open-consul/command/connect/envoy/bootstrap_config.go

886 lines
30 KiB
Go
Raw Normal View History

// Copyright (c) HashiCorp, Inc.
// SPDX-License-Identifier: MPL-2.0
package envoy
import (
"bytes"
"encoding/json"
"fmt"
"net"
"net/url"
"os"
Allow HCP metrics collection for Envoy proxies Co-authored-by: Ashvitha Sridharan <ashvitha.sridharan@hashicorp.com> Co-authored-by: Freddy <freddygv@users.noreply.github.com> Add a new envoy flag: "envoy_hcp_metrics_bind_socket_dir", a directory where a unix socket will be created with the name `<namespace>_<proxy_id>.sock` to forward Envoy metrics. If set, this will configure: - In bootstrap configuration a local stats_sink and static cluster. These will forward metrics to a loopback listener sent over xDS. - A dynamic listener listening at the socket path that the previously defined static cluster is sending metrics to. - A dynamic cluster that will forward traffic received at this listener to the hcp-metrics-collector service. Reasons for having a static cluster pointing at a dynamic listener: - We want to secure the metrics stream using TLS, but the stats sink can only be defined in bootstrap config. With dynamic listeners/clusters we can use the proxy's leaf certificate issued by the Connect CA, which isn't available at bootstrap time. - We want to intelligently route to the HCP collector. Configuring its addreess at bootstrap time limits our flexibility routing-wise. More on this below. Reasons for defining the collector as an upstream in `proxycfg`: - The HCP collector will be deployed as a mesh service. - Certificate management is taken care of, as mentioned above. - Service discovery and routing logic is automatically taken care of, meaning that no code changes are required in the xds package. - Custom routing rules can be added for the collector using discovery chain config entries. Initially the collector is expected to be deployed to each admin partition, but in the future could be deployed centrally in the default partition. These config entries could even be managed by HCP itself.
2023-03-10 20:52:54 +00:00
"path"
"strings"
"text/template"
Allow HCP metrics collection for Envoy proxies Co-authored-by: Ashvitha Sridharan <ashvitha.sridharan@hashicorp.com> Co-authored-by: Freddy <freddygv@users.noreply.github.com> Add a new envoy flag: "envoy_hcp_metrics_bind_socket_dir", a directory where a unix socket will be created with the name `<namespace>_<proxy_id>.sock` to forward Envoy metrics. If set, this will configure: - In bootstrap configuration a local stats_sink and static cluster. These will forward metrics to a loopback listener sent over xDS. - A dynamic listener listening at the socket path that the previously defined static cluster is sending metrics to. - A dynamic cluster that will forward traffic received at this listener to the hcp-metrics-collector service. Reasons for having a static cluster pointing at a dynamic listener: - We want to secure the metrics stream using TLS, but the stats sink can only be defined in bootstrap config. With dynamic listeners/clusters we can use the proxy's leaf certificate issued by the Connect CA, which isn't available at bootstrap time. - We want to intelligently route to the HCP collector. Configuring its addreess at bootstrap time limits our flexibility routing-wise. More on this below. Reasons for defining the collector as an upstream in `proxycfg`: - The HCP collector will be deployed as a mesh service. - Certificate management is taken care of, as mentioned above. - Service discovery and routing logic is automatically taken care of, meaning that no code changes are required in the xds package. - Custom routing rules can be added for the collector using discovery chain config entries. Initially the collector is expected to be deployed to each admin partition, but in the future could be deployed centrally in the default partition. These config entries could even be managed by HCP itself.
2023-03-10 20:52:54 +00:00
"github.com/hashicorp/consul/acl"
"github.com/hashicorp/consul/api"
)
const (
selfAdminName = "self_admin"
)
// BootstrapConfig is the set of keys we care about in a Connect.Proxy.Config
// map. Note that this only includes config keys that affects Envoy bootstrap
// generation. For Envoy config keys that affect runtime xDS behavior see
// agent/xds/config.go.
type BootstrapConfig struct {
// StatsdURL allows simple configuration of the statsd metrics sink. If
// tagging is required, use DogstatsdURL instead. The URL must be in one of
// the following forms:
// - udp://<ip>:<port>
// - $ENV_VAR_NAME in this case the ENV var named will have it's
// value taken and is expected to contain a URL in
// one of the supported forms above.
StatsdURL string `mapstructure:"envoy_statsd_url"`
// DogstatsdURL allows simple configuration of the dogstatsd metrics sink
// which allows tags and Unix domain sockets. The URL must be in one of the
// following forms:
// - udp://<ip>:<port>
// - unix:///full/path/to/unix.sock
// - $ENV_VAR_NAME in this case the ENV var named will have it's
// value taken and is expected to contain a URL in
// one of the supported forms above.
DogstatsdURL string `mapstructure:"envoy_dogstatsd_url"`
// StatsTags is a slice of string values that will be added as tags to
// metrics. They are used to configure
// https://www.envoyproxy.io/docs/envoy/v1.9.0/api-v2/config/metrics/v2/stats.proto#envoy-api-msg-config-metrics-v2-statsconfig
// and add to the basic tags Consul adds by default like the local_cluster
// name. Only exact values are supported here. Full configuration of
// stats_config.stats_tags can be made by overriding envoy_stats_config_json.
StatsTags []string `mapstructure:"envoy_stats_tags"`
Allow HCP metrics collection for Envoy proxies Co-authored-by: Ashvitha Sridharan <ashvitha.sridharan@hashicorp.com> Co-authored-by: Freddy <freddygv@users.noreply.github.com> Add a new envoy flag: "envoy_hcp_metrics_bind_socket_dir", a directory where a unix socket will be created with the name `<namespace>_<proxy_id>.sock` to forward Envoy metrics. If set, this will configure: - In bootstrap configuration a local stats_sink and static cluster. These will forward metrics to a loopback listener sent over xDS. - A dynamic listener listening at the socket path that the previously defined static cluster is sending metrics to. - A dynamic cluster that will forward traffic received at this listener to the hcp-metrics-collector service. Reasons for having a static cluster pointing at a dynamic listener: - We want to secure the metrics stream using TLS, but the stats sink can only be defined in bootstrap config. With dynamic listeners/clusters we can use the proxy's leaf certificate issued by the Connect CA, which isn't available at bootstrap time. - We want to intelligently route to the HCP collector. Configuring its addreess at bootstrap time limits our flexibility routing-wise. More on this below. Reasons for defining the collector as an upstream in `proxycfg`: - The HCP collector will be deployed as a mesh service. - Certificate management is taken care of, as mentioned above. - Service discovery and routing logic is automatically taken care of, meaning that no code changes are required in the xds package. - Custom routing rules can be added for the collector using discovery chain config entries. Initially the collector is expected to be deployed to each admin partition, but in the future could be deployed centrally in the default partition. These config entries could even be managed by HCP itself.
2023-03-10 20:52:54 +00:00
// HCPMetricsBindSocketDir is a string that configures the directory for a
// unix socket where Envoy will forward metrics. These metrics get pushed to
// the HCP Metrics collector to show service mesh metrics on HCP.
HCPMetricsBindSocketDir string `mapstructure:"envoy_hcp_metrics_bind_socket_dir"`
// PrometheusBindAddr configures an <ip>:<port> on which the Envoy will listen
// and expose a single /metrics HTTP endpoint for Prometheus to scrape. It
// does this by proxying that URL to the internal admin server's prometheus
// endpoint which allows exposing metrics on the network without the security
// risk of exposing the full admin server API. Any other URL requested will be
// a 404.
//
// Note that as of Envoy 1.9.0, the built in Prometheus endpoint only exports
// counters and gauges but not timing information via histograms. This is
// fixed in 1.10-dev currently in Envoy master. Other changes since 1.9.0 make
// master incompatible with the current release of Consul Connect. This will
// be fixed in a future Consul version as Envoy 1.10 reaches stable release.
PrometheusBindAddr string `mapstructure:"envoy_prometheus_bind_addr"`
// StatsBindAddr configures an <ip>:<port> on which the Envoy will listen
// and expose the /stats HTTP path prefix for any agent to access. It
// does this by proxying that path prefix to the internal admin server
// which allows exposing metrics on the network without the security
// risk of exposing the full admin server API. Any other URL requested will be
// a 404.
StatsBindAddr string `mapstructure:"envoy_stats_bind_addr"`
// ReadyBindAddr configures an <ip>:<port> on which Envoy will listen and
// expose a single /ready HTTP endpoint. This is useful for checking the
// liveness of an Envoy instance when no other listeners are garaunteed to be
// configured, as is the case with ingress gateways.
//
// Note that we do not allow this to be configured via the service
// definition config map currently.
ReadyBindAddr string `mapstructure:"-"`
// OverrideJSONTpl allows replacing the base template used to render the
// bootstrap. This is an "escape hatch" allowing arbitrary control over the
// proxy's configuration but will the most effort to maintain and correctly
// configure the aspects that Connect relies upon to work. It's recommended
// that this only be used if necessary, and that it be based on the default
// template in
// https://github.com/hashicorp/consul/blob/main/command/connect/envoy/bootstrap_tpl.go
// for the correct version of Consul and Envoy being used.
OverrideJSONTpl string `mapstructure:"envoy_bootstrap_json_tpl"`
// StaticClustersJSON is a JSON string containing zero or more Cluster
// definitions. They are appended to the "static_resources.clusters" list. A
// single cluster should be given as a plain object, if more than one is to be
// added, they should be separated by a comma suitable for direct injection
// into a JSON array.
//
// Note that cluster names should be chosen in such a way that they won't
// collide with service names since we use plain service names as cluster
// names in xDS to make metrics population simpler and cluster names mush be
// unique.
//
// This is mostly intended for providing clusters for tracing or metrics
// services.
//
// See https://www.envoyproxy.io/docs/envoy/v1.9.0/api-v2/api/v2/cds.proto.
StaticClustersJSON string `mapstructure:"envoy_extra_static_clusters_json"`
// StaticListenersJSON is a JSON string containing zero or more Listener
// definitions. They are appended to the "static_resources.listeners" list. A
// single listener should be given as a plain object, if more than one is to
// be added, they should be separated by a comma suitable for direct injection
// into a JSON array.
//
// See https://www.envoyproxy.io/docs/envoy/v1.9.0/api-v2/api/v2/lds.proto.
StaticListenersJSON string `mapstructure:"envoy_extra_static_listeners_json"`
// StatsSinksJSON is a JSON string containing zero or more StatsSink
// definititions. They are appended to the `stats_sinks` array at the top
// level of the bootstrap config. A single sink should be given as a plain
// object, if more than one is to be added, they should be separated by a
// comma suitable for direct injection into a JSON array.
//
// See
// https://www.envoyproxy.io/docs/envoy/v1.9.0/api-v2/config/metrics/v2/stats.proto#config-metrics-v2-statssink.
//
// If this is non-empty then it will override anything configured in
// StatsTags.
StatsSinksJSON string `mapstructure:"envoy_extra_stats_sinks_json"`
// StatsConfigJSON is a JSON string containing an object in the right format
// to be rendered as the body of the `stats_config` field at the top level of
// the bootstrap config. It's format may vary based on Envoy version used. See
// https://www.envoyproxy.io/docs/envoy/v1.9.0/api-v2/config/metrics/v2/stats.proto#envoy-api-msg-config-metrics-v2-statsconfig.
//
// If this is non-empty then it will override anything configured in
// StatsdURL or DogstatsdURL.
StatsConfigJSON string `mapstructure:"envoy_stats_config_json"`
// StatsFlushInterval is the time duration between Envoy stats flushes. It is
// in proto3 "duration" string format for example "1.12s" See
// https://developers.google.com/protocol-buffers/docs/proto3#json and
// https://www.envoyproxy.io/docs/envoy/v1.9.0/api-v2/config/bootstrap/v2/bootstrap.proto#bootstrap
StatsFlushInterval string `mapstructure:"envoy_stats_flush_interval"`
// TracingConfigJSON is a JSON string containing an object in the right format
// to be rendered as the body of the `tracing` field at the top level of
// the bootstrap config. It's format may vary based on Envoy version used.
// See https://www.envoyproxy.io/docs/envoy/v1.9.0/api-v2/config/trace/v2/trace.proto.
TracingConfigJSON string `mapstructure:"envoy_tracing_json"`
}
// Template returns the bootstrap template to use as a base.
func (c *BootstrapConfig) Template() string {
if c.OverrideJSONTpl != "" {
return c.OverrideJSONTpl
}
return bootstrapTemplate
}
func (c *BootstrapConfig) GenerateJSON(args *BootstrapTplArgs, omitDeprecatedTags bool) ([]byte, error) {
if err := c.ConfigureArgs(args, omitDeprecatedTags); err != nil {
return nil, err
}
t, err := template.New("bootstrap").Parse(c.Template())
if err != nil {
return nil, err
}
var buf bytes.Buffer
err = t.Execute(&buf, args)
if err != nil {
return nil, err
}
// Pretty print the JSON.
var buf2 bytes.Buffer
if err := json.Indent(&buf2, buf.Bytes(), "", " "); err != nil {
return nil, err
}
return buf2.Bytes(), nil
}
// ConfigureArgs takes the basic template arguments generated from the command
// arguments and environment and modifies them according to the BootstrapConfig.
func (c *BootstrapConfig) ConfigureArgs(args *BootstrapTplArgs, omitDeprecatedTags bool) error {
// Attempt to setup sink(s) from high-level config. Note the args are passed
// by ref and modified in place.
if err := c.generateStatsSinks(args); err != nil {
return err
}
if c.StatsConfigJSON != "" {
// StatsConfig overridden explicitly
args.StatsConfigJSON = c.StatsConfigJSON
} else {
// Attempt to setup tags from high-level config. Note the args are passed by
// ref and modified in place.
stats, err := generateStatsTags(args, c.StatsTags, omitDeprecatedTags)
if err != nil {
return err
}
args.StatsConfigJSON = formatStatsTags(stats)
}
if c.StaticClustersJSON != "" {
args.StaticClustersJSON = c.StaticClustersJSON
}
if c.StaticListenersJSON != "" {
args.StaticListenersJSON = c.StaticListenersJSON
}
// Setup prometheus if needed. This MUST happen after the Static*JSON is set above
if c.PrometheusBindAddr != "" {
if err := c.generateListenerConfig(args, c.PrometheusBindAddr, "envoy_prometheus_metrics", "path", args.PrometheusScrapePath, "/stats/prometheus", args.PrometheusBackendPort); err != nil {
return err
}
}
// Setup /stats proxy listener if needed. This MUST happen after the Static*JSON is set above
if c.StatsBindAddr != "" {
if err := c.generateListenerConfig(args, c.StatsBindAddr, "envoy_metrics", "prefix", "/stats", "/stats", ""); err != nil {
return err
}
}
// Setup /ready proxy listener if needed. This MUST happen after the Static*JSON is set above
if c.ReadyBindAddr != "" {
if err := c.generateListenerConfig(args, c.ReadyBindAddr, "envoy_ready", "path", "/ready", "/ready", ""); err != nil {
return err
}
}
if c.TracingConfigJSON != "" {
args.TracingConfigJSON = c.TracingConfigJSON
}
if c.StatsFlushInterval != "" {
args.StatsFlushInterval = c.StatsFlushInterval
}
Allow HCP metrics collection for Envoy proxies Co-authored-by: Ashvitha Sridharan <ashvitha.sridharan@hashicorp.com> Co-authored-by: Freddy <freddygv@users.noreply.github.com> Add a new envoy flag: "envoy_hcp_metrics_bind_socket_dir", a directory where a unix socket will be created with the name `<namespace>_<proxy_id>.sock` to forward Envoy metrics. If set, this will configure: - In bootstrap configuration a local stats_sink and static cluster. These will forward metrics to a loopback listener sent over xDS. - A dynamic listener listening at the socket path that the previously defined static cluster is sending metrics to. - A dynamic cluster that will forward traffic received at this listener to the hcp-metrics-collector service. Reasons for having a static cluster pointing at a dynamic listener: - We want to secure the metrics stream using TLS, but the stats sink can only be defined in bootstrap config. With dynamic listeners/clusters we can use the proxy's leaf certificate issued by the Connect CA, which isn't available at bootstrap time. - We want to intelligently route to the HCP collector. Configuring its addreess at bootstrap time limits our flexibility routing-wise. More on this below. Reasons for defining the collector as an upstream in `proxycfg`: - The HCP collector will be deployed as a mesh service. - Certificate management is taken care of, as mentioned above. - Service discovery and routing logic is automatically taken care of, meaning that no code changes are required in the xds package. - Custom routing rules can be added for the collector using discovery chain config entries. Initially the collector is expected to be deployed to each admin partition, but in the future could be deployed centrally in the default partition. These config entries could even be managed by HCP itself.
2023-03-10 20:52:54 +00:00
// Setup HCP Metrics if needed. This MUST happen after the Static*JSON is set above
if c.HCPMetricsBindSocketDir != "" {
appendHCPMetricsConfig(args, c.HCPMetricsBindSocketDir)
}
return nil
}
func (c *BootstrapConfig) generateStatsSinks(args *BootstrapTplArgs) error {
var stats_sinks []string
if c.StatsdURL != "" {
sinkJSON, err := c.generateStatsSinkJSON(
"envoy.stat_sinks.statsd",
"type.googleapis.com/envoy.config.metrics.v3.StatsdSink",
c.StatsdURL,
)
if err != nil {
return err
}
stats_sinks = append(stats_sinks, sinkJSON)
}
if c.DogstatsdURL != "" {
sinkJSON, err := c.generateStatsSinkJSON(
"envoy.stat_sinks.dog_statsd",
"type.googleapis.com/envoy.config.metrics.v3.DogStatsdSink",
c.DogstatsdURL,
)
if err != nil {
return err
}
stats_sinks = append(stats_sinks, sinkJSON)
}
if c.StatsSinksJSON != "" {
stats_sinks = append(stats_sinks, c.StatsSinksJSON)
}
if len(stats_sinks) > 0 {
Allow HCP metrics collection for Envoy proxies Co-authored-by: Ashvitha Sridharan <ashvitha.sridharan@hashicorp.com> Co-authored-by: Freddy <freddygv@users.noreply.github.com> Add a new envoy flag: "envoy_hcp_metrics_bind_socket_dir", a directory where a unix socket will be created with the name `<namespace>_<proxy_id>.sock` to forward Envoy metrics. If set, this will configure: - In bootstrap configuration a local stats_sink and static cluster. These will forward metrics to a loopback listener sent over xDS. - A dynamic listener listening at the socket path that the previously defined static cluster is sending metrics to. - A dynamic cluster that will forward traffic received at this listener to the hcp-metrics-collector service. Reasons for having a static cluster pointing at a dynamic listener: - We want to secure the metrics stream using TLS, but the stats sink can only be defined in bootstrap config. With dynamic listeners/clusters we can use the proxy's leaf certificate issued by the Connect CA, which isn't available at bootstrap time. - We want to intelligently route to the HCP collector. Configuring its addreess at bootstrap time limits our flexibility routing-wise. More on this below. Reasons for defining the collector as an upstream in `proxycfg`: - The HCP collector will be deployed as a mesh service. - Certificate management is taken care of, as mentioned above. - Service discovery and routing logic is automatically taken care of, meaning that no code changes are required in the xds package. - Custom routing rules can be added for the collector using discovery chain config entries. Initially the collector is expected to be deployed to each admin partition, but in the future could be deployed centrally in the default partition. These config entries could even be managed by HCP itself.
2023-03-10 20:52:54 +00:00
args.StatsSinksJSON = strings.Join(stats_sinks, ",\n")
}
return nil
}
func (c *BootstrapConfig) generateStatsSinkJSON(name string, typeName string, addr string) (string, error) {
// Resolve address ENV var
if len(addr) > 2 && addr[0] == '$' {
addr = os.Getenv(addr[1:])
} else {
addr = os.Expand(addr, statsSinkEnvMapping)
}
u, err := url.Parse(addr)
if err != nil {
return "", fmt.Errorf("failed to parse %s sink address %q", name, addr)
}
var addrJSON string
switch u.Scheme {
case "udp":
addrJSON = `
"socket_address": {
"address": "` + u.Hostname() + `",
"port_value": ` + u.Port() + `
}
`
case "unix":
addrJSON = `
"pipe": {
"path": "` + u.Path + `"
}
`
default:
return "", fmt.Errorf("unsupported address protocol %q for %s sink",
u.Scheme, name)
}
return `{
"name": "` + name + `",
"typedConfig": {
"@type": "` + typeName + `",
"address": {
` + addrJSON + `
}
}
}`, nil
}
func statsSinkEnvMapping(s string) string {
allowedStatsSinkEnvVars := map[string]bool{
"HOST_IP": true,
}
if !allowedStatsSinkEnvVars[s] {
// if the specified env var isn't explicitly allowed, unexpand it
return fmt.Sprintf("${%s}", s)
}
return os.Getenv(s)
}
// resourceTagSpecifiers returns patterns used to generate tags from cluster and filter metric names.
func resourceTagSpecifiers(omitDeprecatedTags bool) ([]string, error) {
const (
reSegment = `[^.]+`
)
// For all rules:
// - The outer capture group is removed from the final metric name.
// - The inner capture group is extracted into labels.
rules := [][]string{
// Cluster metrics are prefixed by consul.destination
//
// Cluster metric name format:
// <subset>.<service>.<namespace>.<partition>.<datacenter|peering>.<internal|internal-<version>|external>.<trustdomain>.consul
//
// Examples:
// (default partition)
// - cluster.pong.default.dc2.internal.e5b08d03-bfc3-c870-1833-baddb116e648.consul.bind_errors: 0
// - cluster.f8f8f8f8~pong.default.dc2.internal.e5b08d03-bfc3-c870-1833-baddb116e648.consul.bind_errors: 0
// - cluster.v2.pong.default.dc2.internal.e5b08d03-bfc3-c870-1833-baddb116e648.consul.bind_errors: 0
// - cluster.f8f8f8f8~v2.pong.default.dc2.internal.e5b08d03-bfc3-c870-1833-baddb116e648.consul.bind_errors: 0
// - cluster.passthrough~pong.default.dc2.internal.e5b08d03-bfc3-c870-1833-baddb116e648.consul.bind_errors: 0
// (non-default partition)
// - cluster.pong.default.partA.dc2.internal-v1.e5b08d03-bfc3-c870-1833-baddb116e648.consul.bind_errors: 0
// - cluster.f8f8f8f8~pong.default.partA.dc2.internal-v1.e5b08d03-bfc3-c870-1833-baddb116e648.consul.bind_errors: 0
// - cluster.v2.pong.default.partA.dc2.internal-v1.e5b08d03-bfc3-c870-1833-baddb116e648.consul.bind_errors: 0
// - cluster.f8f8f8f8~v2.pong.default.partA.dc2.internal-v1.e5b08d03-bfc3-c870-1833-baddb116e648.consul.bind_errors: 0
// - cluster.passthrough~pong.default.partA.dc2.internal-v1.e5b08d03-bfc3-c870-1833-baddb116e648.consul.bind_errors: 0
// (peered)
// - cluster.pong.default.cloudpeer.external.e5b08d03-bfc3-c870-1833-baddb116e648.consul.bind_errors: 0
{"consul.destination.custom_hash",
fmt.Sprintf(`^cluster\.(?:passthrough~)?((?:(%s)~)?(?:%s\.)?%s\.%s\.(?:%s\.)?%s\.%s\.%s\.consul\.)`,
reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment)},
{"consul.destination.service_subset",
fmt.Sprintf(`^cluster\.(?:passthrough~)?((?:%s~)?(?:(%s)\.)?%s\.%s\.(?:%s\.)?%s\.%s\.%s\.consul\.)`,
reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment)},
{"consul.destination.service",
fmt.Sprintf(`^cluster\.(?:passthrough~)?((?:%s~)?(?:%s\.)?(%s)\.%s\.(?:%s\.)?%s\.%s\.%s\.consul\.)`,
reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment)},
{"consul.destination.namespace",
fmt.Sprintf(`^cluster\.(?:passthrough~)?((?:%s~)?(?:%s\.)?%s\.(%s)\.(?:%s\.)?%s\.%s\.%s\.consul\.)`,
reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment)},
{"consul.destination.partition",
fmt.Sprintf(`^cluster\.(?:passthrough~)?((?:%s~)?(?:%s\.)?%s\.%s\.(?:(%s)\.)?%s\.internal[^.]*\.%s\.consul\.)`,
reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment)},
{"consul.destination.datacenter",
fmt.Sprintf(`^cluster\.(?:passthrough~)?((?:%s~)?(?:%s\.)?%s\.%s\.(?:%s\.)?(%s)\.internal[^.]*\.%s\.consul\.)`,
reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment)},
{"consul.destination.peer",
fmt.Sprintf(`^cluster\.(%s\.(?:%s\.)?(%s)\.external\.%s\.consul\.)`,
reSegment, reSegment, reSegment, reSegment)},
{"consul.destination.routing_type",
fmt.Sprintf(`^cluster\.(?:passthrough~)?((?:%s~)?(?:%s\.)?%s\.%s\.(?:%s\.)?%s\.(%s)\.%s\.consul\.)`,
reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment)},
{"consul.destination.trust_domain",
fmt.Sprintf(`^cluster\.(?:passthrough~)?((?:%s~)?(?:%s\.)?%s\.%s\.(?:%s\.)?%s\.%s\.(%s)\.consul\.)`,
reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment)},
{"consul.destination.target",
fmt.Sprintf(`^cluster\.(?:passthrough~)?(((?:%s~)?(?:%s\.)?%s\.%s\.(?:%s\.)?%s)\.%s\.%s\.consul\.)`,
reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment)},
{"consul.destination.full_target",
fmt.Sprintf(`^cluster\.(?:passthrough~)?(((?:%s~)?(?:%s\.)?%s\.%s\.(?:%s\.)?%s\.%s\.%s)\.consul\.)`,
reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment)},
// Upstream listener metrics are prefixed by consul.upstream
//
// Listener metric name format:
// <tcp|http>.upstream.<service>.<namespace>.<partition>.<datacenter>
//
// Examples:
// - tcp.upstream.db.dc1.downstream_cx_total: 0
// - http.upstream.web.frontend.west.dc1.downstream_cx_total: 0
//
// Peered Listener metric name format:
// <tcp|http>.upstream_peered.<service>.<namespace>.peer
//
// Examples:
// - http.upstream_peered.web.frontend.cloudpeer.downstream_cx_total: 0
{"consul.upstream.service",
fmt.Sprintf(`^(?:tcp|http)\.upstream(?:_peered)?\.((%s)(?:\.%s)?(?:\.%s)?\.%s\.)`,
reSegment, reSegment, reSegment, reSegment)},
{"consul.upstream.datacenter",
fmt.Sprintf(`^(?:tcp|http)\.upstream\.(%s(?:\.%s)?(?:\.%s)?\.(%s)\.)`,
reSegment, reSegment, reSegment, reSegment)},
{"consul.upstream.peer",
fmt.Sprintf(`^(?:tcp|http)\.upstream_peered\.(%s(?:\.%s)?\.(%s)\.)`,
reSegment, reSegment, reSegment)},
{"consul.upstream.namespace",
fmt.Sprintf(`^(?:tcp|http)\.upstream(?:_peered)?\.(%s(?:\.(%s))?(?:\.%s)?\.%s\.)`,
reSegment, reSegment, reSegment, reSegment)},
{"consul.upstream.partition",
fmt.Sprintf(`^(?:tcp|http)\.upstream\.(%s(?:\.%s)?(?:\.(%s))?\.%s\.)`,
reSegment, reSegment, reSegment, reSegment)},
}
// These tags were deprecated in Consul 1.9.0
// We are leaving them enabled by default for backwards compatibility
if !omitDeprecatedTags {
deprecatedRules := [][]string{
{"consul.custom_hash",
fmt.Sprintf(`^cluster\.((?:(%s)~)?(?:%s\.)?%s\.%s\.(?:%s\.)?%s\.%s\.%s\.consul\.)`,
reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment)},
{"consul.service_subset",
fmt.Sprintf(`^cluster\.((?:%s~)?(?:(%s)\.)?%s\.%s\.(?:%s\.)?%s\.%s\.%s\.consul\.)`,
reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment)},
{"consul.service",
fmt.Sprintf(`^cluster\.((?:%s~)?(?:%s\.)?(%s)\.%s\.(?:%s\.)?%s\.%s\.%s\.consul\.)`,
reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment)},
{"consul.namespace",
fmt.Sprintf(`^cluster\.((?:%s~)?(?:%s\.)?%s\.(%s)\.(?:%s\.)?%s\.%s\.%s\.consul\.)`,
reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment)},
{"consul.datacenter",
fmt.Sprintf(`^cluster\.((?:%s~)?(?:%s\.)?%s\.%s\.(?:%s\.)?(%s)\.internal[^.]*\.%s\.consul\.)`,
reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment)},
{"consul.routing_type",
fmt.Sprintf(`^cluster\.((?:%s~)?(?:%s\.)?%s\.%s\.(?:%s\.)?%s\.(%s)\.%s\.consul\.)`,
reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment)},
{"consul.trust_domain",
fmt.Sprintf(`^cluster\.((?:%s~)?(?:%s\.)?%s\.%s\.(?:%s\.)?%s\.%s\.(%s)\.consul\.)`,
reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment)},
{"consul.target",
fmt.Sprintf(`^cluster\.(((?:%s~)?(?:%s\.)?%s\.%s\.(?:%s\.)?%s)\.%s\.%s\.consul\.)`,
reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment)},
{"consul.full_target",
fmt.Sprintf(`^cluster\.(((?:%s~)?(?:%s\.)?%s\.%s\.(?:%s\.)?%s\.%s\.%s)\.consul\.)`,
reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment, reSegment)},
}
rules = append(rules, deprecatedRules...)
}
var tags []string
for _, rule := range rules {
m := map[string]string{
"tag_name": rule[0],
"regex": rule[1],
}
d, err := json.Marshal(m)
if err != nil {
return nil, err
}
tags = append(tags, string(d))
}
return tags, nil
}
func formatStatsTags(tags []string) string {
var output string
if len(tags) > 0 {
// use_all_default_tags is true by default but we'll make it explicit!
output = `{
"stats_tags": [
` + strings.Join(tags, ",\n") + `
],
"use_all_default_tags": true
}`
}
return output
}
func generateStatsTags(args *BootstrapTplArgs, initialTags []string, omitDeprecatedTags bool) ([]string, error) {
var (
// Track tags we are setting explicitly to exclude them from defaults
tagNames = make(map[string]struct{})
tagJSONs []string
)
for _, tag := range initialTags {
parts := strings.SplitN(tag, "=", 2)
// If there is no equals, treat it as a boolean tag and just assign value of
// 1 e.g. "canary" will out put the tag "canary: 1"
v := "1"
if len(parts) == 2 {
v = parts[1]
}
k := strings.ToLower(parts[0])
tagJSON := `{
"tag_name": "` + k + `",
"fixed_value": "` + v + `"
}`
tagJSONs = append(tagJSONs, tagJSON)
tagNames[k] = struct{}{}
}
// Explode listener and cluster portions.
tags, err := resourceTagSpecifiers(omitDeprecatedTags)
if err != nil {
return nil, fmt.Errorf("failed to generate resource-specific envoy tags: %v", err)
}
tagJSONs = append(tagJSONs, tags...)
// Default the namespace and partition here since it is also done for cluster SNI
ns := args.Namespace
if ns == "" {
ns = api.IntentionDefaultNamespace
}
ap := args.Partition
if ap == "" {
ap = api.IntentionDefaultNamespace
}
// Add some default tags if not already overridden. Note this is a slice not a
// map since we need ordering to be deterministic.
defaults := []struct {
name string
val string
}{
// local_cluster is for backwards compatibility. We originally choose this
// name as it matched a few other Envoy metrics examples given in docs but
// it's a little confusing in context of setting up metrics dashboards.
{
name: "local_cluster",
val: args.ProxyCluster,
},
{
name: "consul.source.service",
val: args.ProxySourceService,
},
{
name: "consul.source.namespace",
val: ns,
},
{
name: "consul.source.partition",
val: ap,
},
{
name: "consul.source.datacenter",
val: args.Datacenter,
},
}
for _, kv := range defaults {
if kv.val == "" {
// Skip stuff we just didn't have data for.
continue
}
if _, ok := tagNames[kv.name]; ok {
// Skip anything already set explicitly.
continue
}
tagJSON := `{
"tag_name": "` + kv.name + `",
"fixed_value": "` + kv.val + `"
}`
tagJSONs = append(tagJSONs, tagJSON)
}
return tagJSONs, nil
}
func (c *BootstrapConfig) generateListenerConfig(args *BootstrapTplArgs, bindAddr, name, matchType, matchValue, prefixRewrite, prometheusBackendPort string) error {
host, port, err := net.SplitHostPort(bindAddr)
if err != nil {
return fmt.Errorf("invalid %s bind address: %s", name, err)
}
// If prometheusBackendPort is set (not empty string), create
// "prometheus_backend" cluster with the prometheusBackendPort that the
// listener will point to, rather than the "self_admin" cluster. This is for
// the merged metrics feature in consul-k8s, so the
// envoy_prometheus_bind_addr listener will point to the merged Envoy and
// service metrics endpoint rather than the Envoy admin endpoint for
// metrics. This cluster will only be created once since it's only created
// when prometheusBackendPort is set, and prometheusBackendPort is only set
// when calling this function if c.PrometheusBindAddr is set.
clusterAddress := args.AdminBindAddress
clusterPort := args.AdminBindPort
clusterName := selfAdminName
if prometheusBackendPort != "" {
clusterPort = prometheusBackendPort
clusterName = "prometheus_backend"
}
clusterJSON := `{
"name": "` + clusterName + `",
"ignore_health_on_host_removal": false,
"connect_timeout": "5s",
"type": "STATIC",
"http_protocol_options": {},
"loadAssignment": {
"clusterName": "` + clusterName + `",
"endpoints": [
{
"lbEndpoints": [
{
"endpoint": {
"address": {
"socket_address": {
"address": "` + clusterAddress + `",
"port_value": ` + clusterPort + `
}
}
}
}
]
}
]
}
}`
// Enable TLS on the prometheus listener if cert/private key are provided.
var tlsConfig string
if args.PrometheusCertFile != "" {
tlsConfig = `,
"transportSocket": {
"name": "tls",
"typedConfig": {
"@type": "type.googleapis.com/envoy.extensions.transport_sockets.tls.v3.DownstreamTlsContext",
"commonTlsContext": {
"tlsCertificateSdsSecretConfigs": [
{
"name": "prometheus_cert"
}
],
"validationContextSdsSecretConfig": {
"name": "prometheus_validation_context"
}
}
}
}`
}
listenerJSON := `{
"name": "` + name + `_listener",
"address": {
"socket_address": {
"address": "` + host + `",
"port_value": ` + port + `
}
},
"filter_chains": [
{
"filters": [
{
"name": "envoy.filters.network.http_connection_manager",
"typedConfig": {
"@type": "type.googleapis.com/envoy.extensions.filters.network.http_connection_manager.v3.HttpConnectionManager",
"stat_prefix": "` + name + `",
"codec_type": "HTTP1",
"route_config": {
"name": "self_admin_route",
"virtual_hosts": [
{
"name": "self_admin",
"domains": [
"*"
],
"routes": [
{
"match": {
"` + matchType + `": "` + matchValue + `"
},
"route": {
"cluster": "` + clusterName + `",
"prefix_rewrite": "` + prefixRewrite + `"
}
},
{
"match": {
"prefix": "/"
},
"direct_response": {
"status": 404
}
}
]
}
]
},
"http_filters": [
{
"name": "envoy.filters.http.router",
"typedConfig": {
"@type": "type.googleapis.com/envoy.extensions.filters.http.router.v3.Router"
}
}
]
}
}
]` + tlsConfig + `
}
]
}`
secretsTemplate := `{
"name": "prometheus_cert",
"tlsCertificate": {
"certificateChain": {
"filename": "%s"
},
"privateKey": {
"filename": "%s"
}
}
},
{
"name": "prometheus_validation_context",
"validationContext": {
%s
}
}`
var validationContext string
if args.PrometheusCAPath != "" {
validationContext = fmt.Sprintf(`"watchedDirectory": {
"path": "%s"
}`, args.PrometheusCAPath)
} else {
validationContext = fmt.Sprintf(`"trustedCa": {
"filename": "%s"
}`, args.PrometheusCAFile)
}
var secretsJSON string
if args.PrometheusCertFile != "" {
secretsJSON = fmt.Sprintf(secretsTemplate, args.PrometheusCertFile, args.PrometheusKeyFile, validationContext)
}
// Make sure we do not append the same cluster multiple times, as that will
// cause envoy startup to fail.
selfAdminClusterExists, err := containsSelfAdminCluster(args.StaticClustersJSON)
if err != nil {
return err
}
if args.StaticClustersJSON == "" {
args.StaticClustersJSON = clusterJSON
} else if !selfAdminClusterExists {
args.StaticClustersJSON += ",\n" + clusterJSON
}
if args.StaticListenersJSON != "" {
listenerJSON = ",\n" + listenerJSON
}
args.StaticListenersJSON += listenerJSON
if args.StaticSecretsJSON != "" {
secretsJSON = ",\n" + secretsJSON
}
args.StaticSecretsJSON += secretsJSON
return nil
}
Allow HCP metrics collection for Envoy proxies Co-authored-by: Ashvitha Sridharan <ashvitha.sridharan@hashicorp.com> Co-authored-by: Freddy <freddygv@users.noreply.github.com> Add a new envoy flag: "envoy_hcp_metrics_bind_socket_dir", a directory where a unix socket will be created with the name `<namespace>_<proxy_id>.sock` to forward Envoy metrics. If set, this will configure: - In bootstrap configuration a local stats_sink and static cluster. These will forward metrics to a loopback listener sent over xDS. - A dynamic listener listening at the socket path that the previously defined static cluster is sending metrics to. - A dynamic cluster that will forward traffic received at this listener to the hcp-metrics-collector service. Reasons for having a static cluster pointing at a dynamic listener: - We want to secure the metrics stream using TLS, but the stats sink can only be defined in bootstrap config. With dynamic listeners/clusters we can use the proxy's leaf certificate issued by the Connect CA, which isn't available at bootstrap time. - We want to intelligently route to the HCP collector. Configuring its addreess at bootstrap time limits our flexibility routing-wise. More on this below. Reasons for defining the collector as an upstream in `proxycfg`: - The HCP collector will be deployed as a mesh service. - Certificate management is taken care of, as mentioned above. - Service discovery and routing logic is automatically taken care of, meaning that no code changes are required in the xds package. - Custom routing rules can be added for the collector using discovery chain config entries. Initially the collector is expected to be deployed to each admin partition, but in the future could be deployed centrally in the default partition. These config entries could even be managed by HCP itself.
2023-03-10 20:52:54 +00:00
// appendHCPMetricsConfig generates config to enable a socket at path: <hcpMetricsBindSocketDir>/<namespace>_<proxy_id>.sock
// or <hcpMetricsBindSocketDir>/<proxy_id>.sock, if namespace is empty.
func appendHCPMetricsConfig(args *BootstrapTplArgs, hcpMetricsBindSocketDir string) {
// Normalize namespace to "default". This ensures we match the namespace behaviour in proxycfg package,
// where a dynamic listener will be created at the same socket path via xDS.
sock := fmt.Sprintf("%s_%s.sock", acl.NamespaceOrDefault(args.Namespace), args.ProxyID)
path := path.Join(hcpMetricsBindSocketDir, sock)
if args.StatsSinksJSON != "" {
args.StatsSinksJSON += ",\n"
}
args.StatsSinksJSON += `{
"name": "envoy.stat_sinks.metrics_service",
"typed_config": {
"@type": "type.googleapis.com/envoy.config.metrics.v3.MetricsServiceConfig",
"transport_api_version": "V3",
"grpc_service": {
"envoy_grpc": {
"cluster_name": "hcp_metrics_collector"
}
}
}
}`
if args.StaticClustersJSON != "" {
args.StaticClustersJSON += ",\n"
}
args.StaticClustersJSON += fmt.Sprintf(`{
"name": "hcp_metrics_collector",
"type": "STATIC",
"http2_protocol_options": {},
"loadAssignment": {
"clusterName": "hcp_metrics_collector",
"endpoints": [
{
"lbEndpoints": [
{
"endpoint": {
"address": {
"pipe": {
"path": "%s"
}
}
}
}
]
}
]
}
}`, path)
}
func containsSelfAdminCluster(clustersJSON string) (bool, error) {
clusterNames := []struct {
Name string
}{}
// StaticClustersJSON is defined as a comma-separated list of clusters, so we
// need to wrap it in JSON array brackets
err := json.Unmarshal([]byte("["+clustersJSON+"]"), &clusterNames)
if err != nil {
return false, fmt.Errorf("failed to parse static clusters: %s", err)
}
for _, cluster := range clusterNames {
if cluster.Name == selfAdminName {
return true, nil
}
}
return false, nil
}