You are viewing a plain text version of this content. The canonical link for it is here.
Posted to github@beam.apache.org by GitBox <gi...@apache.org> on 2020/04/21 18:50:34 UTC
[GitHub] [beam] lostluck commented on a change in pull request #11474: [BEAM-9642] Enabling SDF exec runners and fixing small bug.
lostluck commented on a change in pull request #11474:
URL: https://github.com/apache/beam/pull/11474#discussion_r412316843
##########
File path: sdks/go/pkg/beam/core/runtime/exec/sdf.go
##########
@@ -263,20 +266,33 @@ func (n *ProcessSizedElementsAndRestrictions) ProcessElement(ctx context.Context
return errors.Errorf("invalid status for ParDo %v: %v, want Active", n.PDo.UID, n.PDo.status)
}
- userElm := elm.Elm.(*FullValue).Elm.(*FullValue)
rest := elm.Elm.(*FullValue).Elm2
rt := n.inv.Invoke(rest)
+ var mainIn = &MainInput{
Review comment:
The idiomatic form for this would be to do the following:
```suggestion
mainIn := &MainInput{
```
var declarations are usually for uninitialized variables.
##########
File path: sdks/go/pkg/beam/core/runtime/exec/sdf.go
##########
@@ -263,20 +266,33 @@ func (n *ProcessSizedElementsAndRestrictions) ProcessElement(ctx context.Context
return errors.Errorf("invalid status for ParDo %v: %v, want Active", n.PDo.UID, n.PDo.status)
}
- userElm := elm.Elm.(*FullValue).Elm.(*FullValue)
rest := elm.Elm.(*FullValue).Elm2
rt := n.inv.Invoke(rest)
+ var mainIn = &MainInput{
+ Values: values,
+ RTracker: rt,
+ }
- return n.PDo.processMainInput(&MainInput{
- Key: FullValue{ // User userElm's values but the top-level windows and timestamp.
+ // For the key, the way we fill it out depends on whether the input element
+ // is a KV or single-element. Single-elements might have been lifted out of
+ // their FullValue if they were decoded, so we need to have a case for that.
+ // Also, we use the the top-level windows and timestamp.
+ if userElm, ok := elm.Elm.(*FullValue).Elm.(*FullValue); ok {
Review comment:
Note for the performance pass: We can know KV vs not KV from the input PCollection coderat exec/translate.go time, so we can avoid doing this if branch per element.
##########
File path: sdks/go/pkg/beam/core/runtime/exec/translate.go
##########
@@ -372,32 +386,43 @@ func (b *builder) makeLink(from string, id linkID) (Node, error) {
switch op {
case graph.ParDo:
- n := &ParDo{UID: b.idgen.New(), Inbound: in, Out: out}
- n.Fn, err = graph.AsDoFn(fn, graph.MainUnknown)
+ dofn, err := graph.AsDoFn(fn, graph.MainUnknown)
if err != nil {
return nil, err
}
- n.PID = transform.GetUniqueName()
-
- input := unmarshalKeyedValues(transform.GetInputs())
- for i := 1; i < len(input); i++ {
- // TODO(herohde) 8/8/2018: handle different windows, view_fn and window_mapping_fn.
- // For now, assume we don't need any information in the pardo payload.
-
- ec, wc, err := b.makeCoderForPCollection(input[i])
- if err != nil {
- return nil, err
+ switch urn {
+ case urnPairWithRestriction:
+ u = &PairWithRestriction{UID: b.idgen.New(), Fn: dofn, Out: out[0]}
+ case urnSplitAndSizeRestrictions:
+ u = &SplitAndSizeRestrictions{UID: b.idgen.New(), Fn: dofn, Out: out[0]}
+ default:
+ n := &ParDo{UID: b.idgen.New(), Fn: dofn, Inbound: in, Out: out}
+ n.PID = transform.GetUniqueName()
+
+ input := unmarshalKeyedValues(transform.GetInputs())
+ for i := 1; i < len(input); i++ {
+ // TODO(herohde) 8/8/2018: handle different windows, view_fn and window_mapping_fn.
Review comment:
NoActionRequiredNit: I don't like that this gets to 6 indentations deep, which is a bit of a readability concern, but I'm not certain that moving this to a helper function would be that helpful. We'd probably need to break it into two helpers, one from handling the graphx.URNDoFn, and then this graph.ParDo block.
These are a changes for another PR though.
##########
File path: sdks/go/pkg/beam/core/runtime/exec/translate.go
##########
@@ -372,32 +386,43 @@ func (b *builder) makeLink(from string, id linkID) (Node, error) {
switch op {
case graph.ParDo:
- n := &ParDo{UID: b.idgen.New(), Inbound: in, Out: out}
- n.Fn, err = graph.AsDoFn(fn, graph.MainUnknown)
+ dofn, err := graph.AsDoFn(fn, graph.MainUnknown)
if err != nil {
return nil, err
}
- n.PID = transform.GetUniqueName()
-
- input := unmarshalKeyedValues(transform.GetInputs())
- for i := 1; i < len(input); i++ {
- // TODO(herohde) 8/8/2018: handle different windows, view_fn and window_mapping_fn.
- // For now, assume we don't need any information in the pardo payload.
-
- ec, wc, err := b.makeCoderForPCollection(input[i])
- if err != nil {
- return nil, err
+ switch urn {
+ case urnPairWithRestriction:
+ u = &PairWithRestriction{UID: b.idgen.New(), Fn: dofn, Out: out[0]}
+ case urnSplitAndSizeRestrictions:
+ u = &SplitAndSizeRestrictions{UID: b.idgen.New(), Fn: dofn, Out: out[0]}
+ default:
+ n := &ParDo{UID: b.idgen.New(), Fn: dofn, Inbound: in, Out: out}
+ n.PID = transform.GetUniqueName()
+
+ input := unmarshalKeyedValues(transform.GetInputs())
+ for i := 1; i < len(input); i++ {
+ // TODO(herohde) 8/8/2018: handle different windows, view_fn and window_mapping_fn.
+ // For now, assume we don't need any information in the pardo payload.
+
+ ec, wc, err := b.makeCoderForPCollection(input[i])
+ if err != nil {
+ return nil, err
+ }
+
+ sid := StreamID{
+ Port: Port{URL: b.desc.GetStateApiServiceDescriptor().GetUrl()},
+ PtransformID: id.to,
+ }
+ sideInputID := fmt.Sprintf("i%v", i) // SideInputID (= local id, "iN")
+ side := NewSideInputAdapter(sid, sideInputID, coder.NewW(ec, wc))
+ n.Side = append(n.Side, side)
}
-
- sid := StreamID{
- Port: Port{URL: b.desc.GetStateApiServiceDescriptor().GetUrl()},
- PtransformID: id.to,
+ if urn == urnProcessSizedElementsAndRestrictions {
+ u = &ProcessSizedElementsAndRestrictions{PDo: n}
+ } else {
+ u = n
}
Review comment:
Consider moving the common case to always be used, and then corrected afterwards.
```suggestion
u = n
if urn == urnProcessSizedElementsAndRestrictions {
u = &ProcessSizedElementsAndRestrictions{PDo: n}
}
```
This idiom avoids the else block, and indicates that if there were a ?: operator in Go, that's what would happen. The compiler will optimize this appropriately.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
users@infra.apache.org