342 lines
18 KiB
YAML
342 lines
18 KiB
YAML
---
|
|
# yamllint disable rule:document-start rule:line-length rule:trailing-spaces rule:empty-lines
|
|
suite: Test Loki Integration - Metrics
|
|
templates:
|
|
- configmap.yaml
|
|
tests:
|
|
- it: should create the loki metrics config
|
|
set:
|
|
deployAsConfigMap: true
|
|
loki:
|
|
instances:
|
|
- name: loki
|
|
labelSelectors:
|
|
app.kubernetes.io/name: loki
|
|
asserts:
|
|
- isKind:
|
|
of: ConfigMap
|
|
- equal:
|
|
path: data["metrics.alloy"]
|
|
value: |-
|
|
declare "loki_integration" {
|
|
argument "metrics_destinations" {
|
|
comment = "Must be a list of metric destinations where collected metrics should be forwarded to"
|
|
}
|
|
|
|
declare "loki_integration_discovery" {
|
|
argument "namespaces" {
|
|
comment = "The namespaces to look for targets in (default: [] is all namespaces)"
|
|
optional = true
|
|
}
|
|
|
|
argument "field_selectors" {
|
|
comment = "The field selectors to use to find matching targets (default: [])"
|
|
optional = true
|
|
}
|
|
|
|
argument "label_selectors" {
|
|
comment = "The label selectors to use to find matching targets (default: [\"app.kubernetes.io/name=loki\"])"
|
|
optional = true
|
|
}
|
|
|
|
argument "port_name" {
|
|
comment = "The of the port to scrape metrics from (default: http-metrics)"
|
|
optional = true
|
|
}
|
|
|
|
// loki service discovery for all of the pods
|
|
discovery.kubernetes "loki_pods" {
|
|
role = "pod"
|
|
|
|
selectors {
|
|
role = "pod"
|
|
field = string.join(coalesce(argument.field_selectors.value, []), ",")
|
|
label = string.join(coalesce(argument.label_selectors.value, ["app.kubernetes.io/name=loki"]), ",")
|
|
}
|
|
|
|
namespaces {
|
|
names = coalesce(argument.namespaces.value, [])
|
|
}
|
|
}
|
|
|
|
// loki relabelings (pre-scrape)
|
|
discovery.relabel "loki_pods" {
|
|
targets = discovery.kubernetes.loki_pods.targets
|
|
|
|
// keep only the specified metrics port name, and pods that are Running and ready
|
|
rule {
|
|
source_labels = [
|
|
"__meta_kubernetes_pod_container_port_name",
|
|
"__meta_kubernetes_pod_phase",
|
|
"__meta_kubernetes_pod_ready",
|
|
"__meta_kubernetes_pod_container_init",
|
|
]
|
|
separator = "@"
|
|
regex = coalesce(argument.port_name.value, "http-metrics") + "@Running@true@false"
|
|
action = "keep"
|
|
}
|
|
|
|
// the loki-mixin expects the job label to be namespace/component
|
|
rule {
|
|
source_labels = ["__meta_kubernetes_namespace","__meta_kubernetes_pod_label_app_kubernetes_io_component"]
|
|
separator = "/"
|
|
target_label = "job"
|
|
}
|
|
|
|
|
|
|
|
rule {
|
|
source_labels = ["__meta_kubernetes_namespace"]
|
|
target_label = "namespace"
|
|
}
|
|
|
|
rule {
|
|
source_labels = ["__meta_kubernetes_pod_name"]
|
|
target_label = "pod"
|
|
}
|
|
|
|
rule {
|
|
source_labels = ["__meta_kubernetes_pod_container_name"]
|
|
target_label = "container"
|
|
}
|
|
|
|
// set the workload to the controller kind and name
|
|
rule {
|
|
action = "lowercase"
|
|
source_labels = ["__meta_kubernetes_pod_controller_kind"]
|
|
target_label = "workload_type"
|
|
}
|
|
|
|
rule {
|
|
source_labels = ["__meta_kubernetes_pod_controller_name"]
|
|
target_label = "workload"
|
|
}
|
|
|
|
// remove the hash from the ReplicaSet
|
|
rule {
|
|
source_labels = [
|
|
"workload_type",
|
|
"workload",
|
|
]
|
|
separator = "/"
|
|
regex = "replicaset/(.+)-.+$"
|
|
target_label = "workload"
|
|
}
|
|
|
|
// set the app name if specified as metadata labels "app:" or "app.kubernetes.io/name:" or "k8s-app:"
|
|
rule {
|
|
action = "replace"
|
|
source_labels = [
|
|
"__meta_kubernetes_pod_label_app_kubernetes_io_name",
|
|
"__meta_kubernetes_pod_label_k8s_app",
|
|
"__meta_kubernetes_pod_label_app",
|
|
]
|
|
separator = ";"
|
|
regex = "^(?:;*)?([^;]+).*$"
|
|
replacement = "$1"
|
|
target_label = "app"
|
|
}
|
|
|
|
// set the component if specified as metadata labels "component:" or "app.kubernetes.io/component:" or "k8s-component:"
|
|
rule {
|
|
action = "replace"
|
|
source_labels = [
|
|
"__meta_kubernetes_pod_label_app_kubernetes_io_component",
|
|
"__meta_kubernetes_pod_label_k8s_component",
|
|
"__meta_kubernetes_pod_label_component",
|
|
]
|
|
regex = "^(?:;*)?([^;]+).*$"
|
|
replacement = "$1"
|
|
target_label = "component"
|
|
}
|
|
|
|
// set a source label
|
|
rule {
|
|
action = "replace"
|
|
replacement = "kubernetes"
|
|
target_label = "source"
|
|
}
|
|
}
|
|
|
|
export "output" {
|
|
value = discovery.relabel.loki_pods.output
|
|
}
|
|
}
|
|
|
|
declare "loki_integration_scrape" {
|
|
argument "targets" {
|
|
comment = "Must be a list() of targets"
|
|
}
|
|
|
|
argument "forward_to" {
|
|
comment = "Must be a list(MetricsReceiver) where collected metrics should be forwarded to"
|
|
}
|
|
|
|
argument "job_label" {
|
|
comment = "The job label to add for all Loki metrics (default: integrations/loki)"
|
|
optional = true
|
|
}
|
|
|
|
argument "keep_metrics" {
|
|
comment = "A regular expression of metrics to keep (default: see below)"
|
|
optional = true
|
|
}
|
|
|
|
argument "drop_metrics" {
|
|
comment = "A regular expression of metrics to drop (default: see below)"
|
|
optional = true
|
|
}
|
|
|
|
argument "scrape_interval" {
|
|
comment = "How often to scrape metrics from the targets (default: 60s)"
|
|
optional = true
|
|
}
|
|
|
|
argument "max_cache_size" {
|
|
comment = "The maximum number of elements to hold in the relabeling cache (default: 100000). This should be at least 2x-5x your largest scrape target or samples appended rate."
|
|
optional = true
|
|
}
|
|
|
|
argument "clustering" {
|
|
comment = "Whether or not clustering should be enabled (default: false)"
|
|
optional = true
|
|
}
|
|
|
|
prometheus.scrape "loki" {
|
|
job_name = coalesce(argument.job_label.value, "integrations/loki")
|
|
forward_to = [prometheus.relabel.loki.receiver]
|
|
targets = argument.targets.value
|
|
scrape_interval = coalesce(argument.scrape_interval.value, "60s")
|
|
|
|
clustering {
|
|
enabled = coalesce(argument.clustering.value, false)
|
|
}
|
|
}
|
|
|
|
// loki metric relabelings (post-scrape)
|
|
prometheus.relabel "loki" {
|
|
forward_to = argument.forward_to.value
|
|
max_cache_size = coalesce(argument.max_cache_size.value, 100000)
|
|
|
|
// drop metrics that match the drop_metrics regex
|
|
rule {
|
|
source_labels = ["__name__"]
|
|
regex = coalesce(argument.drop_metrics.value, "")
|
|
action = "drop"
|
|
}
|
|
|
|
// keep only metrics that match the keep_metrics regex
|
|
rule {
|
|
source_labels = ["__name__"]
|
|
regex = coalesce(argument.keep_metrics.value, "(.+)")
|
|
action = "keep"
|
|
}
|
|
|
|
// the loki-mixin expects the instance label to be the node name
|
|
rule {
|
|
source_labels = ["node"]
|
|
target_label = "instance"
|
|
replacement = "$1"
|
|
}
|
|
rule {
|
|
action = "labeldrop"
|
|
regex = "node"
|
|
}
|
|
|
|
// set the memcached exporter container name from container="exporter" to container="memcached"
|
|
rule {
|
|
source_labels = ["component", "container"]
|
|
separator = ";"
|
|
regex = "memcached-[^;]+;exporter"
|
|
target_label = "container"
|
|
replacement = "memcached"
|
|
}
|
|
}
|
|
}
|
|
|
|
loki_integration_discovery "loki" {
|
|
namespaces = []
|
|
label_selectors = ["app.kubernetes.io/name=loki"]
|
|
port_name = "http-metrics"
|
|
}
|
|
|
|
loki_integration_scrape "loki" {
|
|
targets = loki_integration_discovery.loki.output
|
|
job_label = "integrations/loki"
|
|
clustering = true
|
|
keep_metrics = "up|scrape_samples_scraped|go_gc_cycles_total_gc_cycles_total|go_gc_duration_seconds|go_gc_duration_seconds_count|go_gc_duration_seconds_sum|go_gc_pauses_seconds_bucket|go_goroutines|go_memstats_heap_inuse_bytes|loki_azure_blob_request_duration_seconds_bucket|loki_azure_blob_request_duration_seconds_count|loki_bigtable_request_duration_seconds_bucket|loki_bigtable_request_duration_seconds_count|loki_bloom_blocks_cache_added_total|loki_bloom_blocks_cache_entries|loki_bloom_blocks_cache_evicted_total|loki_bloom_blocks_cache_fetched_total|loki_bloom_blocks_cache_usage_bytes|loki_bloom_chunks_indexed_total|loki_bloom_gateway_block_query_latency_seconds_bucket|loki_bloom_gateway_dequeue_duration_seconds_bucket|loki_bloom_gateway_filtered_chunks_sum|loki_bloom_gateway_filtered_series_sum|loki_bloom_gateway_inflight_tasks|loki_bloom_gateway_process_duration_seconds_bucket|loki_bloom_gateway_process_duration_seconds_count|loki_bloom_gateway_querier_chunks_filtered_total|loki_bloom_gateway_querier_chunks_skipped_total|loki_bloom_gateway_querier_chunks_total|loki_bloom_gateway_querier_series_filtered_total|loki_bloom_gateway_querier_series_skipped_total|loki_bloom_gateway_querier_series_total|loki_bloom_gateway_queue_duration_seconds_bucket|loki_bloom_gateway_queue_duration_seconds_count|loki_bloom_gateway_queue_duration_seconds_sum|loki_bloom_gateway_queue_length|loki_bloom_gateway_requested_chunks_sum|loki_bloom_gateway_requested_series_sum|loki_bloom_gateway_tasks_dequeued_bucket|loki_bloom_gateway_tasks_dequeued_total|loki_bloom_gateway_tasks_processed_total|loki_bloom_inserts_total|loki_bloom_recorder_chunks_total|loki_bloom_recorder_series_total|loki_bloom_size_bucket|loki_bloom_store_blocks_fetched_size_bytes_bucket|loki_bloom_store_blocks_fetched_sum|loki_bloom_store_download_queue_size_sum|loki_bloom_store_metas_fetched_bucket|loki_bloom_store_metas_fetched_size_bytes_bucket|loki_bloom_store_metas_fetched_sum|loki_bloom_tokens_total|loki_bloombuilder_blocks_created_total|loki_bloombuilder_blocks_reused_total|loki_bloombuilder_bytes_per_task_bucket|loki_bloombuilder_chunk_series_size_sum|loki_bloombuilder_metas_created_total|loki_bloombuilder_processing_task|loki_bloombuilder_series_per_task_bucket|loki_bloomplanner_blocks_deleted_total|loki_bloomplanner_connected_builders|loki_bloomplanner_inflight_tasks|loki_bloomplanner_metas_deleted_total|loki_bloomplanner_queue_length|loki_bloomplanner_retention_running|loki_bloomplanner_retention_time_seconds_bucket|loki_bloomplanner_tenant_tasks_completed|loki_bloomplanner_tenant_tasks_planned|loki_boltdb_shipper_compact_tables_operation_duration_seconds|loki_boltdb_shipper_compact_tables_operation_last_successful_run_timestamp_seconds|loki_boltdb_shipper_compact_tables_operation_total|loki_boltdb_shipper_request_duration_seconds_bucket|loki_boltdb_shipper_request_duration_seconds_count|loki_boltdb_shipper_request_duration_seconds_sum|loki_boltdb_shipper_retention_marker_count_total|loki_boltdb_shipper_retention_marker_table_processed_duration_seconds_bucket|loki_boltdb_shipper_retention_marker_table_processed_duration_seconds_count|loki_boltdb_shipper_retention_marker_table_processed_duration_seconds_sum|loki_boltdb_shipper_retention_marker_table_processed_total|loki_boltdb_shipper_retention_sweeper_chunk_deleted_duration_seconds_bucket|loki_boltdb_shipper_retention_sweeper_chunk_deleted_duration_seconds_count|loki_boltdb_shipper_retention_sweeper_chunk_deleted_duration_seconds_sum|loki_boltdb_shipper_retention_sweeper_marker_file_processing_current_time|loki_boltdb_shipper_retention_sweeper_marker_files_current|loki_build_info|loki_chunk_store_deduped_chunks_total|loki_chunk_store_index_entries_per_chunk_count|loki_chunk_store_index_entries_per_chunk_sum|loki_compactor_apply_retention_last_successful_run_timestamp_seconds|loki_compactor_apply_retention_operation_duration_seconds|loki_compactor_apply_retention_operation_total|loki_compactor_delete_requests_processed_total|loki_compactor_delete_requests_received_total|loki_compactor_deleted_lines|loki_compactor_load_pending_requests_attempts_total|loki_compactor_locked_table_successive_compaction_skips|loki_compactor_oldest_pending_delete_request_age_seconds|loki_compactor_pending_delete_requests_count|loki_consul_request_duration_seconds_bucket|loki_discarded_samples_total|loki_distributor_bytes_received_total|loki_distributor_ingester_append_failures_total|loki_distributor_lines_received_total|loki_distributor_structured_metadata_bytes_received_total|loki_dynamo_consumed_capacity_total|loki_dynamo_dropped_requests_total|loki_dynamo_failures_total|loki_dynamo_query_pages_count|loki_dynamo_request_duration_seconds_bucket|loki_dynamo_request_duration_seconds_count|loki_dynamo_throttled_total|loki_embeddedcache_entries|loki_embeddedcache_memory_bytes|loki_gcs_request_duration_seconds_bucket|loki_gcs_request_duration_seconds_count|loki_index_gateway_postfilter_chunks_sum|loki_index_gateway_prefilter_chunks_sum|loki_index_request_duration_seconds_bucket|loki_index_request_duration_seconds_count|loki_index_request_duration_seconds_sum|loki_ingester_chunk_age_seconds_bucket|loki_ingester_chunk_age_seconds_count|loki_ingester_chunk_age_seconds_sum|loki_ingester_chunk_bounds_hours_bucket|loki_ingester_chunk_bounds_hours_count|loki_ingester_chunk_bounds_hours_sum|loki_ingester_chunk_entries_bucket|loki_ingester_chunk_entries_count|loki_ingester_chunk_entries_sum|loki_ingester_chunk_size_bytes_bucket|loki_ingester_chunk_utilization_bucket|loki_ingester_chunk_utilization_count|loki_ingester_chunk_utilization_sum|loki_ingester_chunks_flushed_total|loki_ingester_flush_queue_length|loki_ingester_memory_chunks|loki_ingester_memory_streams|loki_ingester_streams_created_total|loki_memcache_request_duration_seconds_bucket|loki_memcache_request_duration_seconds_count|loki_panic_total|loki_prometheus_rule_group_rules|loki_request_duration_seconds_bucket|loki_request_duration_seconds_count|loki_request_duration_seconds_sum|loki_ruler_wal_appender_ready|loki_ruler_wal_disk_size|loki_ruler_wal_prometheus_remote_storage_highest_timestamp_in_seconds|loki_ruler_wal_prometheus_remote_storage_queue_highest_sent_timestamp_seconds|loki_ruler_wal_prometheus_remote_storage_samples_pending|loki_ruler_wal_prometheus_remote_storage_samples_total|loki_ruler_wal_samples_appended_total|loki_ruler_wal_storage_created_series_total|loki_s3_request_duration_seconds_bucket|loki_s3_request_duration_seconds_count"
|
|
scrape_interval = "60s"
|
|
max_cache_size = 100000
|
|
forward_to = argument.metrics_destinations.value
|
|
}
|
|
}
|
|
|
|
- it: should allow you to restrict the namespaces for metrics
|
|
set:
|
|
deployAsConfigMap: true
|
|
loki:
|
|
instances:
|
|
- name: loki
|
|
namespaces:
|
|
- k8smon
|
|
asserts:
|
|
- isKind:
|
|
of: ConfigMap
|
|
- matchRegex:
|
|
path: data["metrics.alloy"]
|
|
pattern: namespaces = \["k8smon"\]
|
|
|
|
- it: should allow you to set the default allow list
|
|
set:
|
|
deployAsConfigMap: true
|
|
loki:
|
|
instances:
|
|
- name: loki
|
|
metrics:
|
|
tuning:
|
|
useDefaultAllowList: true
|
|
asserts:
|
|
- isKind:
|
|
of: ConfigMap
|
|
- matchRegex:
|
|
path: data["metrics.alloy"]
|
|
pattern: keep_metrics = "up\|scrape_samples_scraped\|go_gc_cycles_total_gc_cycles_total.+\|loki_s3_request_duration_seconds_count"
|
|
|
|
- it: should allow you to specific which metrics to include
|
|
set:
|
|
deployAsConfigMap: true
|
|
loki:
|
|
instances:
|
|
- name: loki
|
|
metrics:
|
|
tuning:
|
|
useDefaultAllowList: false
|
|
includeMetrics:
|
|
- foo
|
|
- bar
|
|
asserts:
|
|
- isKind:
|
|
of: ConfigMap
|
|
- matchRegex:
|
|
path: data["metrics.alloy"]
|
|
pattern: keep_metrics = "up\|scrape_samples_scraped\|foo\|bar"
|
|
|
|
- it: should allow you to specific which metrics to exclude
|
|
set:
|
|
deployAsConfigMap: true
|
|
loki:
|
|
instances:
|
|
- name: loki
|
|
metrics:
|
|
tuning:
|
|
useDefaultAllowList: false
|
|
excludeMetrics:
|
|
- foo
|
|
- bar
|
|
asserts:
|
|
- isKind:
|
|
of: ConfigMap
|
|
- matchRegex:
|
|
path: data["metrics.alloy"]
|
|
pattern: drop_metrics = "foo\|bar"
|