Skip to content

Commit

Permalink
Tail-based sampling (#526)
Browse files Browse the repository at this point in the history
* Add tail sampling in tempo pipelines

* Add load balancing for traces when tail sampling

Implements load balancing of spans by trace ID between agent instances for tail sampling

* Lint things

* Config recevier listening port

* Add sampling in scraping service compose example

* Don't load balance when it's not needed

Many times users will have single instance deployements that do not require
to load balance spans.

Load balancing block is optional

* Config reference for tail sampling

* Update CHANGELOG

* Fix panic

* Move exporter to its own block

* Make tail-sampling work in k3d example

* Fix image name

* Lower collector log level

* Remove unnecessary pvcs
  • Loading branch information
mapno authored Apr 16, 2021
1 parent 6b70747 commit 697685c
Show file tree
Hide file tree
Showing 49 changed files with 8,661 additions and 10 deletions.
2 changes: 2 additions & 0 deletions CHANGELOG.md
Original file line number Diff line number Diff line change
Expand Up @@ -15,6 +15,8 @@ cross-compilation issue, but will return in v0.13.0.
- [BUGFIX] Validate that incoming scraped metrics do not have an empty label
set or a label set with duplicate labels, mirroring the behavior of
Prometheus. (@rfratto)

- [FEATURE] Tail-based sampling for tracing pipelines (@mapno)

# v0.13.1 (2021-04-09)

Expand Down
53 changes: 53 additions & 0 deletions docs/configuration-reference.md
Original file line number Diff line number Diff line change
Expand Up @@ -2078,6 +2078,59 @@ spanmetrics:
# They can be further namespaced, i.e. `{namespace}_tempo_spanmetrics`
[ namespace: <prometheusexporter.namespace> ]
[ send_timestamps: <prometheusexporter.send_timestamps> ]

# tail_sampling supports tail-based sampling of traces in the agent.
# Policies can be defined that determine what traces are sampled and sent to the backends and what traces are dropped.
# In order to make a correct sampling decision it's important that the agent has a complete trace.
# This is achieved by waiting a given time for all the spans before evaluating the trace.
# Tail sampling also supports multiple agent deployments, allowing to group all spans of a trace
# in the same agent by load balancing the spans by trace ID between the instances.
tail_sampling:
# policies define the rules by which traces will be sampled. Multiple policies can be added to the same pipeline
# They are the same as the policies in Open-Telemetry's tailsamplingprocessor.
# https://github.com/open-telemetry/opentelemetry-collector-contrib/tree/main/processor/tailsamplingprocessor
policies:
- [<tailsamplingprocessor.policies>]
# decision_wait is the time that will be waited before making a decision for a trace.
# Longer times reduce the probability of sampling an incomplete trace at the cost of higher memory usage.
decision_wait: [ <string> | default="5s" ]
# load_balancing configures load balancing of spans across multiple agents.
# It ensures that all spans of a trace are sampled in the same instance.
# It's not necessary when only one agent is receiving traces (e.g. single instance deployments).
load_balancing:
# resolver configures the resolution strategy for the involved backends
# It can be static, with a fixed list of hostnames, or DNS, with a hostname (and port) that will resolve to all IP addresses.
# It's the same as the config in loadbalancingexporter.
# https://github.com/open-telemetry/opentelemetry-collector-contrib/tree/main/exporter/loadbalancingexporter
resolver:
static:
hostnames:
[ - <string> ... ]
dns:
hostname: <string>
[ port: <int> ]

# Load balancing is done via an otlp exporter.
# The remaining configuration is common with the remote_write block.
exporter:
# Controls whether compression is enabled.
[ compression: <string> | default = "gzip" | supported = "none", "gzip"]

# Controls whether or not TLS is required. See https://godoc.org/google.golang.org/grpc#WithInsecure
[ insecure: <boolean> | default = false ]

# Disable validation of the server certificate. Only used when insecure is set
# to false.
[ insecure_skip_verify: <bool> | default = false ]

# Sets the `Authorization` header on every trace push with the
# configured username and password.
# password and password_file are mutually exclusive.
basic_auth:
[ username: <string> ]
[ password: <secret> ]
[ password_file: <string> ]

```

### integrations_config
Expand Down
42 changes: 42 additions & 0 deletions example/docker-compose/agent/config/agent-scraping-service.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -21,3 +21,45 @@ prometheus:
endpoints:
- etcd:2379

tempo:
configs:
- name: default
receivers:
jaeger:
protocols:
thrift_http:
attributes:
actions:
- action: upsert
key: env
value: prod
remote_write:
- endpoint: otel-collector:55680
insecure: true
batch:
timeout: 5s
send_batch_size: 100
tail_sampling:
policies:
- always_sample:
# Enter more policies to sample traces (ref: https://github.com/open-telemetry/opentelemetry-collector-contrib/tree/main/processor/tailsamplingprocessor)
# Example of rate limiting policy. Uncomment to start rate limiting!
# - name: policy-2
# type: rate_limiting
# rate_limiting:
# - spans_per_second: 35
load_balancing:
exporter:
insecure: true
resolver:
static:
hostnames:
- agent-1:4318
- agent-2:4318
- agent-3:4318
spanmetrics:
dimensions:
- name: http.url
metrics_exporter:
endpoint: 0.0.0.0:8889

19 changes: 19 additions & 0 deletions example/docker-compose/docker-compose.scraping-service.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -80,3 +80,22 @@ services:
image: quay.io/freshtracks.io/avalanche:latest
avalanche-5:
image: quay.io/freshtracks.io/avalanche:latest

# tracing load generator
synthetic-load-generator:
image: omnition/synthetic-load-generator:1.0.25
volumes:
- ./load-generator:/etc/load-generator
environment:
- TOPOLOGY_FILE=/etc/load-generator/load-generator.json
- JAEGER_COLLECTOR_URL=http://agent-1:14268
depends_on:
- agent-1

# tracing backend
otel-collector:
image: otel/opentelemetry-collector:0.9.0
volumes:
- ./otel-collector:/etc/otel-collector
command:
- --config=/etc/otel-collector/config.yaml
48 changes: 47 additions & 1 deletion example/k3d/environment/main.jsonnet
Original file line number Diff line number Diff line change
@@ -1,11 +1,14 @@
local collector = import 'collector/main.libsonnet';
local default = import 'default/main.libsonnet';
local etcd = import 'etcd/main.libsonnet';
local agent_cluster = import 'grafana-agent/scraping-svc/main.libsonnet';
local k = import 'ksonnet-util/kausal.libsonnet';
local load_generator = import 'load-generator/main.libsonnet';

local loki_config = import 'default/loki_config.libsonnet';
local grafana_agent = import 'grafana-agent/v1/main.libsonnet';

local containerPort = k.core.v1.containerPort;
local ingress = k.networking.v1beta1.ingress;
local path = k.networking.v1beta1.httpIngressPath;
local rule = k.networking.v1beta1.ingressRule;
Expand Down Expand Up @@ -65,11 +68,54 @@ local images = {
scheme: 'http',
hostname: 'loki.default.svc.cluster.local',
external_labels: { cluster: cluster_label },
})),
})) +
grafana_agent.withTempoConfig({
receivers: {
jaeger: {
protocols: {
thrift_http: null,
},
},
},
batch: {
timeout: '5s',
send_batch_size: 1000,
},
}) +
grafana_agent.withPortsMixin([
containerPort.new('thrift-http', 14268) + containerPort.withProtocol('TCP'),
containerPort.new('otlp-lb', 4318) + containerPort.withProtocol('TCP'),
]) +
grafana_agent.withTempoRemoteWrite([
{
endpoint: 'collector.default.svc.cluster.local:55680',
insecure: true,
},
]) +
grafana_agent.withTempoTailSamplingConfig({
policies: [{
always_sample: null,
}],
load_balancing: {
exporter: {
insecure: true,
},
resolver: {
dns: {
hostname: 'grafana-agent.default.svc.cluster.local',
port: 4318,
},
},
},
}),

// Need to run ETCD for agent_cluster
etcd: etcd.new('default'),

collector: collector.new('default'),

load_generator: load_generator.new('default'),

agent_cluster:
agent_cluster.new('default', 'kube-system') +
agent_cluster.withImagesMixin(images) +
Expand Down
28 changes: 28 additions & 0 deletions example/k3d/lib/collector/collector-config.libsonnet
Original file line number Diff line number Diff line change
@@ -0,0 +1,28 @@
{
receivers: {
otlp: {
protocols: {
grpc: null,
},
},
},

exporters: {
logging: {
loglevel: "info",
},
},

service: {
pipelines: {
traces: {
receivers: [
"otlp",
],
exporters: [
"logging",
],
},
},
},
}
47 changes: 47 additions & 0 deletions example/k3d/lib/collector/main.libsonnet
Original file line number Diff line number Diff line change
@@ -0,0 +1,47 @@
local k = import 'ksonnet-util/kausal.libsonnet';

local configMap = k.core.v1.configMap;
local container = k.core.v1.container;
local containerPort = k.core.v1.containerPort;
local deployment = k.apps.v1.deployment;
local pvc = k.core.v1.persistentVolumeClaim;
local service = k.core.v1.service;
local volumeMount = k.core.v1.volumeMount;
local volume = k.core.v1.volume;

{
new(namespace=''):: {
local this = self,

_images:: {
collector: 'otel/opentelemetry-collector:0.9.0',
},
_config:: (import './collector-config.libsonnet'),

configMap:
configMap.new('collector') +
configMap.mixin.metadata.withNamespace(namespace) +
configMap.withData({
'config.yaml': k.util.manifestYaml(this._config),
}),

container::
container.new('collector', this._images.collector) +
container.withPorts([
containerPort.newNamed(name='grpc', containerPort=55680),
]) +
container.withArgsMixin(
'--config=/etc/collector/config.yaml',
),

deployment:
deployment.new('collector', 1, [self.container]) +
deployment.mixin.metadata.withNamespace(namespace) +
k.util.configMapVolumeMount(this.configMap, '/etc/collector'),


service:
k.util.serviceFor(self.deployment) +
service.mixin.metadata.withNamespace(namespace),
},
}
Loading

0 comments on commit 697685c

Please sign in to comment.