns the ids for the returned transaction cids
func (ecr *CIDRetriever) RetrieveTxCIDs(tx *sqlx.Tx, txFilter TxFilter, headerID int64) ([]TxModel, error) {
log.Debug("retrieving transaction cids for header id ", headerID)
args := make([]interface{}, 0, 3)
results := make([]TxModel, 0)
id := 1
pgStr := fmt.Sprintf(`SELECT transaction_cids.id, transaction_cids.header_id,
transaction_cids.tx_hash, transaction_cids.cid,
transaction_cids.dst, transaction_cids.src, transaction_cids.index
FROM eth.transaction_cids INNER JOIN eth.header_cids ON (transaction_cids.header_id = header_cids.id)
WHERE header_cids.id = $%d`, id)
args = append(args, headerID)
id++
if len(txFilter.Dst) > 0 {
pgStr += fmt.Sprintf(` AND transaction_cids.dst = ANY($%d::VARCHAR(66)[])`, id)
args = append(args, pq.Array(txFilter.Dst))
id++
}
if len(txFilter.Src) > 0 {
pgStr += fmt.Sprintf(` AND transaction_cids.src = ANY($%d::VARCHAR(66)[])`, id)
args = append(args, pq.Array(txFilter.Src))
}
pgStr += ` ORDER BY transaction_cids.index`
return results, tx.Select(&results, pgStr, args...)
}
// RetrieveRctCIDsByHeaderID retrieves and returns all of the rct cids at the provided header ID that conform to the provided
// filter parameters and correspond to the provided tx ids
func (ecr *CIDRetriever) RetrieveRctCIDsByHeaderID(tx *sqlx.Tx, rctFilter ReceiptFilter, headerID int64, trxIds []int64) ([]ReceiptModel, error) {
log.Debug("retrieving receipt cids for header id ", headerID)
args := make([]interface{}, 0, 4)
pgStr := `SELECT receipt_cids.id, receipt_cids.tx_id, receipt_cids.cid, receipt_cids.contract,
receipt_cids.contract_hash, receipt_cids.topic0s, receipt_cids.topic1s,
receipt_cids.topic2s, receipt_cids.topic3s, receipt_cids.log_contracts
FROM eth.receipt_cids, eth.transaction_cids, eth.header_cids
WHERE receipt_cids.tx_id = transaction_cids.id
AND transaction_cids.header_id = header_cids.id
AND header_cids.id = $1`
id := 2
args = append(args, headerID)
if len(rctFilter.LogAddresses) > 0 {
// Filter on log contract addresses if there are any
pgStr += fmt.Sprintf(` AND ((receipt_cids.log_contracts && $%d::VARCHAR(66)[]`, id)
args = append(args, pq.Array(rctFilter.LogAddresses))
id++
// Filter on topics if there are any
if hasTopics(rctFilter.Topics) {
pgStr += " AND ("
first := true
for i, topicSet := range rctFilter.Topics {
if i < 4 && len(topicSet) > 0 {
if first {
pgStr += fmt.Sprintf(`receipt_cids.topic%ds && $%d::VARCHAR(66)[]`, i, id)
first = false
} else {
pgStr += fmt.Sprintf(` AND receipt_cids.topic%ds && $%d::VARCHAR(66)[]`, i, id)
}
args = append(args, pq.Array(topicSet))
id++
}
}
pgStr += ")"
}
pgStr += ")"
// Filter on txIDs if there are any and we are matching txs
if rctFilter.MatchTxs && len(trxIds) > 0 {
pgStr += fmt.Sprintf(` OR receipt_cids.tx_id = ANY($%d::INTEGER[])`, id)
args = append(args, pq.Array(trxIds))
}
pgStr += ")"
} else { // If there are no contract addresses to filter on
// Filter on topics if there are any
if hasTopics(rctFilter.Topics) {
pgStr += " AND (("
first := true
for i, topicSet := range rctFilter.Topics {
if i < 4 && len(topicSet) > 0 {
if first {
pgStr += fmt.Sprintf(`receipt_cids.topic%ds && $%d::VARCHAR(66)[]`, i, id)
first = false
} else {
pgStr += fmt.Sprintf(` AND receipt_cids.topic%ds && $%d::VARCHAR(66)[]`, i, id)
}
args = append(args, pq.Array(topicSet))
id++
}
}
pgStr += ")"
// Filter on txIDs if there are any and we are matching txs
if rctFilter.MatchTxs && len(trxIds) > 0 {
pgStr += fmt.Sprintf(` OR receipt_cids.tx_id = ANY($%d::INTEGER[])`, id)
args = append(args, pq.Array(trxIds))
}
pgStr += ")"
} else if rctFilter.MatchTxs && len(trxIds) > 0 {
// If there are no contract addresses or topics to filter on,
// Filter on txIDs if there are any and we are matching txs
pgStr += fmt.Sprintf(` AND receipt_cids.tx_id = ANY($%d::INTEGER[])`, id)
args = append(args, pq.Array(trxIds))
}
}
pgStr += ` ORDER BY transaction_cids.index`
receiptCids := make([]ReceiptModel, 0)
return receiptCids, tx.Select(&receiptCids, pgStr, args...)
}
// RetrieveRctCIDs retrieves and returns all of the rct cids at the provided blockheight or block hash that conform to the provided
// filter parameters and correspond to the provided tx ids
func (ecr *CIDRetriever) RetrieveRctCIDs(tx *sqlx.Tx, rctFilter ReceiptFilter, blockNumber int64, blockHash *common.Hash, trxIds []int64) ([]ReceiptModel, error) {
log.Debug("retrieving receipt cids for block ", blockNumber)
args := make([]interface{}, 0, 5)
pgStr := `SELECT receipt_cids.id, receipt_cids.tx_id, receipt_cids.cid, receipt_cids.contract,
receipt_cids.contract_hash, receipt_cids.topic0s, receipt_cids.topic1s,
receipt_cids.topic2s, receipt_cids.topic3s, receipt_cids.log_contracts
FROM eth.receipt_cids, eth.transaction_cids, eth.header_cids
WHERE receipt_cids.tx_id = transaction_cids.id
AND transaction_cids.header_id = header_cids.id`
id := 1
if blockNumber > 0 {
pgStr += fmt.Sprintf(` AND header_cids.block_number = $%d`, id)
args = append(args, blockNumber)
id++
}
if blockHash != nil {
pgStr += fmt.Sprintf(` AND header_cids.block_hash = $%d`, id)
args = append(args, blockHash.String())
id++
}
if len(rctFilter.LogAddresses) > 0 {
// Filter on log contract addresses if there are any
pgStr += fmt.Sprintf(` AND ((receipt_cids.log_contracts && $%d::VARCHAR(66)[]`, id)
args = append(args, pq.Array(rctFilter.LogAddresses))
id++
// Filter on topics if there are any
if hasTopics(rctFilter.Topics) {
pgStr += " AND ("
first := true
for i, topicSet := range rctFilter.Topics {
if i < 4 && len(topicSet) > 0 {
if first {
pgStr += fmt.Sprintf(`receipt_cids.topic%ds && $%d::VARCHAR(66)[]`, i, id)
first = false
} else {
pgStr += fmt.Sprintf(` AND receipt_cids.topic%ds && $%d::VARCHAR(66)[]`, i, id)
}
args = append(args, pq.Array(topicSet))
id++
}
}
pgStr += ")"
}
pgStr += ")"
// Filter on txIDs if there are any and we are matching txs
if rctFilter.MatchTxs && len(trxIds) > 0 {
pgStr += fmt.Sprintf(` OR receipt_cids.tx_id = ANY($%d::INTEGER[])`, id)
args = append(args, pq.Array(trxIds))
}
pgStr += ")"
} else { // If there are no contract addresses to filter on
// Filter on topics if there are any
if hasTopics(rctFilter.Topics) {
pgStr += " AND (("
first := true
for i, topicSet := range rctFilter.Topics {
if i < 4 && len(topicSet) > 0 {
if first {
pgStr += fmt.Sprintf(`receipt_cids.topic%ds && $%d::VARCHAR(66)[]`, i, id)
first = false
} else {
pgStr += fmt.Sprintf(` AND receipt_cids.topic%ds && $%d::VARCHAR(66)[]`, i, id)
}
args = append(args, pq.Array(topicSet))
id++
}
}
pgStr += ")"
// Filter on txIDs if there are any and we are matching txs
if rctFilter.MatchTxs && len(trxIds) > 0 {
pgStr += fmt.Sprintf(` OR receipt_cids.tx_id = ANY($%d::INTEGER[])`, id)
args = append(args, pq.Array(trxIds))
}
pgStr += ")"
} else if rctFilter.MatchTxs && len(trxIds) > 0 {
// If there are no contract addresses or topics to filter on,
// Filter on txIDs if there are any and we are matching txs
pgStr += fmt.Sprintf(` AND receipt_cids.tx_id = ANY($%d::INTEGER[])`, id)
args = append(args, pq.Array(trxIds))
}
}
pgStr += ` ORDER BY transaction_cids.index`
receiptCids := make([]ReceiptModel, 0)
return receiptCids, tx.Select(&receiptCids, pgStr, args...)
}
func hasTopics(topics [][]string) bool {
for _, topicSet := range topics {
if len(topicSet) > 0 {
return true
}
}
return false
}
// RetrieveStateCIDs retrieves and returns all of the state node cids at the provided header ID that conform to the provided filter parameters
func (ecr *CIDRetriever) RetrieveStateCIDs(tx *sqlx.Tx, stateFilter StateFilter, headerID int64) ([]StateNodeModel, error) {
log.Debug("retrieving state cids for header id ", headerID)
args := make([]interface{}, 0, 2)
pgStr := `SELECT state_cids.id, state_cids.header_id,
state_cids.state_leaf_key, state_cids.node_type, state_cids.cid, state_cids.state_path
FROM eth.state_cids INNER JOIN eth.header_cids ON (state_cids.header_id = header_cids.id)
WHERE header_cids.id = $1`
args = append(args, headerID)
addrLen := len(stateFilter.Addresses)
if addrLen > 0 {
keys := make([]string, addrLen)
for i, addr := range stateFilter.Addresses {
keys[i] = crypto.Keccak256Hash(common.HexToAddress(addr).Bytes()).String()
}
pgStr += ` AND state_cids.state_leaf_key = ANY($2::VARCHAR(66)[])`
args = append(args, pq.Array(keys))
}
if !stateFilter.IntermediateNodes {
pgStr += ` AND state_cids.node_type = 2`
}
stateNodeCIDs := make([]StateNodeModel, 0)
return stateNodeCIDs, tx.Select(&stateNodeCIDs, pgStr, args...)
}
// RetrieveStorageCIDs retrieves and returns all of the storage node cids at the provided header id that conform to the provided filter parameters
func (ecr *CIDRetriever) RetrieveStorageCIDs(tx *sqlx.Tx, storageFilter StorageFilter, headerID int64) ([]StorageNodeWithStateKeyModel, error) {
log.Debug("retrieving storage cids for header id ", headerID)
args := make([]interface{}, 0, 3)
pgStr := `SELECT storage_cids.id, storage_cids.state_id, storage_cids.storage_leaf_key,
storage_cids.node_type, storage_cids.cid, storage_cids.storage_path, state_cids.state_leaf_key
FROM eth.storage_cids, eth.state_cids, eth.header_cids
WHERE storage_cids.state_id = state_cids.id
AND state_cids.header_id = header_cids.id
AND header_cids.id = $1`
args = append(args, headerID)
id := 2
addrLen := len(storageFilter.Addresses)
if addrLen > 0 {
keys := make([]string, addrLen)
for i, addr := range storageFilter.Addresses {
keys[i] = crypto.Keccak256Hash(common.HexToAddress(addr).Bytes()).String()
}
pgStr += fmt.Sprintf(` AND state_cids.state_leaf_key = ANY($%d::VARCHAR(66)[])`, id)
args = append(args, pq.Array(keys))
id++
}
if len(storageFilter.StorageKeys) > 0 {
pgStr += fmt.Sprintf(` AND storage_cids.storage_leaf_key = ANY($%d::VARCHAR(66)[])`, id)
args = append(args, pq.Array(storageFilter.StorageKeys))
}
if !storageFilter.IntermediateNodes {
pgStr += ` AND storage_cids.node_type = 2`
}
storageNodeCIDs := make([]StorageNodeWithStateKeyModel, 0)
return storageNodeCIDs, tx.Select(&storageNodeCIDs, pgStr, args...)
}
// RetrieveGapsInData is used to find the the block numbers at which we are missing data in the db
// it finds the union of heights where no data exists and where the times_validated is lower than the validation level
func (ecr *CIDRetriever) RetrieveGapsInData(validationLevel int) ([]shared.Gap, error) {
log.Info("searching for gaps in the eth super node database")
startingBlock, err := ecr.RetrieveFirstBlockNumber()
if err != nil {
return nil, fmt.Errorf("eth CIDRetriever RetrieveFirstBlockNumber error: %v", err)
}
var initialGap []shared.Gap
if startingBlock != 0 {
stop := uint64(startingBlock - 1)
log.Infof("found gap at the beginning of the eth sync from 0 to %d", stop)
initialGap = []shared.Gap{{
Start: 0,
Stop: stop,
}}
}
pgStr := `SELECT header_cids.block_number + 1 AS start, min(fr.block_number) - 1 AS stop FROM eth.header_cids
LEFT JOIN eth.header_cids r on eth.header_cids.block_number = r.block_number - 1
LEFT JOIN eth.header_cids fr on eth.header_cids.block_number < fr.block_number
WHERE r.block_number is NULL and fr.block_number IS NOT NULL
GROUP BY header_cids.block_number, r.block_number`
results := make([]struct {
Start uint64 `db:"start"`
Stop uint64 `db:"stop"`
}, 0)
if err := ecr.db.Select(&results, pgStr); err != nil && err != sql.ErrNoRows {
return nil, err
}
emptyGaps := make([]shared.Gap, len(results))
for i, res := range results {
emptyGaps[i] = shared.Gap{
Start: res.Start,
Stop: res.Stop,
}
}
// Find sections of blocks where we are below the validation level
// There will be no overlap between these "gaps" and the ones above
pgStr = `SELECT block_number FROM eth.header_cids
WHERE times_validated < $1
ORDER BY block_number`
var heights []uint64
if err := ecr.db.Select(&heights, pgStr, validationLevel); err != nil && err != sql.ErrNoRows {
return nil, err
}
return append(append(initialGap, emptyGaps...), utils.MissingHeightsToGaps(heights)...), nil
}
// RetrieveBlockByHash returns all of the CIDs needed to compose an entire block, for a given block hash
func (ecr *CIDRetriever) RetrieveBlockByHash(blockHash common.Hash) (HeaderModel, []UncleModel, []TxModel, []ReceiptModel, error) {
log.Debug("retrieving block cids for block hash ", blockHash.String())
// Begin new db tx
tx, err := ecr.db.Beginx()
if err != nil {
return HeaderModel{}, nil, nil, nil, err
}
defer func() {
if p := recover(); p != nil {
shared.Rollback(tx)
panic(p)
} else if err != nil {
shared.Rollback(tx)
} else {
err = tx.Commit()
}
}()
headerCID, err := ecr.RetrieveHeaderCIDByHash(tx, blockHash)
if err != nil {
log.Error("header cid retrieval error")
return HeaderModel{}, nil, nil, nil, err
}
uncleCIDs, err := ecr.RetrieveUncleCIDsByHeaderID(tx, headerCID.ID)
if err != nil {
log.Error("uncle cid retrieval error")
return HeaderModel{}, nil, nil, nil, err
}
txCIDs, err := ecr.RetrieveTxCIDsByHeaderID(tx, headerCID.ID)
if err != nil {
log.Error("tx cid retrieval error")
return HeaderModel{}, nil, nil, nil, err
}
txIDs := make([]int64, len(txCIDs))
for i, txCID := range txCIDs {
txIDs[i] = txCID.ID
}
rctCIDs, err := ecr.RetrieveReceiptCIDsByTxIDs(tx, txIDs)
if err != nil {
log.Error("rct cid retrieval error")
}
return headerCID, uncleCIDs, txCIDs, rctCIDs, err
}
// RetrieveBlockByNumber returns all of the CIDs needed to compose an entire block, for a given block number
func (ecr *CIDRetriever) RetrieveBlockByNumber(blockNumber int64) (HeaderModel, []UncleModel, []TxModel, []ReceiptModel, error) {
log.Debug("retrieving block cids for block number ", blockNumber)
// Begin new db tx
tx, err := ecr.db.Beginx()
if err != nil {
return HeaderModel{}, nil, nil, nil, err
}
defer func() {
if p := recover(); p != nil {
shared.Rollback(tx)
panic(p)
} else if err != nil {
shared.Rollback(tx)
} else {
err = tx.Commit()
}
}()
headerCID, err := ecr.RetrieveHeaderCIDs(tx, blockNumber)
if err != nil {
log.Error("header cid retrieval error")
return HeaderModel{}, nil, nil, nil, err
}
if len(headerCID) < 1 {
return HeaderModel{}, nil, nil, nil, fmt.Errorf("header cid retrieval error, no header CIDs found at block %d", blockNumber)
}
uncleCIDs, err := ecr.RetrieveUncleCIDsByHeaderID(tx, headerCID[0].ID)
if err != nil {
log.Error("uncle cid retrieval error")
return HeaderModel{}, nil, nil, nil, err
}
txCIDs, err := ecr.RetrieveTxCIDsByHeaderID(tx, headerCID[0].ID)
if err != nil {
log.Error("tx cid retrieval error")
return HeaderModel{}, nil, nil, nil, err
}
txIDs := make([]int64, len(txCIDs))
for i, txCID := range txCIDs {
txIDs[i] = txCID.ID
}
rctCIDs, err := ecr.RetrieveReceiptCIDsByTxIDs(tx, txIDs)
if err != nil {
log.Error("rct cid retrieval error")
}
return headerCID[0], uncleCIDs, txCIDs, rctCIDs, err
}
// RetrieveHeaderCIDByHash returns the header for the given block hash
func (ecr *CIDRetriever) RetrieveHeaderCIDByHash(tx *sqlx.Tx, blockHash common.Hash) (HeaderModel, error) {
log.Debug("retrieving header cids for block hash ", blockHash.String())
pgStr := `SELECT * FROM eth.header_cids
WHERE block_hash = $1`
var headerCID HeaderModel
return headerCID, tx.Get(&headerCID, pgStr, blockHash.String())
}
// RetrieveTxCIDsByHeaderID retrieves all tx CIDs for the given header id
func (ecr *CIDRetriever) RetrieveTxCIDsByHeaderID(tx *sqlx.Tx, headerID int64) ([]TxModel, error) {
log.Debug("retrieving tx cids for block id ", headerID)
pgStr := `SELECT * FROM eth.transaction_cids
WHERE header_id = $1
ORDER BY index`
var txCIDs []TxModel
return txCIDs, tx.Select(&txCIDs, pgStr, headerID)
}
// RetrieveReceiptCIDsByTxIDs retrieves receipt CIDs by their associated tx IDs
func (ecr *CIDRetriever) RetrieveReceiptCIDsByTxIDs(tx *sqlx.Tx, txIDs []int64) ([]ReceiptModel, error) {
log.Debugf("retrieving receipt cids for tx ids %v", txIDs)
pgStr := `SELECT receipt_cids.id, receipt_cids.tx_id, receipt_cids.cid, receipt_cids.contract,
receipt_cids.contract_hash, receipt_cids.topic0s, receipt_cids.topic1s,
receipt_cids.topic2s, receipt_cids.topic3s, receipt_cids.log_contracts
FROM eth.receipt_cids, eth.transaction_cids
WHERE tx_id = ANY($1::INTEGER[])
AND receipt_cids.tx_id = transaction_cids.id
ORDER BY transaction_cids.index`
var rctCIDs []ReceiptModel
return rctCIDs, tx.Select(&rctCIDs, pgStr, pq.Array(txIDs))
}