// ProcessData defers to util.SQLInsertData
func (s *SQLWriter) ProcessData(d data.JSON, outputChan chan data.JSON, killChan chan error) {
// handle panics a bit more gracefully
defer func() {
if err := recover(); err != nil {
util.KillPipelineIfErr(err.(error), killChan)
}
}()
// First check for SQLWriterData
var wd SQLWriterData
err := data.ParseJSONSilent(d, &wd)
logger.Info("SQLWriter: Writing data...")
if err == nil && wd.TableName != "" && wd.InsertData != nil {
logger.Debug("SQLWriter: SQLWriterData scenario")
dd, err := data.NewJSON(wd.InsertData)
util.KillPipelineIfErr(err, killChan)
err = util.SQLInsertData(s.writeDB, dd, wd.TableName, s.OnDupKeyUpdate, s.OnDupKeyFields, s.BatchSize)
util.KillPipelineIfErr(err, killChan)
} else {
logger.Debug("SQLWriter: normal data scenario")
err = util.SQLInsertData(s.writeDB, d, s.TableName, s.OnDupKeyUpdate, s.OnDupKeyFields, s.BatchSize)
util.KillPipelineIfErr(err, killChan)
}
logger.Info("SQLWriter: Write complete")
}
func insertObjects(db *sql.DB, objects []map[string]interface{}, tableName string, onDupKeyUpdate bool, onDupKeyFields []string) error {
logger.Info("SQLInsertData: building INSERT for len(objects) =", len(objects))
insertSQL, vals := buildInsertSQL(objects, tableName, onDupKeyUpdate, onDupKeyFields)
logger.Debug("SQLInsertData:", insertSQL)
logger.Debug("SQLInsertData: values", vals)
stmt, err := db.Prepare(insertSQL)
if err != nil {
logger.Debug("SQLInsertData: error preparing SQL")
return err
}
defer stmt.Close()
res, err := stmt.Exec(vals...)
if err != nil {
return err
}
lastID, err := res.LastInsertId()
if err != nil {
return err
}
rowCnt, err := res.RowsAffected()
if err != nil {
return err
}
logger.Info(fmt.Sprintf("SQLInsertData: rows affected = %d, last insert ID = %d", rowCnt, lastID))
return nil
}
// ParseJSON is a simple wrapper for json.Unmarshal
func ParseJSON(d JSON, v interface{}) error {
err := json.Unmarshal(d, v)
if err != nil {
logger.Debug(fmt.Sprintf("data: failure to unmarshal JSON into %+v - error is \"%v\"", v, err.Error()))
logger.Debug(fmt.Sprintf(" Failed Data: %+v", string(d)))
}
return err
}
// NewJSON is a simple wrapper for json.Marshal.
func NewJSON(v interface{}) (JSON, error) {
d, err := json.Marshal(v)
if err != nil {
logger.Debug(fmt.Sprintf("data: failure to marshal JSON %+v - error is \"%v\"", v, err.Error()))
logger.Debug(fmt.Sprintf(" Failed val: %+v", v))
}
return d, err
}
// ProcessData runs the SQL statements, deferring to util.ExecuteSQLQuery
func (s *SQLExecutor) ProcessData(d data.JSON, outputChan chan data.JSON, killChan chan error) {
// handle panics a bit more gracefully
defer func() {
if err := recover(); err != nil {
util.KillPipelineIfErr(err.(error), killChan)
}
}()
sql := ""
var err error
if s.query == "" && s.sqlGenerator != nil {
sql, err = s.sqlGenerator(d)
util.KillPipelineIfErr(err, killChan)
} else if s.query != "" {
sql = s.query
} else {
killChan <- errors.New("SQLExecutor: must have either static query or sqlGenerator func")
}
logger.Debug("SQLExecutor: Running - ", sql)
// See sql.go
err = util.ExecuteSQLQuery(s.readDB, sql)
util.KillPipelineIfErr(err, killChan)
logger.Info("SQLExecutor: Query complete")
}
// GetS3Object returns the object output for the given object key
func GetS3Object(client *s3.S3, bucket, objKey string) (*s3.GetObjectOutput, error) {
logger.Debug("GetS3Object: ", bucket, "-", objKey)
params := &s3.GetObjectInput{
Bucket: aws.String(bucket), // Required
Key: aws.String(objKey), // Required
// IfMatch: aws.String("IfMatch"),
// IfModifiedSince: aws.Time(time.Now()),
// IfNoneMatch: aws.String("IfNoneMatch"),
// IfUnmodifiedSince: aws.Time(time.Now()),
// Range: aws.String("Range"),
// RequestPayer: aws.String("RequestPayer"),
// ResponseCacheControl: aws.String("ResponseCacheControl"),
// ResponseContentDisposition: aws.String("ResponseContentDisposition"),
// ResponseContentEncoding: aws.String("ResponseContentEncoding"),
// ResponseContentLanguage: aws.String("ResponseContentLanguage"),
// ResponseContentType: aws.String("ResponseContentType"),
// ResponseExpires: aws.Time(time.Now()),
// SSECustomerAlgorithm: aws.String("SSECustomerAlgorithm"),
// SSECustomerKey: aws.String("SSECustomerKey"),
// SSECustomerKeyMD5: aws.String("SSECustomerKeyMD5"),
// VersionId: aws.String("ObjectVersionId"),
}
return client.GetObject(params)
}
// ObjectsFromJSON is a helper for parsing JSON into a slice of
// generic maps/objects. The use-case is when a stage is expecting
// to receive either a JSON object or an array of JSON objects, and
// want to deal with it in a generic fashion.
func ObjectsFromJSON(d JSON) ([]map[string]interface{}, error) {
var objects []map[string]interface{}
// return if we have null instead of object(s).
if bytes.Equal(d, []byte("null")) {
logger.Debug("ObjectsFromJSON: received null. Expected object or objects. Skipping.")
return objects, nil
}
var v interface{}
err := ParseJSON(d, &v)
if err != nil {
return nil, err
}
// check if we have a single object or a slice of objects
switch vv := v.(type) {
case []interface{}:
for _, o := range vv {
objects = append(objects, o.(map[string]interface{}))
}
case map[string]interface{}:
objects = []map[string]interface{}{vv}
case []map[string]interface{}:
objects = vv
default:
err = fmt.Errorf("ObjectsFromJSON: unsupported data type: %T", vv)
return nil, err
}
return objects, nil
}
func (p *Pipeline) runStages(killChan chan error) {
for n, stage := range p.layout.stages {
for _, dp := range stage.processors {
p.wg.Add(1)
// Each DataProcessor runs in a separate gorountine.
go func(n int, dp *dataProcessor) {
// This is where the main DataProcessor interface
// functions are called.
logger.Info(p.Name, "- stage", n+1, dp, "waiting to receive data")
for d := range dp.inputChan {
logger.Info(p.Name, "- stage", n+1, dp, "received data")
if p.PrintData {
logger.Debug(p.Name, "- stage", n+1, dp, "data =", string(d))
}
dp.recordDataReceived(d)
dp.processData(d, killChan)
}
logger.Info(p.Name, "- stage", n+1, dp, "input closed, calling Finish")
dp.Finish(dp.outputChan, killChan)
if dp.outputChan != nil {
logger.Info(p.Name, "- stage", n+1, dp, "closing output")
close(dp.outputChan)
}
p.wg.Done()
}(n, dp)
}
}
}
// ForEachQueryData handles generating the SQL (in case of dynamic mode),
// running the query and retrieving the data in data.JSON format, and then
// passing the results back witih the function call to forEach.
func (s *SQLReader) ForEachQueryData(d data.JSON, killChan chan error, forEach func(d data.JSON)) {
sql := ""
var err error
if s.query == "" && s.sqlGenerator != nil {
sql, err = s.sqlGenerator(d)
util.KillPipelineIfErr(err, killChan)
} else if s.query != "" {
sql = s.query
} else {
killChan <- errors.New("SQLReader: must have either static query or sqlGenerator func")
}
logger.Debug("SQLReader: Running - ", sql)
// See sql.go
dataChan, err := util.GetDataFromSQLQuery(s.readDB, sql, s.BatchSize, s.StructDestination)
util.KillPipelineIfErr(err, killChan)
for d := range dataChan {
// First check if an error was returned back from the SQL processing
// helper, then if not call forEach with the received data.
var derr dataErr
if err := data.ParseJSONSilent(d, &derr); err == nil {
util.KillPipelineIfErr(errors.New(derr.Error), killChan)
} else {
forEach(d)
}
}
}
// Run finalizes the channel connections between PipelineStages
// and kicks off execution.
// Run will return a killChan that should be waited on so your calling function doesn't
// return prematurely. Any stage of the pipeline can send to the killChan to halt
// execution. Your calling function should check if the sent value is an error or nil to know if
// execution was a failure or a success (nil being the success value).
func (p *Pipeline) Run() (killChan chan error) {
p.timer = util.StartTimer()
killChan = make(chan error)
p.connectStages()
p.runStages(killChan)
for _, dp := range p.layout.stages[0].processors {
logger.Debug(p.Name, ": sending", StartSignal, "to", dp)
dp.inputChan <- data.JSON(StartSignal)
dp.Finish(dp.outputChan, killChan)
close(dp.inputChan)
}
// After all the stages are running, send the StartSignal
// to the initial stage processors to kick off execution, and
// then wait until all the processing goroutines are done to
// signal successful pipeline completion.
go func() {
p.wg.Wait()
p.timer.Stop()
killChan <- nil
}()
handleInterrupt(killChan)
return killChan
}
// At this point in pipeline initialization, every dataProcessor has an input
// and output channel, but there is nothing connecting them together. In order
// to support branching and merging between stages (as defined by each
// dataProcessor's outputs), we set up some intermediary channels that will
// manage copying and passing data between stages, as well as properly closing
// channels when all data is received.
func (p *Pipeline) connectStages() {
logger.Debug(p.Name, ": connecting stages")
// First, setup the bridgeing channels & brancher/merger's to aid in
// managing channel communication between processors.
for _, stage := range p.layout.stages {
for _, from := range stage.processors {
if from.outputs != nil {
from.branchOutChans = []chan data.JSON{}
for _, to := range p.dataProcessorOutputs(from) {
if to.mergeInChans == nil {
to.mergeInChans = []chan data.JSON{}
}
c := p.initDataChan()
from.branchOutChans = append(from.branchOutChans, c)
to.mergeInChans = append(to.mergeInChans, c)
}
}
}
}
// Loop through again and setup goroutines to handle data management
// between the branchers and mergers
for _, stage := range p.layout.stages {
for _, dp := range stage.processors {
if dp.branchOutChans != nil {
dp.branchOut()
}
if dp.mergeInChans != nil {
dp.mergeIn()
}
}
}
}
// ForEachQueryData handles generating the SQL (in case of dynamic mode),
// running the query and retrieving the data in data.JSON format, and then
// passing the results back witih the function call to forEach.
func (r *BigQueryReader) ForEachQueryData(d data.JSON, killChan chan error, forEach func(d data.JSON)) {
sql := ""
var err error
if r.query == "" && r.sqlGenerator != nil {
sql, err = r.sqlGenerator(d)
util.KillPipelineIfErr(err, killChan)
} else if r.query != "" {
sql = r.query
} else {
killChan <- errors.New("BigQueryReader: must have either static query or sqlGenerator func")
}
logger.Debug("BigQueryReader: Running -", sql)
bqDataChan := make(chan bigquery.Data)
go r.bqClient().AsyncQuery(r.PageSize, r.config.DatasetID, r.config.ProjectID, sql, bqDataChan)
aggregatedData := bigquery.Data{}
for bqd := range bqDataChan {
util.KillPipelineIfErr(bqd.Err, killChan)
logger.Info("BigQueryReader: received bqData: len(rows) =", len(bqd.Rows))
// logger.Debug(" %+v", bqd)
if bqd.Rows != nil && bqd.Headers != nil && len(bqd.Rows) > 0 {
if r.AggregateResults {
logger.Debug("BigQueryReader: aggregating results")
aggregatedData.Headers = bqd.Headers
aggregatedData.Rows = append(aggregatedData.Rows, bqd.Rows...)
} else {
// Send data as soon as we get it back
logger.Debug("BigQueryReader: sending data without aggregation")
d, err := data.JSONFromHeaderAndRows(bqd.Headers, bqd.Rows)
util.KillPipelineIfErr(err, killChan)
forEach(d) // pass back out via the forEach func
}
}
}
if r.AggregateResults {
logger.Info("BigQueryReader: sending aggregated results: len(rows) =", len(aggregatedData.Rows))
d, err := data.JSONFromHeaderAndRows(aggregatedData.Headers, aggregatedData.Rows)
util.KillPipelineIfErr(err, killChan)
forEach(d) // pass back out via the forEach func
}
}
// ProcessData sends the data it receives to the outputChan only if it matches the supplied regex
func (r *RegexpMatcher) ProcessData(d data.JSON, outputChan chan data.JSON, killChan chan error) {
matches, err := regexp.Match(r.pattern, d)
util.KillPipelineIfErr(err, killChan)
if r.DebugLog {
logger.Debug("RegexpMatcher: checking if", string(d), "matches pattern", r.pattern, ". MATCH=", matches)
}
if matches {
outputChan <- d
}
}
// ProcessData writes the data
func (w *IoWriter) ProcessData(d data.JSON, outputChan chan data.JSON, killChan chan error) {
var bytesWritten int
var err error
if w.AddNewline {
bytesWritten, err = fmt.Fprintln(w.Writer, string(d))
} else {
bytesWritten, err = w.Writer.Write(d)
}
util.KillPipelineIfErr(err, killChan)
logger.Debug("IoWriter:", bytesWritten, "bytes written")
}
// ProcessData writes data as is directly to the output file
func (f *FtpWriter) ProcessData(d data.JSON, outputChan chan data.JSON, killChan chan error) {
logger.Debug("FTPWriter Process data:", string(d))
if !f.authenticated {
f.connect(killChan)
}
_, e := f.fileWriter.Write([]byte(d))
if e != nil {
util.KillPipelineIfErr(e, killChan)
}
}
// sendResults handles sending work that is completed, as well as
// guaranteeing a FIFO order of the resulting data sent over the
// original outputChan.
func (dp *dataProcessor) sendResults() {
dp.Lock()
logger.Debug("dataProcessor: sendResults checking for valid data to send")
e := dp.workList.Front()
for e != nil && e.Value.(*result).done {
logger.Debug("dataHandler: sendResults sending data")
res := dp.workList.Remove(e).(*result)
for _, d := range res.data {
res.outputChan <- d
}
if !res.open {
logger.Debug("dataProcessor: sendResults closing outputChan")
close(res.outputChan)
}
e = dp.workList.Front()
}
dp.Unlock()
if dp.inputClosed && dp.workList.Len() == 0 {
dp.doneChan <- true
}
}
// ProcessData reads an entire directory if a prefix is provided (sending each file in that
// directory to outputChan), or just sends the single file to outputChan if a complete
// file path is provided (not a prefix/directory).
//
// It optionally deletes all processed objects once the contents have been sent to outputChan
func (r *S3Reader) ProcessData(d data.JSON, outputChan chan data.JSON, killChan chan error) {
if r.prefix != "" {
logger.Debug("S3Reader: process data for prefix", r.prefix)
objects, err := util.ListS3Objects(r.client, r.bucket, r.prefix)
logger.Debug("S3Reader: list =", objects)
util.KillPipelineIfErr(err, killChan)
for _, o := range objects {
obj, err := util.GetS3Object(r.client, r.bucket, o)
util.KillPipelineIfErr(err, killChan)
r.processObject(obj, outputChan, killChan)
r.processedObjectKeys = append(r.processedObjectKeys, o)
}
} else {
logger.Debug("S3Reader: process data for object", r.object)
obj, err := util.GetS3Object(r.client, r.bucket, r.object)
util.KillPipelineIfErr(err, killChan)
r.processObject(obj, outputChan, killChan)
r.processedObjectKeys = append(r.processedObjectKeys, r.object)
}
if r.DeleteObjects {
_, err := util.DeleteS3Objects(r.client, r.bucket, r.processedObjectKeys)
util.KillPipelineIfErr(err, killChan)
}
}
func (dp *dataProcessor) processData(d data.JSON, killChan chan error) {
logger.Debug("dataProcessor: processData", dp, "with concurrency =", dp.concurrency)
// If no concurrency is needed, simply call stage.ProcessData and return...
if dp.concurrency <= 1 {
dp.recordExecution(func() {
dp.ProcessData(d, dp.outputChan, killChan)
})
return
}
// ... otherwise process the data in a concurrent queue/pool of goroutines
logger.Debug("dataProcessor: processData", dp, "waiting for work")
// wait for room in the queue
dp.workThrottle <- workSignal{}
logger.Debug("dataProcessor: processData", dp, "work obtained")
rc := make(chan data.JSON)
done := make(chan bool)
exit := make(chan bool)
// setup goroutine to handle result
go func() {
res := result{outputChan: dp.outputChan, data: []data.JSON{}, open: true}
dp.Lock()
dp.workList.PushBack(&res)
dp.Unlock()
logger.Debug("dataProcessor: processData", dp, "waiting to receive data on result chan")
for {
select {
case d, open := <-rc:
logger.Debug("dataProcessor: processData", dp, "received data on result chan")
res.data = append(res.data, d)
// outputChan will need to be closed if the rc chan was closed
res.open = open
case <-done:
res.done = true
logger.Debug("dataProcessor: processData", dp, "done, releasing work")
<-dp.workThrottle
dp.sendResults()
exit <- true
return
}
}
}()
// do normal data processing, passing in new result chan
// instead of the original outputChan
go dp.recordExecution(func() {
dp.ProcessData(d, rc, killChan)
done <- true
})
// wait on processing to complete
<-exit
}
// DeleteS3Objects deletes the objects specified by the given object keys
func DeleteS3Objects(client *s3.S3, bucket string, objKeys []string) (*s3.DeleteObjectsOutput, error) {
logger.Debug("DeleteS3Objects: ", bucket, "-", objKeys)
s3Ids := make([]*s3.ObjectIdentifier, len(objKeys))
for i, key := range objKeys {
s3Ids[i] = &s3.ObjectIdentifier{Key: aws.String(key)}
}
params := &s3.DeleteObjectsInput{
Bucket: aws.String(bucket), // Required
Delete: &s3.Delete{ // Required
Objects: s3Ids,
Quiet: aws.Bool(true),
},
// MFA: aws.String("MFA"),
// RequestPayer: aws.String("RequestPayer"),
}
return client.DeleteObjects(params)
}
// ListS3Objects returns all object keys matching the given prefix. Note that
// delimiter is set to "/". See http://docs.aws.amazon.com/AmazonS3/latest/dev/ListingKeysHierarchy.html
func ListS3Objects(client *s3.S3, bucket, keyPrefix string) ([]string, error) {
logger.Debug("ListS3Objects: ", bucket, "-", keyPrefix)
params := &s3.ListObjectsInput{
Bucket: aws.String(bucket), // Required
Delimiter: aws.String("/"),
// EncodingType: aws.String("EncodingType"),
// Marker: aws.String("Marker"),
MaxKeys: aws.Int64(1000),
Prefix: aws.String(keyPrefix),
}
objects := []string{}
err := client.ListObjectsPages(params, func(page *s3.ListObjectsOutput, lastPage bool) bool {
for _, o := range page.Contents {
objects = append(objects, *o.Key)
}
return lastPage
})
if err != nil {
return nil, err
}
return objects, nil
}
// ProcessData writes data as is directly to the output file
func (w *SftpWriter) ProcessData(d data.JSON, outputChan chan data.JSON, killChan chan error) {
logger.Debug("SftpWriter Process data:", string(d))
w.ensureInitialized(killChan)
_, e := w.file.Write([]byte(d))
util.KillPipelineIfErr(e, killChan)
}
