// Plan creates an execution plan for the given SelectStatement and returns an Executor.
func (q *QueryExecutor) plan(stmt *influxql.SelectStatement, chunkSize int) (Executor, error) {
shards := map[uint64]meta.ShardInfo{} // Shards requiring mappers.
// Replace instances of "now()" with the current time, and check the resultant times.
stmt.Condition = influxql.Reduce(stmt.Condition, &influxql.NowValuer{Now: time.Now().UTC()})
tmin, tmax := influxql.TimeRange(stmt.Condition)
if tmax.IsZero() {
tmax = time.Now()
}
if tmin.IsZero() {
tmin = time.Unix(0, 0)
}
for _, src := range stmt.Sources {
mm, ok := src.(*influxql.Measurement)
if !ok {
return nil, fmt.Errorf("invalid source type: %#v", src)
}
// Build the set of target shards. Using shard IDs as keys ensures each shard ID
// occurs only once.
shardGroups, err := q.MetaStore.ShardGroupsByTimeRange(mm.Database, mm.RetentionPolicy, tmin, tmax)
if err != nil {
return nil, err
}
for _, g := range shardGroups {
for _, sh := range g.Shards {
shards[sh.ID] = sh
}
}
}
// Build the Mappers, one per shard.
mappers := []Mapper{}
for _, sh := range shards {
m, err := q.ShardMapper.CreateMapper(sh, stmt.String(), chunkSize)
if err != nil {
return nil, err
}
if m == nil {
// No data for this shard, skip it.
continue
}
mappers = append(mappers, m)
}
var executor Executor
if len(mappers) > 0 {
// All Mapper are of same type, so check first to determine correct Executor type.
if _, ok := mappers[0].(*RawMapper); ok {
executor = NewRawExecutor(stmt, mappers, chunkSize)
} else {
executor = NewAggregateExecutor(stmt, mappers)
}
} else {
// With no mappers, the Executor type doesn't matter.
executor = NewRawExecutor(stmt, nil, chunkSize)
}
return executor, nil
}
// Ensure the time range of an expression can be extracted.
func TestTimeRange(t *testing.T) {
for i, tt := range []struct {
expr string
min, max string
}{
// LHS VarRef
{expr: `time > '2000-01-01 00:00:00'`, min: `2000-01-01T00:00:00.000000001Z`, max: `0001-01-01T00:00:00Z`},
{expr: `time >= '2000-01-01 00:00:00'`, min: `2000-01-01T00:00:00Z`, max: `0001-01-01T00:00:00Z`},
{expr: `time < '2000-01-01 00:00:00'`, min: `0001-01-01T00:00:00Z`, max: `1999-12-31T23:59:59.999999999Z`},
{expr: `time <= '2000-01-01 00:00:00'`, min: `0001-01-01T00:00:00Z`, max: `2000-01-01T00:00:00Z`},
// RHS VarRef
{expr: `'2000-01-01 00:00:00' > time`, min: `0001-01-01T00:00:00Z`, max: `1999-12-31T23:59:59.999999999Z`},
{expr: `'2000-01-01 00:00:00' >= time`, min: `0001-01-01T00:00:00Z`, max: `2000-01-01T00:00:00Z`},
{expr: `'2000-01-01 00:00:00' < time`, min: `2000-01-01T00:00:00.000000001Z`, max: `0001-01-01T00:00:00Z`},
{expr: `'2000-01-01 00:00:00' <= time`, min: `2000-01-01T00:00:00Z`, max: `0001-01-01T00:00:00Z`},
// Equality
{expr: `time = '2000-01-01 00:00:00'`, min: `2000-01-01T00:00:00Z`, max: `2000-01-01T00:00:00Z`},
// Multiple time expressions.
{expr: `time >= '2000-01-01 00:00:00' AND time < '2000-01-02 00:00:00'`, min: `2000-01-01T00:00:00Z`, max: `2000-01-01T23:59:59.999999999Z`},
// Min/max crossover
{expr: `time >= '2000-01-01 00:00:00' AND time <= '1999-01-01 00:00:00'`, min: `2000-01-01T00:00:00Z`, max: `1999-01-01T00:00:00Z`},
// Absolute time
{expr: `time = 1388534400s`, min: `2014-01-01T00:00:00Z`, max: `2014-01-01T00:00:00Z`},
// Non-comparative expressions.
{expr: `time`, min: `0001-01-01T00:00:00Z`, max: `0001-01-01T00:00:00Z`},
{expr: `time + 2`, min: `0001-01-01T00:00:00Z`, max: `0001-01-01T00:00:00Z`},
{expr: `time - '2000-01-01 00:00:00'`, min: `0001-01-01T00:00:00Z`, max: `0001-01-01T00:00:00Z`},
{expr: `time AND '2000-01-01 00:00:00'`, min: `0001-01-01T00:00:00Z`, max: `0001-01-01T00:00:00Z`},
} {
// Extract time range.
expr := MustParseExpr(tt.expr)
min, max := influxql.TimeRange(expr)
// Compare with expected min/max.
if min := min.Format(time.RFC3339Nano); tt.min != min {
t.Errorf("%d. %s: unexpected min:\n\nexp=%s\n\ngot=%s\n\n", i, tt.expr, tt.min, min)
continue
}
if max := max.Format(time.RFC3339Nano); tt.max != max {
t.Errorf("%d. %s: unexpected max:\n\nexp=%s\n\ngot=%s\n\n", i, tt.expr, tt.max, max)
continue
}
}
}
// Ensure the SELECT statement can have its start and end time set
func TestSelectStatement_SetTimeRange(t *testing.T) {
q := "SELECT sum(value) from foo where time < now() GROUP BY time(10m)"
stmt, err := influxql.NewParser(strings.NewReader(q)).ParseStatement()
if err != nil {
t.Fatalf("invalid statement: %q: %s", stmt, err)
}
s := stmt.(*influxql.SelectStatement)
min, max := influxql.TimeRange(s.Condition)
start := time.Now().Add(-20 * time.Hour).Round(time.Second).UTC()
end := time.Now().Add(10 * time.Hour).Round(time.Second).UTC()
s.SetTimeRange(start, end)
min, max = influxql.TimeRange(s.Condition)
if min != start {
t.Fatalf("start time wasn't set properly.\n exp: %s\n got: %s", start, min)
}
// the end range is actually one nanosecond before the given one since end is exclusive
end = end.Add(-time.Nanosecond)
if max != end {
t.Fatalf("end time wasn't set properly.\n exp: %s\n got: %s", end, max)
}
// ensure we can set a time on a select that already has one set
start = time.Now().Add(-20 * time.Hour).Round(time.Second).UTC()
end = time.Now().Add(10 * time.Hour).Round(time.Second).UTC()
q = fmt.Sprintf("SELECT sum(value) from foo WHERE time >= %ds and time <= %ds GROUP BY time(10m)", start.Unix(), end.Unix())
stmt, err = influxql.NewParser(strings.NewReader(q)).ParseStatement()
if err != nil {
t.Fatalf("invalid statement: %q: %s", stmt, err)
}
s = stmt.(*influxql.SelectStatement)
min, max = influxql.TimeRange(s.Condition)
if start != min || end != max {
t.Fatalf("start and end times weren't equal:\n exp: %s\n got: %s\n exp: %s\n got:%s\n", start, min, end, max)
}
// update and ensure it saves it
start = time.Now().Add(-40 * time.Hour).Round(time.Second).UTC()
end = time.Now().Add(20 * time.Hour).Round(time.Second).UTC()
s.SetTimeRange(start, end)
min, max = influxql.TimeRange(s.Condition)
// TODO: right now the SetTimeRange can't override the start time if it's more recent than what they're trying to set it to.
// shouldn't matter for our purposes with continuous queries, but fix this later
if min != start {
t.Fatalf("start time wasn't set properly.\n exp: %s\n got: %s", start, min)
}
// the end range is actually one nanosecond before the given one since end is exclusive
end = end.Add(-time.Nanosecond)
if max != end {
t.Fatalf("end time wasn't set properly.\n exp: %s\n got: %s", end, max)
}
// ensure that when we set a time range other where clause conditions are still there
q = "SELECT sum(value) from foo WHERE foo = 'bar' and time < now() GROUP BY time(10m)"
stmt, err = influxql.NewParser(strings.NewReader(q)).ParseStatement()
if err != nil {
t.Fatalf("invalid statement: %q: %s", stmt, err)
}
s = stmt.(*influxql.SelectStatement)
// update and ensure it saves it
start = time.Now().Add(-40 * time.Hour).Round(time.Second).UTC()
end = time.Now().Add(20 * time.Hour).Round(time.Second).UTC()
s.SetTimeRange(start, end)
min, max = influxql.TimeRange(s.Condition)
if min != start {
t.Fatalf("start time wasn't set properly.\n exp: %s\n got: %s", start, min)
}
// the end range is actually one nanosecond before the given one since end is exclusive
end = end.Add(-time.Nanosecond)
if max != end {
t.Fatalf("end time wasn't set properly.\n exp: %s\n got: %s", end, max)
}
// ensure the where clause is there
hasWhere := false
influxql.WalkFunc(s.Condition, func(n influxql.Node) {
if ex, ok := n.(*influxql.BinaryExpr); ok {
if lhs, ok := ex.LHS.(*influxql.VarRef); ok {
if lhs.Val == "foo" {
if rhs, ok := ex.RHS.(*influxql.StringLiteral); ok {
if rhs.Val == "bar" {
hasWhere = true
}
}
}
}
}
})
if !hasWhere {
t.Fatal("set time range cleared out the where clause")
}
}