Like Prometheus, but for logs.
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
loki/pkg/engine/internal/executor/executor.go

461 lines
14 KiB

package executor
import (
"context"
"errors"
"fmt"
"strings"
"github.com/go-kit/log"
"github.com/grafana/dskit/user"
"github.com/thanos-io/objstore"
"go.opentelemetry.io/otel"
"go.opentelemetry.io/otel/attribute"
"github.com/grafana/loki/v3/pkg/dataobj"
refactor(dataobj): invert dependency between dataobj and sections (#17762) Originally, the dataobj package was a higher-level API around sections. This design caused it to become a bottleneck: * Implementing any new public behaviour for a section required bubbling it up to the dataobj API for it to be exposed, making it tedious to add new sections or update existing ones. * The `dataobj.Builder` pattern was focused on constructing dataobjs for storing log data, which will cause friction as we build objects around other use cases. This PR builds on top of the foundation laid out by #17704 and #17708, fully inverting the dependency between dataobj and sections: * The `dataobj` package has no knowledge of what sections exist, and can now be used for writing and reading generic sections. Section packages now create higher-level APIs around the abstractions provided by `dataobj`. * Section packages are now public, and callers interact directly with these packages for writing and reading section-specific data. * All logic for a section (encoding, decoding, buffering, reading) is now fully self-contained inside the section package. Previously, the implementation of each section was spread across three packages (`pkg/dataobj/internal/encoding`, `pkg/dataobj/internal/sections/SECTION`, `pkg/dataobj`). * Cutting a section is now a decision made by the caller rather than the section implementation. Previously, the logs section builder would create multiple sections. For the most part, this change is a no-op, with two exceptions: 1. Section cutting is now performed by the caller; however, this shouldn't result in any issues. 2. Removing the high-level `dataobj.Stream` and `dataobj.Record` types will temporarily reduce the allocation gains from #16988. I will address this after this PR is merged.
7 months ago
"github.com/grafana/loki/v3/pkg/dataobj/sections/logs"
"github.com/grafana/loki/v3/pkg/dataobj/sections/streams"
"github.com/grafana/loki/v3/pkg/engine/internal/planner/physical"
"github.com/grafana/loki/v3/pkg/xcap"
)
var tracer = otel.Tracer("pkg/engine/internal/executor")
type Config struct {
BatchSize int64
Bucket objstore.Bucket
MergePrefetchCount int
// GetExternalInputs is an optional function called for each node in the
// plan. If GetExternalInputs returns a non-nil slice of Pipelines, they
// will be used as inputs to the pipeline of node.
GetExternalInputs func(ctx context.Context, node physical.Node) []Pipeline
}
func Run(ctx context.Context, cfg Config, plan *physical.Plan, logger log.Logger) Pipeline {
c := &Context{
plan: plan,
batchSize: cfg.BatchSize,
mergePrefetchCount: cfg.MergePrefetchCount,
bucket: cfg.Bucket,
logger: logger,
evaluator: newExpressionEvaluator(),
getExternalInputs: cfg.GetExternalInputs,
}
if plan == nil {
return errorPipeline(ctx, errors.New("plan is nil"))
}
node, err := plan.Root()
if err != nil {
return errorPipeline(ctx, err)
}
return c.execute(ctx, node)
}
// Context is the execution context
type Context struct {
batchSize int64
logger log.Logger
plan *physical.Plan
evaluator expressionEvaluator
bucket objstore.Bucket
getExternalInputs func(ctx context.Context, node physical.Node) []Pipeline
mergePrefetchCount int
}
func (c *Context) execute(ctx context.Context, node physical.Node) Pipeline {
// Start a new xcap.Region for this node.
// Region is created in preorder traversal to maintain the parent-child relationship.
ctx, nodeRegion := startRegionForNode(ctx, node)
children := c.plan.Children(node)
inputs := make([]Pipeline, 0, len(children))
for _, child := range children {
inputs = append(inputs, c.execute(ctx, child))
}
if c.getExternalInputs != nil {
inputs = append(inputs, c.getExternalInputs(ctx, node)...)
}
switch n := node.(type) {
case *physical.DataObjScan:
// DataObjScan reads from object storage to determine the full pipeline to
// construct, making it expensive to call during planning time.
//
// TODO(rfratto): find a way to remove the logic from executeDataObjScan
// which wraps the pipeline with a topk/limit without reintroducing
// planning cost for thousands of scan nodes.
return newLazyPipeline(func(ctx context.Context, _ []Pipeline) Pipeline {
return newObservedPipeline(c.executeDataObjScan(ctx, n, nodeRegion))
}, inputs)
case *physical.TopK:
return newObservedPipeline(c.executeTopK(ctx, n, inputs, nodeRegion))
case *physical.Limit:
return newObservedPipeline(c.executeLimit(ctx, n, inputs, nodeRegion))
case *physical.Filter:
return newObservedPipeline(c.executeFilter(ctx, n, inputs, nodeRegion))
case *physical.Projection:
return newObservedPipeline(c.executeProjection(ctx, n, inputs, nodeRegion))
case *physical.RangeAggregation:
return newObservedPipeline(c.executeRangeAggregation(ctx, n, inputs, nodeRegion))
case *physical.VectorAggregation:
return newObservedPipeline(c.executeVectorAggregation(ctx, n, inputs, nodeRegion))
case *physical.ColumnCompat:
return newObservedPipeline(c.executeColumnCompat(ctx, n, inputs, nodeRegion))
case *physical.Parallelize:
return c.executeParallelize(ctx, n, inputs)
case *physical.ScanSet:
return c.executeScanSet(ctx, n, nodeRegion)
default:
return errorPipeline(ctx, fmt.Errorf("invalid node type: %T", node))
}
}
func (c *Context) executeDataObjScan(ctx context.Context, node *physical.DataObjScan, region *xcap.Region) Pipeline {
if c.bucket == nil {
return errorPipeline(ctx, errors.New("no object store bucket configured"))
}
obj, err := dataobj.FromBucket(ctx, c.bucket, string(node.Location))
if err != nil {
return errorPipeline(ctx, fmt.Errorf("creating data object: %w", err))
}
region.AddEvent("opened dataobj")
var (
streamsSection *streams.Section
logsSection *logs.Section
)
tenant, err := user.ExtractOrgID(ctx)
if err != nil {
return errorPipeline(ctx, fmt.Errorf("missing org ID: %w", err))
}
for _, sec := range obj.Sections().Filter(streams.CheckSection) {
if sec.Tenant != tenant {
continue
}
if streamsSection != nil {
return errorPipeline(ctx, fmt.Errorf("multiple streams sections found in data object %q", node.Location))
}
var err error
streamsSection, err = streams.Open(ctx, sec)
if err != nil {
return errorPipeline(ctx, fmt.Errorf("opening streams section %q: %w", sec.Type, err))
}
region.AddEvent("opened streams section")
}
if streamsSection == nil {
return errorPipeline(ctx, fmt.Errorf("streams section not found in data object %q", node.Location))
}
for i, sec := range obj.Sections().Filter(logs.CheckSection) {
if i != node.Section {
continue
}
var err error
logsSection, err = logs.Open(ctx, sec)
if err != nil {
return errorPipeline(ctx, fmt.Errorf("opening logs section %q: %w", sec.Type, err))
}
region.AddEvent("opened logs section")
break
}
if logsSection == nil {
return errorPipeline(ctx, fmt.Errorf("logs section %d not found in data object %q", node.Section, node.Location))
}
predicates := make([]logs.Predicate, 0, len(node.Predicates))
for _, p := range node.Predicates {
conv, err := buildLogsPredicate(p, logsSection.Columns())
if err != nil {
return errorPipeline(ctx, err)
}
predicates = append(predicates, conv)
}
region.AddEvent("constructed predicate")
var pipeline Pipeline = newDataobjScanPipeline(dataobjScanOptions{
// TODO(rfratto): passing the streams section means that each DataObjScan
// will read the entire streams section (for IDs being loaded), which is
// going to be quite a bit of wasted effort.
//
// Longer term, there should be a dedicated plan node which handles joining
// streams and log records based on StreamID, which is shared between all
// sections in the same object.
StreamsSection: streamsSection,
LogsSection: logsSection,
StreamIDs: node.StreamIDs,
Predicates: predicates,
Projections: node.Projections,
BatchSize: c.batchSize,
}, log.With(c.logger, "location", string(node.Location), "section", node.Section), region)
return pipeline
}
func (c *Context) executeTopK(ctx context.Context, topK *physical.TopK, inputs []Pipeline, region *xcap.Region) Pipeline {
if len(inputs) == 0 {
return emptyPipeline()
}
pipeline, err := newTopkPipeline(topkOptions{
Inputs: inputs,
SortBy: []physical.ColumnExpression{topK.SortBy},
Ascending: topK.Ascending,
NullsFirst: topK.NullsFirst,
K: topK.K,
MaxUnused: int(c.batchSize) * 2,
Region: region,
})
if err != nil {
return errorPipeline(ctx, err)
}
return pipeline
}
func (c *Context) executeLimit(ctx context.Context, limit *physical.Limit, inputs []Pipeline, region *xcap.Region) Pipeline {
if len(inputs) == 0 {
return emptyPipeline()
}
if len(inputs) > 1 {
return errorPipeline(ctx, fmt.Errorf("limit expects exactly one input, got %d", len(inputs)))
}
return NewLimitPipeline(inputs[0], limit.Skip, limit.Fetch, region)
}
func (c *Context) executeFilter(ctx context.Context, filter *physical.Filter, inputs []Pipeline, region *xcap.Region) Pipeline {
if len(inputs) == 0 {
return emptyPipeline()
}
if len(inputs) > 1 {
return errorPipeline(ctx, fmt.Errorf("filter expects exactly one input, got %d", len(inputs)))
}
return NewFilterPipeline(filter, inputs[0], c.evaluator, region)
}
func (c *Context) executeProjection(ctx context.Context, proj *physical.Projection, inputs []Pipeline, region *xcap.Region) Pipeline {
if len(inputs) == 0 {
return emptyPipeline()
}
if len(inputs) > 1 {
// unsupported for now
return errorPipeline(ctx, fmt.Errorf("projection expects exactly one input, got %d", len(inputs)))
}
if len(proj.Expressions) == 0 {
return errorPipeline(ctx, fmt.Errorf("projection expects at least one expression, got 0"))
}
p, err := NewProjectPipeline(inputs[0], proj, &c.evaluator, region)
if err != nil {
return errorPipeline(ctx, err)
}
return p
}
func (c *Context) executeRangeAggregation(ctx context.Context, plan *physical.RangeAggregation, inputs []Pipeline, region *xcap.Region) Pipeline {
if len(inputs) == 0 {
return emptyPipeline()
}
pipeline, err := newRangeAggregationPipeline(inputs, c.evaluator, rangeAggregationOptions{
partitionBy: plan.PartitionBy,
startTs: plan.Start,
endTs: plan.End,
rangeInterval: plan.Range,
step: plan.Step,
operation: plan.Operation,
}, region)
if err != nil {
return errorPipeline(ctx, err)
}
return pipeline
}
func (c *Context) executeVectorAggregation(ctx context.Context, plan *physical.VectorAggregation, inputs []Pipeline, region *xcap.Region) Pipeline {
if len(inputs) == 0 {
return emptyPipeline()
}
pipeline, err := newVectorAggregationPipeline(inputs, plan.GroupBy, c.evaluator, plan.Operation, region)
if err != nil {
return errorPipeline(ctx, err)
}
return pipeline
}
func (c *Context) executeColumnCompat(ctx context.Context, compat *physical.ColumnCompat, inputs []Pipeline, region *xcap.Region) Pipeline {
if len(inputs) == 0 {
return emptyPipeline()
}
if len(inputs) > 1 {
return errorPipeline(ctx, fmt.Errorf("columncompat expects exactly one input, got %d", len(inputs)))
}
return newColumnCompatibilityPipeline(compat, inputs[0], region)
}
func (c *Context) executeParallelize(ctx context.Context, _ *physical.Parallelize, inputs []Pipeline) Pipeline {
if len(inputs) == 0 {
return emptyPipeline()
} else if len(inputs) > 1 {
return errorPipeline(ctx, fmt.Errorf("parallelize expects exactly one input, got %d", len(inputs)))
}
// Parallelize is a hint node to the scheduler for parallel execution. If we
// see an Parallelize node in the plan, we ignore it and immediately
// propagate up the input.
return inputs[0]
}
func (c *Context) executeScanSet(ctx context.Context, set *physical.ScanSet, _ *xcap.Region) Pipeline {
// ScanSet typically gets partitioned by the scheduler into multiple scan
// nodes.
//
// However, for locally testing unpartitioned pipelines, we still supprt
// running a ScanSet. In this case, we treat internally execute it as a
// Merge on top of multiple sequential scans.
ctx, mergeRegion := xcap.StartRegion(ctx, physical.NodeTypeMerge.String())
var targets []Pipeline
for _, target := range set.Targets {
switch target.Type {
case physical.ScanTypeDataObject:
// Make sure projections and predicates get passed down to the
// individual scan.
partition := target.DataObject
partition.Predicates = set.Predicates
partition.Projections = set.Projections
nodeCtx, partitionRegion := startRegionForNode(ctx, partition)
targets = append(targets, newLazyPipeline(func(_ context.Context, _ []Pipeline) Pipeline {
return newObservedPipeline(c.executeDataObjScan(nodeCtx, partition, partitionRegion))
}, nil))
default:
return errorPipeline(ctx, fmt.Errorf("unrecognized ScanSet target %s", target.Type))
}
}
if len(targets) == 0 {
return emptyPipeline()
}
pipeline, err := newMergePipeline(targets, c.mergePrefetchCount, mergeRegion)
if err != nil {
return errorPipeline(ctx, err)
}
return pipeline
}
// startRegionForNode starts xcap.Region for the given physical plan node.
// It internally calls xcap.StartRegion with attributes relevant to the node type.
func startRegionForNode(ctx context.Context, n physical.Node) (context.Context, *xcap.Region) {
// Include node ID in the region attributes to retain a link to the physical plan node.
attributes := []attribute.KeyValue{
attribute.String("node_id", n.ID().String()),
}
switch n := n.(type) {
case *physical.DataObjScan:
attributes = append(attributes,
attribute.String("location", string(n.Location)),
attribute.Int("section", n.Section),
attribute.Int("num_stream_ids", len(n.StreamIDs)),
attribute.Int("num_predicates", len(n.Predicates)),
attribute.Int("num_projections", len(n.Projections)),
)
case *physical.TopK:
attributes = append(attributes,
attribute.Int("k", n.K),
attribute.Bool("ascending", n.Ascending),
attribute.Bool("nulls_first", n.NullsFirst),
)
if n.SortBy != nil {
attributes = append(attributes, attribute.Stringer("sort_by", n.SortBy))
}
case *physical.Limit:
attributes = append(attributes,
attribute.Int("skip", int(n.Skip)),
attribute.Int("fetch", int(n.Fetch)),
)
case *physical.Filter:
attributes = append(attributes,
attribute.Int("num_predicates", len(n.Predicates)),
)
case *physical.Projection:
attributes = append(attributes,
attribute.Int("num_expressions", len(n.Expressions)),
attribute.Bool("all", n.All),
attribute.Bool("drop", n.Drop),
attribute.Bool("expand", n.Expand),
)
case *physical.RangeAggregation:
attributes = append(attributes,
attribute.String("operation", string(rune(n.Operation))),
attribute.Int64("start_ts", n.Start.UnixNano()),
attribute.Int64("end_ts", n.End.UnixNano()),
attribute.Int64("range_interval", int64(n.Range)),
attribute.Int64("step", int64(n.Step)),
attribute.Int("num_partition_by", len(n.PartitionBy)),
)
case *physical.VectorAggregation:
attributes = append(attributes,
attribute.String("operation", string(rune(n.Operation))),
attribute.Int("num_group_by", len(n.GroupBy)),
)
case *physical.ColumnCompat:
collisionStrs := make([]string, len(n.Collisions))
for i, ct := range n.Collisions {
collisionStrs[i] = ct.String()
}
attributes = append(attributes,
attribute.String("src", n.Source.String()),
attribute.String("dst", n.Destination.String()),
attribute.String("collisions", fmt.Sprintf("[%s]", strings.Join(collisionStrs, ", "))),
)
case *physical.ScanSet:
attributes = append(attributes,
attribute.Int("num_targets", len(n.Targets)),
attribute.Int("num_predicates", len(n.Predicates)),
attribute.Int("num_projections", len(n.Projections)),
)
default:
// do nothing.
}
return xcap.StartRegion(ctx, n.Type().String(), xcap.WithRegionAttributes(attributes...))
}