Do not skip an inaccessible database at agent startup

deploy/demo-lab/.env

@@ -10,6 +10,7 @@ LOG_LEVEL=debug
 #PGSCV_COLLECT_TOP_QUERY=15
 #PGSCV_COLLECT_TOP_TABLE=15
 #PGSCV_COLLECT_TOP_INDEX=15
+#PGSCV_SKIP_CONN_ERROR_MODE=t
 #POSTGRES_DSN_db1=postgresql://pgscv:pgscv@host1:5432/pgbench1
 #POSTGRES_DSN_db2=postgresql://pgscv:pgscv@host1:5432/pgbench2
 #PGSCV_DATABASES="^pgbench.*$"

deploy/demo-lab/pgscv/pgscv.yaml

@@ -8,6 +8,7 @@ no_track_mode: false
 #collect_top_query: 10
 #collect_top_table: 10
 #collect_top_index: 10
+#skip_conn_error_mode: true
 services:
   "postgres12":
     service_type: "postgres"

internal/pgscv/config.go

@@ -35,11 +35,11 @@ type Config struct {
 	CollectorsSettings    model.CollectorsSettings `yaml:"collectors"`         // Collectors settings propagated from main YAML configuration
 	Databases             string                   `yaml:"databases"`          // Regular expression string specifies databases from which metrics should be collected
 	DatabasesRE           *regexp.Regexp           // Regular expression object compiled from Databases
-	AuthConfig            http.AuthConfig          `yaml:"authentication"`    // TLS and Basic auth configuration
-	CollectTopTable       int                      `yaml:"collect_top_table"` // Limit elements on Table collector
-	CollectTopIndex       int                      `yaml:"collect_top_index"` // Limit elements on Indexes collector
-	CollectTopQuery       int                      `yaml:"collect_top_query"` // Limit elements on Statements collector
-
+	AuthConfig            http.AuthConfig          `yaml:"authentication"`       // TLS and Basic auth configuration
+	CollectTopTable       int                      `yaml:"collect_top_table"`    // Limit elements on Table collector
+	CollectTopIndex       int                      `yaml:"collect_top_index"`    // Limit elements on Indexes collector
+	CollectTopQuery       int                      `yaml:"collect_top_query"`    // Limit elements on Statements collector
+	SkipConnErrorMode     bool                     `yaml:"skip_conn_error_mode"` // Skipping connection errors and creating a Service instance.
 }
 
 // NewConfig creates new config based on config file or return default config if config file is not specified.
@@ -109,6 +109,9 @@ func NewConfig(configFilePath string) (*Config, error) {
 		if configFromEnv.CollectTopQuery > 0 {
 			configFromFile.CollectTopQuery = configFromEnv.CollectTopQuery
 		}
+		if configFromEnv.SkipConnErrorMode {
+			configFromFile.SkipConnErrorMode = configFromEnv.SkipConnErrorMode
+		}
 		return configFromFile, nil
 	}
 
@@ -179,6 +182,10 @@ func (c *Config) Validate() error {
 		log.Infoln("no-track disabled, for details check the documentation about 'no_track_mode' option.")
 	}
 
+	if c.SkipConnErrorMode {
+		log.Infoln("skipping connection errors is enabled.")
+	}
+
 	// setup defaults
 	if c.Defaults == nil {
 		c.Defaults = map[string]string{}
@@ -387,12 +394,7 @@ func newConfigFromEnv() (*Config, error) {
 		case "PGSCV_LISTEN_ADDRESS":
 			config.ListenAddress = value
 		case "PGSCV_NO_TRACK_MODE":
-			switch value {
-			case "y", "yes", "Yes", "YES", "t", "true", "True", "TRUE", "1", "on":
-				config.NoTrackMode = true
-			default:
-				config.NoTrackMode = false
-			}
+			config.NoTrackMode = toBool(value)
 		case "PGSCV_DATABASES":
 			config.Databases = value
 		case "PGSCV_DISABLE_COLLECTORS":
@@ -423,12 +425,24 @@ func newConfigFromEnv() (*Config, error) {
 				return nil, fmt.Errorf("invalid setting PGSCV_COLLECT_TOP_INDEX, value '%s', allowed only digits", value)
 			}
 			config.CollectTopIndex = collectTopIndex
+		case "PGSCV_SKIP_CONN_ERROR_MODE":
+			config.SkipConnErrorMode = toBool(value)
 		}
 	}
-
 	return config, nil
 }
 
+func toBool(s string) bool {
+	switch s {
+	case "y", "yes", "Yes", "YES", "t", "true", "True", "TRUE", "1", "on":
+		return true
+	case "n", "no", "No", "NO", "f", "false", "False", "FALSE", "0", "off":
+		return false
+	default:
+		return false
+	}
+}
+
 // newDatabasesRegexp creates new regexp depending on passed string.
 func newDatabasesRegexp(s string) (*regexp.Regexp, error) {
 	if s == "" {

internal/pgscv/config_test.go

@@ -546,25 +546,29 @@ func Test_newConfigFromEnv(t *testing.T) {
 		{
 			valid:   true, // No env variables
 			envvars: map[string]string{},
-			want:    &Config{Defaults: map[string]string{}, ServicesConnsSettings: map[string]service.ConnSetting{}},
+			want: &Config{
+				Defaults:              map[string]string{},
+				ServicesConnsSettings: map[string]service.ConnSetting{},
+			},
 		},
 		{
 			valid: true, // Completely valid variables
 			envvars: map[string]string{
-				"PGSCV_LISTEN_ADDRESS":     "127.0.0.1:12345",
-				"PGSCV_NO_TRACK_MODE":      "yes",
-				"PGSCV_DATABASES":          "exampledb",
-				"PGSCV_DISABLE_COLLECTORS": "example/1,example/2, example/3",
-				"POSTGRES_DSN":             "example_dsn",
-				"POSTGRES_DSN_EXAMPLE1":    "example_dsn",
-				"PGBOUNCER_DSN":            "example_dsn",
-				"PGBOUNCER_DSN_EXAMPLE2":   "example_dsn",
-				"PATRONI_URL":              "example_url",
-				"PATRONI_URL_EXAMPLE3":     "example_url",
-				"PGSCV_AUTH_USERNAME":      "user",
-				"PGSCV_AUTH_PASSWORD":      "pass",
-				"PGSCV_AUTH_KEYFILE":       "keyfile.key",
-				"PGSCV_AUTH_CERTFILE":      "certfile.cert",
+				"PGSCV_LISTEN_ADDRESS":       "127.0.0.1:12345",
+				"PGSCV_NO_TRACK_MODE":        "yes",
+				"PGSCV_DATABASES":            "exampledb",
+				"PGSCV_DISABLE_COLLECTORS":   "example/1,example/2, example/3",
+				"POSTGRES_DSN":               "example_dsn",
+				"POSTGRES_DSN_EXAMPLE1":      "example_dsn",
+				"PGBOUNCER_DSN":              "example_dsn",
+				"PGBOUNCER_DSN_EXAMPLE2":     "example_dsn",
+				"PATRONI_URL":                "example_url",
+				"PATRONI_URL_EXAMPLE3":       "example_url",
+				"PGSCV_AUTH_USERNAME":        "user",
+				"PGSCV_AUTH_PASSWORD":        "pass",
+				"PGSCV_AUTH_KEYFILE":         "keyfile.key",
+				"PGSCV_AUTH_CERTFILE":        "certfile.cert",
+				"PGSCV_SKIP_CONN_ERROR_MODE": "yes",
 			},
 			want: &Config{
 				ListenAddress:     "127.0.0.1:12345",
@@ -585,7 +589,8 @@ func Test_newConfigFromEnv(t *testing.T) {
 					Keyfile:  "keyfile.key",
 					Certfile: "certfile.cert",
 				},
-				Defaults: map[string]string{},
+				Defaults:          map[string]string{},
+				SkipConnErrorMode: true,
 			},
 		},
 		{

internal/pgscv/pgscv.go

@@ -3,10 +3,11 @@ package pgscv
 import (
 	"context"
 	"errors"
+	"sync"
+
 	"github.com/cherts/pgscv/internal/http"
 	"github.com/cherts/pgscv/internal/log"
 	"github.com/cherts/pgscv/internal/service"
-	"sync"
 )
 
 // Start is the application's starting point.
@@ -25,6 +26,7 @@ func Start(ctx context.Context, config *Config) error {
 		CollectTopTable:    config.CollectTopTable,
 		CollectTopIndex:    config.CollectTopIndex,
 		CollectTopQuery:    config.CollectTopQuery,
+		SkipConnErrorMode:  config.SkipConnErrorMode,
 	}
 
 	if len(config.ServicesConnsSettings) == 0 {

internal/service/service.go

@@ -45,6 +45,7 @@ type Config struct {
 	CollectTopTable    int
 	CollectTopIndex    int
 	CollectTopQuery    int
+	SkipConnErrorMode  bool
 }
 
 // Collector is an interface for prometheus.Collector.
@@ -128,7 +129,7 @@ func (repo *Repository) addServicesFromConfig(config Config) {
 		return
 	}
 
-	// Check all passed connection settings and try to connect using them. In case of success, create a 'Service' instance
+	// Check all passed connection settings and try to connect using them. Create a 'Service' instance
 	// in the repo.
 	for k, cs := range config.ConnsSettings {
 		var msg string
@@ -154,15 +155,19 @@ func (repo *Repository) addServicesFromConfig(config Config) {
 			// Check connection using created *ConnConfig, go next if connection failed.
 			db, err := store.NewWithConfig(pgconfig)
 			if err != nil {
-				log.Warnf("%s: %s, skip", cs.Conninfo, err)
-				continue
+				if config.SkipConnErrorMode {
+					log.Warnf("%s: %s", cs.Conninfo, err)
+				} else {
+					log.Warnf("%s: %s skip", cs.Conninfo, err)
+					continue
+				}
+			} else {
+				msg = fmt.Sprintf("service [%s] available through: %s@%s:%d/%s", k, pgconfig.User, pgconfig.Host, pgconfig.Port, pgconfig.Database)
+				db.Close()
 			}
-			db.Close()
-
-			msg = fmt.Sprintf("service [%s] available through: %s@%s:%d/%s", k, pgconfig.User, pgconfig.Host, pgconfig.Port, pgconfig.Database)
 		}
 
-		// Connection was successful, create 'Service' struct with service-related properties and add it to service repo.
+		// Create 'Service' struct with service-related properties and add it to service repo.
 		s := Service{
 			ServiceID:    k,
 			ConnSettings: cs,

internal/service/service_test.go

@@ -1,10 +1,11 @@
 package service
 
 import (
+	"testing"
+
 	"github.com/cherts/pgscv/internal/model"
 	"github.com/prometheus/client_golang/prometheus"
 	"github.com/stretchr/testify/assert"
-	"testing"
 )
 
 func TestRepository_addService(t *testing.T) {
@@ -62,9 +63,22 @@ func TestRepository_addServicesFromConfig(t *testing.T) {
 	}{
 		{
 			name: "valid",
-			config: Config{ConnsSettings: ConnsSettings{
-				"test": {ServiceType: model.ServiceTypePostgresql, Conninfo: "host=127.0.0.1 port=5432 user=pgscv dbname=pgscv_fixtures"},
-			}},
+			config: Config{
+				SkipConnErrorMode: false,
+				ConnsSettings: ConnsSettings{
+					"test": {ServiceType: model.ServiceTypePostgresql,
+						Conninfo: "host=127.0.0.1 port=5432 user=pgscv dbname=pgscv_fixtures"},
+				}},
+			expected: 2,
+		},
+		{
+			name: "valid_skip_error",
+			config: Config{
+				SkipConnErrorMode: true,
+				ConnsSettings: ConnsSettings{
+					"test": {ServiceType: model.ServiceTypePostgresql,
+						Conninfo: "host=127.0.0.1 port=5430 user=pgscv dbname=pgscv_fixtures"},
+				}},
 			expected: 2,
 		},
 		{
@@ -73,13 +87,18 @@ func TestRepository_addServicesFromConfig(t *testing.T) {
 			expected: 1,
 		},
 		{
-			name:     "invalid service",
-			config:   Config{ConnsSettings: ConnsSettings{"test": {ServiceType: model.ServiceTypePostgresql, Conninfo: "invalid conninfo"}}},
+			name: "invalid service",
+			config: Config{
+				SkipConnErrorMode: true,
+				ConnsSettings:     ConnsSettings{"test": {ServiceType: model.ServiceTypePostgresql, Conninfo: "invalid conninfo"}}},
 			expected: 1,
 		},
 		{
-			name:     "unavailable service",
-			config:   Config{ConnsSettings: ConnsSettings{"test": {ServiceType: model.ServiceTypePostgresql, Conninfo: "port=1"}}},
+			name: "unavailable service",
+			config: Config{
+				SkipConnErrorMode: false,
+				ConnsSettings:     ConnsSettings{"test": {ServiceType: model.ServiceTypePostgresql, Conninfo: "port=1"}},
+			},
 			expected: 1,
 		},
 	}