Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

fix: False positive status reason by timing issues in alert rules #1713

Closed
wants to merge 2 commits into from
Closed

fix: False positive status reason by timing issues in alert rules #1713

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
c6315dc
introduce GatewayNoDataDelivered aler rule
hisarbalik Dec 27, 2024
7186254
Merge branch 'main' into refactor-self-monitor-alerts
hisarbalik Dec 27, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 14 additions & 0 deletions internal/selfmonitor/config/otelcol_rule_builder.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,6 +11,7 @@ const (
metricOtelCollectorExporterQueueCapacity = "otelcol_exporter_queue_capacity"
metricOtelCollectorExporterEnqueueFailed = "otelcol_exporter_enqueue_failed"
metricOtelCollectorReceiverRefused = "otelcol_receiver_refused"
metricOtelCollectorReceiverAccepted = "otelcol_receiver_accepted"
)

type otelCollectorRuleBuilder struct {
Expand All @@ -26,6 +27,7 @@ func (rb otelCollectorRuleBuilder) rules() []Rule {
rb.exporterQueueAlmostFullRule(),
rb.exporterEnqueueFailedRule(),
rb.receiverRefusedRule(),
rb.noDataDeliveredRule(),
}
}

Expand Down Expand Up @@ -90,3 +92,15 @@ func (rb otelCollectorRuleBuilder) receiverRefusedRule() Rule {
build(),
}
}

func (rb otelCollectorRuleBuilder) noDataDeliveredRule() Rule {
receivedRule := rate(rb.formatMetricName(metricOtelCollectorReceiverAccepted), selectService(rb.serviceName)).sumBy(labelPipelineName).greaterThan(0).build()

exportedRule := rate(rb.formatMetricName(metricOtelCollectorExporterSent), selectService(rb.serviceName)).sumBy(labelPipelineName).equal(0).build()

return Rule{
Alert: rb.namePrefix + RuleNameGatewayNoDataDelivered,
Expr: and(receivedRule, exportedRule),
For: alertWaitTime,
}
}
1 change: 1 addition & 0 deletions internal/selfmonitor/config/rules.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,7 @@ const (
RuleNameGatewayExporterQueueAlmostFull = "GatewayExporterQueueAlmostFull"
RuleNameGatewayExporterEnqueueFailed = "GatewayExporterEnqueueFailed"
RuleNameGatewayReceiverRefusedData = "GatewayReceiverRefusedData"
RuleNameGatewayNoDataDelivered = "GatewayNoDataDelivered"

// Fluent Bit rule names. Note that the actual full names will be prefixed with Log
RuleNameLogAgentExporterSentLogs = "AgentExporterSentLogs"
Expand Down
48 changes: 27 additions & 21 deletions internal/selfmonitor/config/rules_test.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,7 +14,7 @@ func TestMakeRules(t *testing.T) {
ruleGroup := rules.Groups[0]
require.Equal(t, "default", ruleGroup.Name)

require.Len(t, ruleGroup.Rules, 15)
require.Len(t, ruleGroup.Rules, 17)
require.Equal(t, "MetricGatewayExporterSentData", ruleGroup.Rules[0].Alert)
require.Equal(t, "sum by (pipeline_name) (rate(otelcol_exporter_sent_metric_points{service=\"telemetry-metric-gateway-metrics\"}[5m])) > 0", ruleGroup.Rules[0].Expr)

Expand All @@ -30,35 +30,41 @@ func TestMakeRules(t *testing.T) {
require.Equal(t, "MetricGatewayReceiverRefusedData", ruleGroup.Rules[4].Alert)
require.Equal(t, "sum by (receiver) (rate(otelcol_receiver_refused_metric_points{service=\"telemetry-metric-gateway-metrics\"}[5m])) > 0", ruleGroup.Rules[4].Expr)

require.Equal(t, "TraceGatewayExporterSentData", ruleGroup.Rules[5].Alert)
require.Equal(t, "sum by (pipeline_name) (rate(otelcol_exporter_sent_spans{service=\"telemetry-trace-gateway-metrics\"}[5m])) > 0", ruleGroup.Rules[5].Expr)
require.Equal(t, "MetricGatewayNoDataDelivered", ruleGroup.Rules[5].Alert)
require.Equal(t, "(sum by (pipeline_name) (rate(otelcol_receiver_accepted_metric_points{service=\"telemetry-metric-gateway-metrics\"}[5m])) > 0) and (sum by (pipeline_name) (rate(otelcol_exporter_sent_metric_points{service=\"telemetry-metric-gateway-metrics\"}[5m])) == 0)", ruleGroup.Rules[5].Expr)

require.Equal(t, "TraceGatewayExporterDroppedData", ruleGroup.Rules[6].Alert)
require.Equal(t, "sum by (pipeline_name) (rate(otelcol_exporter_send_failed_spans{service=\"telemetry-trace-gateway-metrics\"}[5m])) > 0", ruleGroup.Rules[6].Expr)
require.Equal(t, "TraceGatewayExporterSentData", ruleGroup.Rules[6].Alert)
require.Equal(t, "sum by (pipeline_name) (rate(otelcol_exporter_sent_spans{service=\"telemetry-trace-gateway-metrics\"}[5m])) > 0", ruleGroup.Rules[6].Expr)

require.Equal(t, "TraceGatewayExporterQueueAlmostFull", ruleGroup.Rules[7].Alert)
require.Equal(t, "max by (pipeline_name) (otelcol_exporter_queue_size{service=\"telemetry-trace-gateway-metrics\"} / ignoring(data_type) otelcol_exporter_queue_capacity{service=\"telemetry-trace-gateway-metrics\"}) > 0.8", ruleGroup.Rules[7].Expr)
require.Equal(t, "TraceGatewayExporterDroppedData", ruleGroup.Rules[7].Alert)
require.Equal(t, "sum by (pipeline_name) (rate(otelcol_exporter_send_failed_spans{service=\"telemetry-trace-gateway-metrics\"}[5m])) > 0", ruleGroup.Rules[7].Expr)

require.Equal(t, "TraceGatewayExporterEnqueueFailed", ruleGroup.Rules[8].Alert)
require.Equal(t, "sum by (pipeline_name) (rate(otelcol_exporter_enqueue_failed_spans{service=\"telemetry-trace-gateway-metrics\"}[5m])) > 0", ruleGroup.Rules[8].Expr)
require.Equal(t, "TraceGatewayExporterQueueAlmostFull", ruleGroup.Rules[8].Alert)
require.Equal(t, "max by (pipeline_name) (otelcol_exporter_queue_size{service=\"telemetry-trace-gateway-metrics\"} / ignoring(data_type) otelcol_exporter_queue_capacity{service=\"telemetry-trace-gateway-metrics\"}) > 0.8", ruleGroup.Rules[8].Expr)

require.Equal(t, "TraceGatewayReceiverRefusedData", ruleGroup.Rules[9].Alert)
require.Equal(t, "sum by (receiver) (rate(otelcol_receiver_refused_spans{service=\"telemetry-trace-gateway-metrics\"}[5m])) > 0", ruleGroup.Rules[9].Expr)
require.Equal(t, "TraceGatewayExporterEnqueueFailed", ruleGroup.Rules[9].Alert)
require.Equal(t, "sum by (pipeline_name) (rate(otelcol_exporter_enqueue_failed_spans{service=\"telemetry-trace-gateway-metrics\"}[5m])) > 0", ruleGroup.Rules[9].Expr)

require.Equal(t, "LogAgentExporterSentLogs", ruleGroup.Rules[10].Alert)
require.Equal(t, "sum by (pipeline_name) (rate(fluentbit_output_proc_bytes_total{service=\"telemetry-fluent-bit-metrics\"}[5m])) > 0", ruleGroup.Rules[10].Expr)
require.Equal(t, "TraceGatewayReceiverRefusedData", ruleGroup.Rules[10].Alert)
require.Equal(t, "sum by (receiver) (rate(otelcol_receiver_refused_spans{service=\"telemetry-trace-gateway-metrics\"}[5m])) > 0", ruleGroup.Rules[10].Expr)

require.Equal(t, "LogAgentExporterDroppedLogs", ruleGroup.Rules[11].Alert)
require.Equal(t, "sum by (pipeline_name) (rate(fluentbit_output_dropped_records_total{service=\"telemetry-fluent-bit-metrics\"}[5m])) > 0", ruleGroup.Rules[11].Expr)
require.Equal(t, "TraceGatewayNoDataDelivered", ruleGroup.Rules[11].Alert)
require.Equal(t, "(sum by (pipeline_name) (rate(otelcol_receiver_accepted_spans{service=\"telemetry-trace-gateway-metrics\"}[5m])) > 0) and (sum by (pipeline_name) (rate(otelcol_exporter_sent_spans{service=\"telemetry-trace-gateway-metrics\"}[5m])) == 0)", ruleGroup.Rules[11].Expr)

require.Equal(t, "LogAgentBufferInUse", ruleGroup.Rules[12].Alert)
require.Equal(t, "telemetry_fsbuffer_usage_bytes{service=\"telemetry-fluent-bit-exporter-metrics\"} > 300000000", ruleGroup.Rules[12].Expr)
require.Equal(t, "LogAgentExporterSentLogs", ruleGroup.Rules[12].Alert)
require.Equal(t, "sum by (pipeline_name) (rate(fluentbit_output_proc_bytes_total{service=\"telemetry-fluent-bit-metrics\"}[5m])) > 0", ruleGroup.Rules[12].Expr)

require.Equal(t, "LogAgentBufferFull", ruleGroup.Rules[13].Alert)
require.Equal(t, "telemetry_fsbuffer_usage_bytes{service=\"telemetry-fluent-bit-exporter-metrics\"} > 900000000", ruleGroup.Rules[13].Expr)
require.Equal(t, "LogAgentExporterDroppedLogs", ruleGroup.Rules[13].Alert)
require.Equal(t, "sum by (pipeline_name) (rate(fluentbit_output_dropped_records_total{service=\"telemetry-fluent-bit-metrics\"}[5m])) > 0", ruleGroup.Rules[13].Expr)

require.Equal(t, "LogAgentNoLogsDelivered", ruleGroup.Rules[14].Alert)
require.Equal(t, "(sum by (pipeline_name) (rate(fluentbit_input_bytes_total{service=\"telemetry-fluent-bit-metrics\"}[5m])) > 0) and (sum by (pipeline_name) (rate(fluentbit_output_proc_bytes_total{service=\"telemetry-fluent-bit-metrics\"}[5m])) == 0)", ruleGroup.Rules[14].Expr)
require.Equal(t, "LogAgentBufferInUse", ruleGroup.Rules[14].Alert)
require.Equal(t, "telemetry_fsbuffer_usage_bytes{service=\"telemetry-fluent-bit-exporter-metrics\"} > 300000000", ruleGroup.Rules[14].Expr)

require.Equal(t, "LogAgentBufferFull", ruleGroup.Rules[15].Alert)
require.Equal(t, "telemetry_fsbuffer_usage_bytes{service=\"telemetry-fluent-bit-exporter-metrics\"} > 900000000", ruleGroup.Rules[15].Expr)

require.Equal(t, "LogAgentNoLogsDelivered", ruleGroup.Rules[16].Alert)
require.Equal(t, "(sum by (pipeline_name) (rate(fluentbit_input_bytes_total{service=\"telemetry-fluent-bit-metrics\"}[5m])) > 0) and (sum by (pipeline_name) (rate(fluentbit_output_proc_bytes_total{service=\"telemetry-fluent-bit-metrics\"}[5m])) == 0)", ruleGroup.Rules[16].Expr)
}

func TestMatchesLogPipelineRule(t *testing.T) {
Expand Down
3 changes: 2 additions & 1 deletion internal/selfmonitor/prober/otel_pipeline_prober.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -88,7 +88,8 @@ func (p *OTelPipelineProber) healthy(alerts []promv1.Alert, pipelineName string)
return !(p.isFiring(alerts, config.RuleNameGatewayExporterDroppedData, pipelineName) ||
p.isFiring(alerts, config.RuleNameGatewayExporterQueueAlmostFull, pipelineName) ||
p.isFiring(alerts, config.RuleNameGatewayExporterEnqueueFailed, pipelineName) ||
p.isFiring(alerts, config.RuleNameGatewayReceiverRefusedData, pipelineName))
p.isFiring(alerts, config.RuleNameGatewayReceiverRefusedData, pipelineName) ||
p.isFiring(alerts, config.RuleNameGatewayNoDataDelivered, pipelineName))
}

func (p *OTelPipelineProber) isFiring(alerts []promv1.Alert, ruleName, pipelineName string) bool {
Expand Down
Loading