featurebase/client/api.go
Travis Turner 00ef2380e5
Batch insert via SQL (multiple tuples) (#2243)
* Formatting adjustments made during code review.

While reviewing the BULK INSERT logic (in order to decide how best to
approach "ingest via sql" in the cloud), I made a few formatting and
comment changes. I'm just adding them here as a separate commit so they
don't muddy up my actual work.

* Parser modifications to support mulitple tuples in INSERT INTO

This commit doesn't include all of the changes required in the
planner. Fow now, the planner is simply modified to continue supporting
a single tuple (the first tuple in the list).

* Update the planner to handle multiple INSERT INTO tuples

This is part 1. It's still using the existing logic which builds an
ImportRequest for every record (and every field!).

The next step will involve using a client.Batch to handle the records.

* Introduce client.Importer interface (used by client.Batch)

Instead of the Batch having a pointer to a client, this puts an
interface there instead (which the client implements). It also allows us
to inject a different importer (i.e. other than a featurebase.client)
into the Batch.

* Decouple batch from client

This commit pulls batch-specific code out of the client package and into
a new batch package. It introduces the batch.Importer interface, the
methods of which replace all the calls that batch was previously making
directly to client methods.

Finally, it contains two implementations of the batch.Importer
interface: one is a wrapper around client, and the other is a wrapper
around featurebase.API.

* Use docker (instead of MustRunCluster) for internal batch tests

Because the `batch` package tests are internal, using
test.MustRunCluster() resulted in an import loop (because it eventually
imports `server`, and we can't have that). So this commit replaces the
use of `test.MustRunCluster()` with docker. The setup is basically the
same as that used in the idk docker tests.

Here we also remove all client-side references to `UseIngestAPI`, which
is an experimental (json) ingest api. It's still suppored on the server,
but here we remove the external usage of it.

* cherry-pick fix

* Use batch.Import() for sql3 INSERT INTO statements

* Thread logger into sql3

* fix batch test

* Fix some shadowing complaint by linter

* Address some test issues related to stringsets

* Exclude batch integration tests from CI

* Address PR feedback

- Added description to batch.README
- Consolidated grep commands in .gitlab-ci.yml
- Removed some debugging comments
- Replaces some inadvertantly removed license headers

* Add batch package to gitlab CI

* Updated CI for batch package

Updated CI include path

Update gitlab ci

Update CI

Update CI

Trying new include path for ci

Updated gitlab ci include path

Made idk race job optional for sonarcloud upload

add testdata directory

remove testenv from dockercompose file

use GIT_STRATEGY clone in batch CI

add testdata volume to dockercompose

Co-authored-by: Fletcher Haynes <fletcher.haynes@generalassemb.ly>
2022-10-29 14:24:33 -05:00

275 lines
7.3 KiB
Go

package client
import (
"context"
featurebase "github.com/molecula/featurebase/v3"
"github.com/molecula/featurebase/v3/errors"
)
var _ featurebase.SchemaAPI = &schemaAPI{}
// schemaAPI is a featurebase client wrapper which implements the
// featurebase.SchemaAPI interface. This was introduced for use in batch tests
// when we decoupled Batch from the client package. In other words, this is only
// used for those tests, it may not be functionally complete, and should not be
// used otherwise without further testing and review of this code.
type schemaAPI struct {
*Client
}
func NewSchemaAPI(c *Client) *schemaAPI {
return &schemaAPI{
Client: c,
}
}
func (s *schemaAPI) CreateIndexAndFields(ctx context.Context, indexName string, options featurebase.IndexOptions, fields []featurebase.CreateFieldObj) error {
schema, err := s.Client.Schema()
if err != nil {
return errors.Wrap(err, "getting schema")
}
// Add the index.
idx := schema.Index(indexName,
OptIndexKeys(options.Keys),
OptIndexTrackExistence(true),
)
if err := s.Client.CreateIndex(idx); err != nil {
return errors.Wrap(err, "creating index")
}
// Now add fields.
for _, f := range fields {
fld, err := s.addFieldToIndex(idx, f.Name, f.Options...)
if err != nil {
return errors.Wrapf(err, "adding field to index")
}
if err := s.Client.CreateField(fld); err != nil {
return errors.Wrapf(err, "creating field")
}
}
return nil
}
func (s *schemaAPI) CreateField(ctx context.Context, indexName string, fieldName string, opts ...featurebase.FieldOption) (*featurebase.Field, error) {
schema, err := s.Client.Schema()
if err != nil {
return nil, errors.Wrap(err, "getting schema")
}
if !schema.HasIndex(indexName) {
return nil, featurebase.ErrIndexNotFound
}
idx := schema.Index(indexName)
fld, err := s.addFieldToIndex(idx, fieldName, opts...)
if err != nil {
return nil, errors.Wrapf(err, "adding field to index")
}
if err := s.Client.CreateField(fld); err != nil {
return nil, errors.Wrapf(err, "creating field")
}
return nil, nil
}
func (s *schemaAPI) addFieldToIndex(idx *Index, fieldName string, opts ...featurebase.FieldOption) (*Field, error) {
ffos := &featurebase.FieldOptions{}
for _, opt := range opts {
opt(ffos)
}
cfos := []FieldOption{}
switch ffos.Type {
case featurebase.FieldTypeBool:
cfos = append(cfos, OptFieldTypeBool())
case featurebase.FieldTypeInt:
cfos = append(cfos, OptFieldTypeInt(ffos.Min.ToInt64(0), ffos.Max.ToInt64(0)))
case featurebase.FieldTypeSet:
cfos = append(cfos,
OptFieldTypeSet(CacheType(ffos.CacheType), int(ffos.CacheSize)),
OptFieldKeys(ffos.Keys),
)
case featurebase.FieldTypeMutex:
cfos = append(cfos,
OptFieldTypeMutex(CacheType(ffos.CacheType), int(ffos.CacheSize)),
OptFieldKeys(ffos.Keys),
)
case featurebase.FieldTypeDecimal:
cfos = append(cfos, OptFieldTypeDecimal(ffos.Scale, ffos.Min, ffos.Max))
case featurebase.FieldTypeTime:
cfos = append(cfos,
OptFieldTypeTime(TimeQuantum(ffos.TimeQuantum), ffos.NoStandardView),
OptFieldKeys(ffos.Keys),
)
case featurebase.FieldTypeTimestamp:
cfos = append(cfos, OptFieldTypeTimestamp(featurebase.DefaultEpoch, ffos.TimeUnit))
default:
return nil, errors.Errorf("unsupported field type: %s", ffos.Type)
}
return idx.Field(fieldName, cfos...), nil
}
func (s *schemaAPI) DeleteField(ctx context.Context, indexName string, fieldName string) error {
schema, err := s.Client.Schema()
if err != nil {
return errors.Wrap(err, "getting schema")
}
if !schema.HasIndex(indexName) {
return featurebase.ErrIndexNotFound
}
idx := schema.Index(indexName)
return s.Client.DeleteField(&Field{
name: fieldName,
index: idx,
})
}
func (s *schemaAPI) DeleteIndex(ctx context.Context, indexName string) error {
return s.Client.DeleteIndexByName(indexName)
}
func (s *schemaAPI) IndexInfo(ctx context.Context, indexName string) (*featurebase.IndexInfo, error) {
schema, err := s.Client.Schema()
if err != nil {
return nil, errors.Wrap(err, "getting schema")
}
if !schema.HasIndex(indexName) {
return nil, featurebase.ErrIndexNotFound
}
idx := schema.Index(indexName)
return FromClientIndex(idx), nil
}
func (s *schemaAPI) Schema(ctx context.Context, withViews bool) ([]*featurebase.IndexInfo, error) {
return nil, errors.New("", "schemaAPI.Schema is not implemented")
}
var _ featurebase.QueryAPI = &queryAPI{}
// queryAPI is a featurebase client wrapper which implements the
// featurebase.QueryAPI interface. This was introduced for use in batch tests
// when we decoupled Batch from the client package. In other words, this is only
// used for those tests, it may not be functionally complete, and should not be
// used otherwise without further testing and review of this code.
type queryAPI struct {
*Client
}
func NewQueryAPI(c *Client) *queryAPI {
return &queryAPI{
Client: c,
}
}
func (q *queryAPI) Query(ctx context.Context, req *featurebase.QueryRequest) (featurebase.QueryResponse, error) {
schema, err := q.Client.Schema()
if err != nil {
return featurebase.QueryResponse{}, errors.Wrap(err, "getting schema")
}
if !schema.HasIndex(req.Index) {
return featurebase.QueryResponse{}, featurebase.ErrIndexNotFound
}
idx := schema.Index(req.Index)
qry := NewPQLBaseQuery(req.Query, idx, nil)
res, err := q.Client.Query(qry)
if err != nil {
return featurebase.QueryResponse{}, errors.Wrap(err, "querying client")
}
var rerr error
if res.ErrorMessage != "" {
rerr = errors.New("", res.ErrorMessage)
}
fbResults := make([]interface{}, 0)
// Row, PairsField, GroupCounts
for _, result := range res.ResultList {
switch result.Type() {
case QueryResultTypeRow:
if len(result.Row().Keys) > 0 {
row := featurebase.NewRow()
row.Keys = result.Row().Keys
fbResults = append(fbResults, row)
} else {
fbResults = append(fbResults, featurebase.NewRow(result.Row().Columns...))
}
case QueryResultTypeUint64:
fbResults = append(fbResults, uint64(result.Count()))
case QueryResultTypeBool:
fbResults = append(fbResults, result.Changed())
case QueryResultTypePairsField:
pairs := []featurebase.Pair{}
for _, ci := range result.CountItems() {
pairs = append(pairs, featurebase.Pair{
ID: ci.ID,
Key: ci.Key,
Count: ci.Count,
})
}
pf := &featurebase.PairsField{
Pairs: pairs,
Field: "",
}
fbResults = append(fbResults, pf)
case QueryResultTypeGroupCounts:
groups := []featurebase.GroupCount{}
for _, grpCnt := range result.GroupCounts() {
fieldRows := []featurebase.FieldRow{}
for _, fr := range grpCnt.Groups {
fieldRows = append(fieldRows, featurebase.FieldRow{
Field: fr.FieldName,
RowID: fr.RowID,
RowKey: fr.RowKey,
Value: fr.Value,
})
}
groups = append(groups, featurebase.GroupCount{
Group: fieldRows,
Count: uint64(grpCnt.Count),
Agg: grpCnt.Agg,
// DecimalAgg: ??,
})
}
gc := featurebase.NewGroupCounts("", groups...)
fbResults = append(fbResults, gc)
case QueryResultTypeValCount:
vc := featurebase.ValCount{
Val: result.Value(),
Count: result.Count(),
}
fbResults = append(fbResults, vc)
default:
return featurebase.QueryResponse{}, errors.Errorf("unsupported query result type: %d", result.Type())
}
}
resp := &featurebase.QueryResponse{
Results: fbResults,
Err: rerr,
}
return *resp, nil
}