| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | package k8s_apiserver |
| 4 | |
| 5 | import ( |
| 6 | "errors" |
| 7 | "os" |
| 8 | |
| 9 | "github.com/netdata/netdata/go/plugins/pkg/prometheus" |
| 10 | "github.com/netdata/netdata/go/plugins/pkg/prometheus/selector" |
| 11 | "github.com/netdata/netdata/go/plugins/pkg/web" |
| 12 | ) |
| 13 | |
| 14 | const ( |
| 15 | defaultTLSCA = "/var/run/secrets/kubernetes.io/serviceaccount/ca.crt" |
| 16 | defaultBearerTokenFile = "/var/run/secrets/kubernetes.io/serviceaccount/token" |
| 17 | ) |
| 18 | |
| 19 | func (c *Collector) validateConfig() error { |
| 20 | if c.URL == "" { |
| 21 | return errors.New("url not set") |
| 22 | } |
| 23 | |
| 24 | // Check if using default in-cluster TLS CA that doesn't exist |
| 25 | if c.TLSConfig.TLSCA == defaultTLSCA { |
| 26 | if _, err := os.Stat(c.TLSConfig.TLSCA); os.IsNotExist(err) { |
| 27 | c.Warningf("default tls_ca '%s' not found (not running inside a Kubernetes pod?); "+ |
| 28 | "set 'tls_ca' to your CA certificate path, or set it to '' if using 'tls_skip_verify: yes'", |
| 29 | defaultTLSCA) |
| 30 | } |
| 31 | } |
| 32 | |
| 33 | // Check if using default bearer token file that doesn't exist |
| 34 | if c.BearerTokenFile == defaultBearerTokenFile { |
| 35 | if _, err := os.Stat(c.BearerTokenFile); os.IsNotExist(err) { |
| 36 | c.Warningf("default bearer_token_file '%s' not found (not running inside a Kubernetes pod?); "+ |
| 37 | "set 'bearer_token_file' to your token path, or use other authentication methods", |
| 38 | defaultBearerTokenFile) |
| 39 | } |
| 40 | } |
| 41 | |
| 42 | return nil |
| 43 | } |
| 44 | |
| 45 | func (c *Collector) initPrometheusClient() (prometheus.Prometheus, error) { |
| 46 | httpClient, err := web.NewHTTPClient(c.ClientConfig) |
| 47 | if err != nil { |
| 48 | return nil, err |
| 49 | } |
| 50 | |
| 51 | // If selector parsing failed or selector is nil, log and collect all metrics |
| 52 | if srParseErr != nil || sr == nil { |
| 53 | if srParseErr != nil { |
| 54 | c.Warningf("selector parse error (collecting all metrics): %v", srParseErr) |
| 55 | } |
| 56 | return prometheus.New(httpClient, c.RequestConfig), nil |
| 57 | } |
| 58 | |
| 59 | return prometheus.NewWithSelector(httpClient, c.RequestConfig, sr), nil |
| 60 | } |
| 61 | |
| 62 | // Selector to filter only the metrics we need, reducing memory usage |
| 63 | // Parse() is called at package init time; if it fails, sr will be nil and all metrics will be collected |
| 64 | var sr, srParseErr = selector.Expr{ |
| 65 | Allow: []string{ |
| 66 | // Request metrics |
| 67 | "apiserver_request_total", |
| 68 | "apiserver_request_count", |
| 69 | "apiserver_request_duration_seconds", |
| 70 | "apiserver_response_sizes", |
| 71 | "apiserver_dropped_requests_total", |
| 72 | "apiserver_flowcontrol_rejected_requests_total", |
| 73 | |
| 74 | // Inflight metrics |
| 75 | "apiserver_current_inflight_requests", |
| 76 | "apiserver_longrunning_requests", |
| 77 | |
| 78 | // REST client metrics |
| 79 | "rest_client_requests_total", |
| 80 | "rest_client_request_duration_seconds", |
| 81 | |
| 82 | // Admission metrics |
| 83 | "apiserver_admission_step_admission_duration_seconds", |
| 84 | "apiserver_admission_controller_admission_duration_seconds", |
| 85 | "apiserver_admission_webhook_admission_duration_seconds", |
| 86 | |
| 87 | // Etcd/storage metrics |
| 88 | "apiserver_storage_objects", |
| 89 | "etcd_object_counts", |
| 90 | |
| 91 | // Workqueue metrics |
| 92 | "workqueue_depth", |
| 93 | "workqueue_adds_total", |
| 94 | "workqueue_retries_total", |
| 95 | "workqueue_queue_duration_seconds", |
| 96 | "workqueue_work_duration_seconds", |
| 97 | |
| 98 | // Audit metrics |
| 99 | "apiserver_audit_event_total", |
| 100 | "apiserver_audit_requests_rejected_total", |
| 101 | |
| 102 | // Auth metrics |
| 103 | "authenticated_user_requests", |
| 104 | "apiserver_client_certificate_expiration_seconds", |
| 105 | |
| 106 | // Process metrics |
| 107 | "go_goroutines", |
| 108 | "go_threads", |
| 109 | "go_gc_duration_seconds", |
| 110 | "go_memstats_heap_alloc_bytes", |
| 111 | "go_memstats_heap_inuse_bytes", |
| 112 | "go_memstats_stack_inuse_bytes", |
| 113 | "process_cpu_seconds_total", |
| 114 | "process_resident_memory_bytes", |
| 115 | "process_virtual_memory_bytes", |
| 116 | "process_open_fds", |
| 117 | "process_max_fds", |
| 118 | }, |
| 119 | }.Parse() |