/
downstreamer.go
244 lines (214 loc) · 6.87 KB
/
downstreamer.go
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
package queryrange
import (
"context"
"fmt"
"reflect"
"time"
"github.com/go-kit/log/level"
"github.com/grafana/dskit/concurrency"
"github.com/grafana/dskit/tenant"
"github.com/opentracing/opentracing-go"
"github.com/prometheus/prometheus/model/labels"
"github.com/prometheus/prometheus/promql"
"github.com/prometheus/prometheus/promql/parser"
"github.com/grafana/loki/v3/pkg/logql"
"github.com/grafana/loki/v3/pkg/logql/syntax"
"github.com/grafana/loki/v3/pkg/logqlmodel"
"github.com/grafana/loki/v3/pkg/querier/plan"
"github.com/grafana/loki/v3/pkg/querier/queryrange/queryrangebase"
"github.com/grafana/loki/v3/pkg/util/spanlogger"
)
const (
DefaultDownstreamConcurrency = 128
)
type DownstreamHandler struct {
limits Limits
next queryrangebase.Handler
splitAlign bool
}
func ParamsToLokiRequest(params logql.Params) queryrangebase.Request {
if logql.GetRangeType(params) == logql.InstantType {
return &LokiInstantRequest{
Query: params.QueryString(),
Limit: params.Limit(),
TimeTs: params.Start(),
Direction: params.Direction(),
Path: "/loki/api/v1/query", // TODO(owen-d): make this derivable
Shards: params.Shards(),
Plan: &plan.QueryPlan{
AST: params.GetExpression(),
},
}
}
return &LokiRequest{
Query: params.QueryString(),
Limit: params.Limit(),
Step: params.Step().Milliseconds(),
Interval: params.Interval().Milliseconds(),
StartTs: params.Start(),
EndTs: params.End(),
Direction: params.Direction(),
Path: "/loki/api/v1/query_range", // TODO(owen-d): make this derivable
Shards: params.Shards(),
Plan: &plan.QueryPlan{
AST: params.GetExpression(),
},
}
}
// Note: After the introduction of the LimitedRoundTripper,
// bounding concurrency in the downstreamer is mostly redundant
// The reason we don't remove it is to prevent malicious queries
// from creating an unreasonably large number of goroutines, such as
// the case of a query like `a / a / a / a / a ..etc`, which could try
// to shard each leg, quickly dispatching an unreasonable number of goroutines.
// In the future, it's probably better to replace this with a channel based API
// so we don't have to do all this ugly edge case handling/accounting
func (h DownstreamHandler) Downstreamer(ctx context.Context) logql.Downstreamer {
p := DefaultDownstreamConcurrency
// We may increase parallelism above the default,
// ensure we don't end up bottlenecking here.
if user, err := tenant.TenantID(ctx); err == nil {
if x := h.limits.MaxQueryParallelism(ctx, user); x > 0 {
p = x
}
}
locks := make(chan struct{}, p)
for i := 0; i < p; i++ {
locks <- struct{}{}
}
return &instance{
parallelism: p,
locks: locks,
handler: h.next,
splitAlign: h.splitAlign,
}
}
// instance is an intermediate struct for controlling concurrency across a single query
type instance struct {
parallelism int
locks chan struct{}
handler queryrangebase.Handler
splitAlign bool
}
// withoutOffset returns the given query string with offsets removed and timestamp adjusted accordingly. If no offset is present in original query, it will be returned as is.
func withoutOffset(query logql.DownstreamQuery) (string, time.Time, time.Time) {
expr := query.Params.GetExpression()
var (
newStart = query.Params.Start()
newEnd = query.Params.End()
)
expr.Walk(func(e syntax.Expr) {
switch rng := e.(type) {
case *syntax.RangeAggregationExpr:
off := rng.Left.Offset
if off != 0 {
rng.Left.Offset = 0 // remove offset
// adjust start and end time
newEnd = newEnd.Add(-off)
newStart = newStart.Add(-off)
}
}
})
return expr.String(), newStart, newEnd
}
func (in instance) Downstream(ctx context.Context, queries []logql.DownstreamQuery, acc logql.Accumulator) ([]logqlmodel.Result, error) {
return in.For(ctx, queries, acc, func(qry logql.DownstreamQuery) (logqlmodel.Result, error) {
var req queryrangebase.Request
if in.splitAlign {
qs, newStart, newEnd := withoutOffset(qry)
req = ParamsToLokiRequest(qry.Params).WithQuery(qs).WithStartEnd(newStart, newEnd)
} else {
req = ParamsToLokiRequest(qry.Params).WithQuery(qry.Params.GetExpression().String())
}
sp, ctx := opentracing.StartSpanFromContext(ctx, "DownstreamHandler.instance")
defer sp.Finish()
logger := spanlogger.FromContext(ctx)
defer logger.Finish()
level.Debug(logger).Log("shards", fmt.Sprintf("%+v", qry.Params.Shards()), "query", req.GetQuery(), "step", req.GetStep(), "handler", reflect.TypeOf(in.handler), "engine", "downstream")
res, err := in.handler.Do(ctx, req)
if err != nil {
return logqlmodel.Result{}, err
}
return ResponseToResult(res)
})
}
// For runs a function against a list of queries, collecting the results or returning an error. The indices are preserved such that input[i] maps to output[i].
func (in instance) For(
ctx context.Context,
queries []logql.DownstreamQuery,
acc logql.Accumulator,
fn func(logql.DownstreamQuery) (logqlmodel.Result, error),
) ([]logqlmodel.Result, error) {
ctx, cancel := context.WithCancel(ctx)
defer cancel()
ch := make(chan logql.Resp)
// ForEachJob blocks until all are done. However, we want to process the
// results as they come in. That's why we start everything in another
// gorouting.
go func() {
err := concurrency.ForEachJob(ctx, len(queries), in.parallelism, func(ctx context.Context, i int) error {
res, err := fn(queries[i])
response := logql.Resp{
I: i,
Res: res,
Err: err,
}
// Feed the result into the channel unless the work has completed.
select {
case <-ctx.Done():
case ch <- response:
}
return err
})
if err != nil {
ch <- logql.Resp{
I: -1,
Err: err,
}
}
close(ch)
}()
for resp := range ch {
if resp.Err != nil {
return nil, resp.Err
}
if err := acc.Accumulate(ctx, resp.Res, resp.I); err != nil {
return nil, err
}
}
return acc.Result(), nil
}
// convert to matrix
func sampleStreamToMatrix(streams []queryrangebase.SampleStream) parser.Value {
xs := make(promql.Matrix, 0, len(streams))
for _, stream := range streams {
x := promql.Series{}
x.Metric = make(labels.Labels, 0, len(stream.Labels))
for _, l := range stream.Labels {
x.Metric = append(x.Metric, labels.Label(l))
}
x.Floats = make([]promql.FPoint, 0, len(stream.Samples))
for _, sample := range stream.Samples {
x.Floats = append(x.Floats, promql.FPoint{
T: sample.TimestampMs,
F: sample.Value,
})
}
xs = append(xs, x)
}
return xs
}
func sampleStreamToVector(streams []queryrangebase.SampleStream) parser.Value {
xs := make(promql.Vector, 0, len(streams))
for _, stream := range streams {
x := promql.Sample{}
x.Metric = make(labels.Labels, 0, len(stream.Labels))
for _, l := range stream.Labels {
x.Metric = append(x.Metric, labels.Label(l))
}
x.T = stream.Samples[0].TimestampMs
x.F = stream.Samples[0].Value
xs = append(xs, x)
}
return xs
}