// applyWindow applies windowing to sg.sorted.
func (sg *SubGraph) applyPagination(ctx context.Context) error {
params := sg.Params
if params.Count == 0 && params.Offset == 0 { // No pagination.
return nil
}
x.AssertTrue(len(sg.SrcUIDs.Uids) == len(sg.uidMatrix))
for _, l := range sg.uidMatrix {
algo.IntersectWith(l, sg.DestUIDs)
start, end := pageRange(&sg.Params, len(l.Uids))
l.Uids = l.Uids[start:end]
}
// Re-merge the UID matrix.
sg.DestUIDs = algo.MergeSorted(sg.uidMatrix)
return nil
}
// ProcessGraph processes the SubGraph instance accumulating result for the query
// from different instances. Note: taskQuery is nil for root node.
func ProcessGraph(ctx context.Context, sg, parent *SubGraph, rch chan error) {
var err error
if len(sg.Attr) == 0 {
// If we have a filter SubGraph which only contains an operator,
// it won't have any attribute to work on.
// This is to allow providing SrcUIDs to the filter children.
sg.DestUIDs = sg.SrcUIDs
} else if parent == nil && len(sg.SrcFunc) == 0 {
// I am root. I don't have any function to execute, and my
// result has been prepared for me already.
sg.DestUIDs = algo.MergeSorted(sg.uidMatrix) // Could also be = sg.SrcUIDs
} else {
taskQuery := createTaskQuery(sg)
result, err := worker.ProcessTaskOverNetwork(ctx, taskQuery)
if err != nil {
x.TraceError(ctx, x.Wrapf(err, "Error while processing task"))
rch <- err
return
}
sg.uidMatrix = result.UidMatrix
sg.values = result.Values
if len(sg.values) > 0 {
v := sg.values[0]
x.Trace(ctx, "Sample value for attr: %v Val: %v", sg.Attr, string(v.Val))
}
sg.counts = result.Counts
if sg.Params.DoCount && len(sg.Filters) == 0 {
// If there is a filter, we need to do more work to get the actual count.
x.Trace(ctx, "Zero uids. Only count requested")
rch <- nil
return
}
if result.IntersectDest {
sg.DestUIDs = algo.IntersectSorted(result.UidMatrix)
} else {
sg.DestUIDs = algo.MergeSorted(result.UidMatrix)
}
}
if len(sg.DestUIDs.Uids) == 0 {
// Looks like we're done here. Be careful with nil srcUIDs!
x.Trace(ctx, "Zero uids for %q. Num attr children: %v", sg.Attr, len(sg.Children))
rch <- nil
return
}
// Apply filters if any.
if len(sg.Filters) > 0 {
// Run all filters in parallel.
filterChan := make(chan error, len(sg.Filters))
for _, filter := range sg.Filters {
filter.SrcUIDs = sg.DestUIDs
go ProcessGraph(ctx, filter, sg, filterChan)
}
for _ = range sg.Filters {
select {
case err = <-filterChan:
if err != nil {
x.TraceError(ctx, x.Wrapf(err, "Error while processing filter task"))
rch <- err
return
}
case <-ctx.Done():
x.TraceError(ctx, x.Wrapf(ctx.Err(), "Context done before full execution"))
rch <- ctx.Err()
return
}
}
// Now apply the results from filter.
var lists []*task.List
for _, filter := range sg.Filters {
lists = append(lists, filter.DestUIDs)
}
if sg.FilterOp == "|" {
sg.DestUIDs = algo.MergeSorted(lists)
} else {
sg.DestUIDs = algo.IntersectSorted(lists)
}
}
if len(sg.Params.Order) == 0 {
// There is no ordering. Just apply pagination and return.
if err = sg.applyPagination(ctx); err != nil {
rch <- err
return
}
} else {
// We need to sort first before pagination.
if err = sg.applyOrderAndPagination(ctx); err != nil {
rch <- err
return
}
}
// Here we consider handling _count_ with filtering. We do this after
// pagination because otherwise, we need to do the count with pagination
// taken into account. For example, a PL might have only 50 entries but the
// user wants to skip 100 entries and return 10 entries. In this case, you
// should return a count of 0, not 10.
if sg.Params.DoCount {
x.AssertTrue(len(sg.Filters) > 0)
sg.counts = make([]uint32, len(sg.uidMatrix))
for i, ul := range sg.uidMatrix {
// A possible optimization is to return the size of the intersection
// without forming the intersection.
algo.IntersectWith(ul, sg.DestUIDs)
sg.counts[i] = uint32(len(ul.Uids))
}
rch <- nil
return
}
childChan := make(chan error, len(sg.Children))
for i := 0; i < len(sg.Children); i++ {
child := sg.Children[i]
child.SrcUIDs = sg.DestUIDs // Make the connection.
go ProcessGraph(ctx, child, sg, childChan)
}
// Now get all the results back.
for _ = range sg.Children {
select {
case err = <-childChan:
if err != nil {
x.TraceError(ctx, x.Wrapf(err, "Error while processing child task"))
rch <- err
return
}
case <-ctx.Done():
x.TraceError(ctx, x.Wrapf(ctx.Err(), "Context done before full execution"))
rch <- ctx.Err()
return
}
}
rch <- nil
}
