diff --git a/dashboards/compliance-agent.json b/dashboards/compliance-agent.json new file mode 100644 index 0000000..344f4e8 --- /dev/null +++ b/dashboards/compliance-agent.json @@ -0,0 +1,388 @@ +{ + "title": "Compliance Agent", + "description": "Monitoring dashboard for the compliance-agent service — API handlers, scan pipeline stages, and error tracking", + "tags": ["compliance", "agent", "api"], + "layout": [ + { + "id": "row-overview", + "type": "row", + "title": "Overview", + "collapsed": false + }, + { + "id": "panel-request-rate", + "type": "graph", + "title": "API Request Rate", + "description": "Requests per second across all compliance-agent endpoints", + "gridPos": { "x": 0, "y": 1, "w": 8, "h": 8 }, + "query": { + "queryType": "builder", + "builder": { + "queryData": [ + { + "dataSource": "traces", + "queryName": "A", + "aggregateOperator": "rate", + "aggregateAttribute": { + "key": "signoz_calls_total", + "dataType": "float64", + "type": "Sum", + "isMonotonic": true + }, + "filters": { + "items": [ + { + "key": { "key": "service.name", "dataType": "string", "type": "resource" }, + "op": "=", + "value": "compliance-agent" + } + ], + "op": "AND" + }, + "groupBy": [], + "expression": "A", + "disabled": false + } + ] + }, + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT toStartOfInterval(timestamp, INTERVAL 1 MINUTE) AS ts, count() / 60 AS rps FROM signoz_traces.distributed_signoz_index_v3 WHERE serviceName = 'compliance-agent' AND timestamp >= now() - INTERVAL 1 HOUR GROUP BY ts ORDER BY ts" + } + ] + }, + "panelType": "time_series", + "unit": "ops/s" + }, + { + "id": "panel-error-rate", + "type": "graph", + "title": "Error Rate", + "description": "Rate of errors (status_code >= 400 or hasError = true) per minute", + "gridPos": { "x": 8, "y": 1, "w": 8, "h": 8 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT toStartOfInterval(timestamp, INTERVAL 1 MINUTE) AS ts, countIf(hasError = true) AS errors, count() AS total, if(total > 0, errors * 100.0 / total, 0) AS error_pct FROM signoz_traces.distributed_signoz_index_v3 WHERE serviceName = 'compliance-agent' AND timestamp >= now() - INTERVAL 1 HOUR GROUP BY ts ORDER BY ts" + } + ] + }, + "panelType": "time_series", + "unit": "%" + }, + { + "id": "panel-p99-latency", + "type": "graph", + "title": "API Latency (p50 / p95 / p99)", + "description": "Response time percentiles for compliance-agent API handlers", + "gridPos": { "x": 16, "y": 1, "w": 8, "h": 8 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT toStartOfInterval(timestamp, INTERVAL 1 MINUTE) AS ts, quantile(0.5)(durationNano) / 1e6 AS p50_ms, quantile(0.95)(durationNano) / 1e6 AS p95_ms, quantile(0.99)(durationNano) / 1e6 AS p99_ms FROM signoz_traces.distributed_signoz_index_v3 WHERE serviceName = 'compliance-agent' AND parentSpanID = '' AND timestamp >= now() - INTERVAL 1 HOUR GROUP BY ts ORDER BY ts" + } + ] + }, + "panelType": "time_series", + "unit": "ms" + }, + { + "id": "row-api-handlers", + "type": "row", + "title": "API Handlers", + "collapsed": false + }, + { + "id": "panel-handler-latency", + "type": "graph", + "title": "Handler Latency by Operation", + "description": "p95 latency per handler function name", + "gridPos": { "x": 0, "y": 10, "w": 12, "h": 10 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT toStartOfInterval(timestamp, INTERVAL 5 MINUTE) AS ts, name AS handler, quantile(0.95)(durationNano) / 1e6 AS p95_ms FROM signoz_traces.distributed_signoz_index_v3 WHERE serviceName = 'compliance-agent' AND name IN ('health', 'stats_overview', 'list_repositories', 'add_repository', 'trigger_scan', 'delete_repository', 'list_findings', 'get_finding', 'update_finding_status', 'bulk_update_finding_status', 'update_finding_feedback', 'list_sbom', 'sbom_filters', 'export_sbom', 'license_summary', 'sbom_diff', 'list_issues', 'list_scan_runs', 'get_ssh_public_key') AND timestamp >= now() - INTERVAL 3 HOUR GROUP BY ts, handler ORDER BY ts" + } + ] + }, + "panelType": "time_series", + "unit": "ms" + }, + { + "id": "panel-handler-calls", + "type": "graph", + "title": "Handler Call Count by Operation", + "description": "Number of calls per handler over time", + "gridPos": { "x": 12, "y": 10, "w": 12, "h": 10 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT toStartOfInterval(timestamp, INTERVAL 5 MINUTE) AS ts, name AS handler, count() AS calls FROM signoz_traces.distributed_signoz_index_v3 WHERE serviceName = 'compliance-agent' AND name IN ('health', 'stats_overview', 'list_repositories', 'add_repository', 'trigger_scan', 'delete_repository', 'list_findings', 'get_finding', 'update_finding_status', 'list_sbom', 'sbom_filters', 'list_issues', 'list_scan_runs', 'get_ssh_public_key') AND timestamp >= now() - INTERVAL 3 HOUR GROUP BY ts, handler ORDER BY ts" + } + ] + }, + "panelType": "time_series", + "unit": "" + }, + { + "id": "panel-handler-errors-table", + "type": "table", + "title": "Handler Errors (Last 24h)", + "description": "Handlers with the most errors in the last 24 hours", + "gridPos": { "x": 0, "y": 20, "w": 12, "h": 8 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT name AS handler, count() AS total_calls, countIf(hasError = true) AS errors, round(errors * 100.0 / total_calls, 2) AS error_pct, quantile(0.95)(durationNano) / 1e6 AS p95_ms FROM signoz_traces.distributed_signoz_index_v3 WHERE serviceName = 'compliance-agent' AND timestamp >= now() - INTERVAL 24 HOUR GROUP BY handler ORDER BY errors DESC LIMIT 20" + } + ] + }, + "panelType": "table" + }, + { + "id": "panel-slow-requests", + "type": "table", + "title": "Slowest Requests (Last 1h)", + "description": "Top 20 slowest individual requests", + "gridPos": { "x": 12, "y": 20, "w": 12, "h": 8 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT name AS handler, traceID, durationNano / 1e6 AS duration_ms, toString(timestamp) AS time, hasError FROM signoz_traces.distributed_signoz_index_v3 WHERE serviceName = 'compliance-agent' AND parentSpanID = '' AND timestamp >= now() - INTERVAL 1 HOUR ORDER BY durationNano DESC LIMIT 20" + } + ] + }, + "panelType": "table" + }, + { + "id": "row-scan-pipeline", + "type": "row", + "title": "Scan Pipeline", + "collapsed": false + }, + { + "id": "panel-scan-pipeline-duration", + "type": "graph", + "title": "Scan Pipeline Stage Duration", + "description": "Time spent in each pipeline stage (SAST, SBOM, CVE, secrets, lint, graph, etc.)", + "gridPos": { "x": 0, "y": 29, "w": 16, "h": 10 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT toStartOfInterval(timestamp, INTERVAL 5 MINUTE) AS ts, name AS stage, quantile(0.95)(durationNano) / 1e9 AS p95_seconds FROM signoz_traces.distributed_signoz_index_v3 WHERE serviceName = 'compliance-agent' AND name IN ('run_pipeline', 'stage_sast', 'stage_sbom_generation', 'stage_cve_scanning', 'stage_pattern_scanning', 'stage_secret_detection', 'stage_lint_scanning', 'stage_code_review', 'stage_graph_building') AND timestamp >= now() - INTERVAL 6 HOUR GROUP BY ts, stage ORDER BY ts" + } + ] + }, + "panelType": "time_series", + "unit": "s" + }, + { + "id": "panel-scan-count", + "type": "value", + "title": "Scans (24h)", + "description": "Total scan pipeline runs in the last 24 hours", + "gridPos": { "x": 16, "y": 29, "w": 4, "h": 5 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT count() AS scans FROM signoz_traces.distributed_signoz_index_v3 WHERE serviceName = 'compliance-agent' AND name = 'run' AND timestamp >= now() - INTERVAL 24 HOUR" + } + ] + }, + "panelType": "value" + }, + { + "id": "panel-scan-errors", + "type": "value", + "title": "Failed Scans (24h)", + "description": "Scans that had errors", + "gridPos": { "x": 20, "y": 29, "w": 4, "h": 5 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT count() AS failed FROM signoz_traces.distributed_signoz_index_v3 WHERE serviceName = 'compliance-agent' AND name = 'run' AND hasError = true AND timestamp >= now() - INTERVAL 24 HOUR" + } + ] + }, + "panelType": "value" + }, + { + "id": "panel-pipeline-runs-table", + "type": "table", + "title": "Recent Pipeline Runs", + "description": "Last 20 scan pipeline executions with duration and status", + "gridPos": { "x": 16, "y": 34, "w": 8, "h": 5 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT traceID, toString(timestamp) AS started, durationNano / 1e9 AS duration_sec, hasError, stringTagMap['repo_id'] AS repo_id, stringTagMap['trigger'] AS trigger FROM signoz_traces.distributed_signoz_index_v3 WHERE serviceName = 'compliance-agent' AND name = 'run' AND timestamp >= now() - INTERVAL 7 DAY ORDER BY timestamp DESC LIMIT 20" + } + ] + }, + "panelType": "table" + }, + { + "id": "panel-stage-breakdown", + "type": "graph", + "title": "Pipeline Stage Error Rate", + "description": "Error rate per pipeline stage", + "gridPos": { "x": 0, "y": 39, "w": 12, "h": 8 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT name AS stage, count() AS total, countIf(hasError = true) AS errors, round(errors * 100.0 / total, 2) AS error_pct FROM signoz_traces.distributed_signoz_index_v3 WHERE serviceName = 'compliance-agent' AND name LIKE 'stage_%' AND timestamp >= now() - INTERVAL 24 HOUR GROUP BY stage ORDER BY error_pct DESC" + } + ] + }, + "panelType": "bar" + }, + { + "id": "panel-stage-avg-duration", + "type": "graph", + "title": "Avg Stage Duration (Last 24h)", + "description": "Average time per pipeline stage", + "gridPos": { "x": 12, "y": 39, "w": 12, "h": 8 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT name AS stage, round(avg(durationNano) / 1e9, 2) AS avg_sec, round(quantile(0.95)(durationNano) / 1e9, 2) AS p95_sec, count() AS runs FROM signoz_traces.distributed_signoz_index_v3 WHERE serviceName = 'compliance-agent' AND name LIKE 'stage_%' AND timestamp >= now() - INTERVAL 24 HOUR GROUP BY stage ORDER BY avg_sec DESC" + } + ] + }, + "panelType": "bar" + }, + { + "id": "row-graph-dast-chat", + "type": "row", + "title": "Graph / DAST / Chat APIs", + "collapsed": true + }, + { + "id": "panel-graph-api", + "type": "graph", + "title": "Graph API Latency", + "description": "p95 latency for graph endpoints", + "gridPos": { "x": 0, "y": 48, "w": 8, "h": 8 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT toStartOfInterval(timestamp, INTERVAL 5 MINUTE) AS ts, name AS handler, quantile(0.95)(durationNano) / 1e6 AS p95_ms FROM signoz_traces.distributed_signoz_index_v3 WHERE serviceName = 'compliance-agent' AND name IN ('get_graph', 'get_nodes', 'get_communities', 'get_impact', 'search_symbols', 'get_file_content', 'trigger_build') AND timestamp >= now() - INTERVAL 3 HOUR GROUP BY ts, handler ORDER BY ts" + } + ] + }, + "panelType": "time_series", + "unit": "ms" + }, + { + "id": "panel-dast-api", + "type": "graph", + "title": "DAST API Latency", + "description": "p95 latency for DAST endpoints", + "gridPos": { "x": 8, "y": 48, "w": 8, "h": 8 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT toStartOfInterval(timestamp, INTERVAL 5 MINUTE) AS ts, name AS handler, quantile(0.95)(durationNano) / 1e6 AS p95_ms FROM signoz_traces.distributed_signoz_index_v3 WHERE serviceName = 'compliance-agent' AND name IN ('list_targets', 'add_target', 'trigger_scan', 'list_scan_runs', 'list_findings', 'get_finding') AND stringTagMap['target_id'] != '' AND timestamp >= now() - INTERVAL 3 HOUR GROUP BY ts, handler ORDER BY ts" + } + ] + }, + "panelType": "time_series", + "unit": "ms" + }, + { + "id": "panel-chat-api", + "type": "graph", + "title": "Chat / RAG API Latency", + "description": "p95 latency for chat and embedding endpoints", + "gridPos": { "x": 16, "y": 48, "w": 8, "h": 8 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT toStartOfInterval(timestamp, INTERVAL 5 MINUTE) AS ts, name AS handler, quantile(0.95)(durationNano) / 1e6 AS p95_ms FROM signoz_traces.distributed_signoz_index_v3 WHERE serviceName = 'compliance-agent' AND name IN ('chat', 'build_embeddings', 'embedding_status') AND timestamp >= now() - INTERVAL 3 HOUR GROUP BY ts, handler ORDER BY ts" + } + ] + }, + "panelType": "time_series", + "unit": "ms" + }, + { + "id": "row-logs", + "type": "row", + "title": "Logs & Warnings", + "collapsed": false + }, + { + "id": "panel-warn-errors-log", + "type": "graph", + "title": "Warn / Error Log Rate", + "description": "Rate of WARN and ERROR level logs over time", + "gridPos": { "x": 0, "y": 57, "w": 12, "h": 8 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT toStartOfInterval(timestamp, INTERVAL 5 MINUTE) AS ts, severity_text AS level, count() AS cnt FROM signoz_logs.distributed_logs_v2 WHERE resources_string['service.name'] = 'compliance-agent' AND severity_text IN ('WARN', 'ERROR') AND timestamp >= now() - INTERVAL 6 HOUR GROUP BY ts, level ORDER BY ts" + } + ] + }, + "panelType": "time_series", + "unit": "" + }, + { + "id": "panel-recent-errors", + "type": "list", + "title": "Recent Error Logs", + "description": "Last 50 ERROR-level log entries", + "gridPos": { "x": 12, "y": 57, "w": 12, "h": 8 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT toString(timestamp) AS time, body AS message, trace_id FROM signoz_logs.distributed_logs_v2 WHERE resources_string['service.name'] = 'compliance-agent' AND severity_text = 'ERROR' AND timestamp >= now() - INTERVAL 24 HOUR ORDER BY timestamp DESC LIMIT 50" + } + ] + }, + "panelType": "list" + }, + { + "id": "panel-warn-messages", + "type": "table", + "title": "Top Warning Messages (24h)", + "description": "Most frequent WARN messages — useful for spotting recurring DB or deserialization issues", + "gridPos": { "x": 0, "y": 65, "w": 24, "h": 8 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT body AS message, count() AS occurrences, max(toString(timestamp)) AS last_seen FROM signoz_logs.distributed_logs_v2 WHERE resources_string['service.name'] = 'compliance-agent' AND severity_text = 'WARN' AND timestamp >= now() - INTERVAL 24 HOUR GROUP BY message ORDER BY occurrences DESC LIMIT 30" + } + ] + }, + "panelType": "table" + } + ], + "variables": { + "time_range": { + "type": "custom", + "label": "Time Range", + "default": "1h", + "options": ["15m", "30m", "1h", "3h", "6h", "12h", "24h", "7d"] + } + } +} diff --git a/dashboards/compliance-dashboard.json b/dashboards/compliance-dashboard.json new file mode 100644 index 0000000..36b8221 --- /dev/null +++ b/dashboards/compliance-dashboard.json @@ -0,0 +1,238 @@ +{ + "title": "Compliance Dashboard (Frontend)", + "description": "Monitoring dashboard for the compliance-dashboard Dioxus SSR/WASM service — page loads, server functions, and error tracking", + "tags": ["compliance", "dashboard", "frontend"], + "layout": [ + { + "id": "row-overview", + "type": "row", + "title": "Overview", + "collapsed": false + }, + { + "id": "panel-request-rate", + "type": "graph", + "title": "Request Rate", + "description": "Requests per second to the compliance-dashboard service", + "gridPos": { "x": 0, "y": 1, "w": 8, "h": 8 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT toStartOfInterval(timestamp, INTERVAL 1 MINUTE) AS ts, count() / 60 AS rps FROM signoz_traces.distributed_signoz_index_v3 WHERE serviceName = 'compliance-dashboard' AND parentSpanID = '' AND timestamp >= now() - INTERVAL 1 HOUR GROUP BY ts ORDER BY ts" + } + ] + }, + "panelType": "time_series", + "unit": "ops/s" + }, + { + "id": "panel-error-rate", + "type": "graph", + "title": "Error Rate", + "description": "Percentage of requests with errors", + "gridPos": { "x": 8, "y": 1, "w": 8, "h": 8 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT toStartOfInterval(timestamp, INTERVAL 1 MINUTE) AS ts, countIf(hasError = true) AS errors, count() AS total, if(total > 0, errors * 100.0 / total, 0) AS error_pct FROM signoz_traces.distributed_signoz_index_v3 WHERE serviceName = 'compliance-dashboard' AND timestamp >= now() - INTERVAL 1 HOUR GROUP BY ts ORDER BY ts" + } + ] + }, + "panelType": "time_series", + "unit": "%" + }, + { + "id": "panel-latency", + "type": "graph", + "title": "Response Latency (p50 / p95 / p99)", + "description": "Response time percentiles for the dashboard service", + "gridPos": { "x": 16, "y": 1, "w": 8, "h": 8 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT toStartOfInterval(timestamp, INTERVAL 1 MINUTE) AS ts, quantile(0.5)(durationNano) / 1e6 AS p50_ms, quantile(0.95)(durationNano) / 1e6 AS p95_ms, quantile(0.99)(durationNano) / 1e6 AS p99_ms FROM signoz_traces.distributed_signoz_index_v3 WHERE serviceName = 'compliance-dashboard' AND parentSpanID = '' AND timestamp >= now() - INTERVAL 1 HOUR GROUP BY ts ORDER BY ts" + } + ] + }, + "panelType": "time_series", + "unit": "ms" + }, + { + "id": "row-server-functions", + "type": "row", + "title": "Server Functions", + "collapsed": false + }, + { + "id": "panel-server-fn-latency", + "type": "graph", + "title": "Server Function Latency by Name", + "description": "p95 latency for each Dioxus server function (fetch_repositories, fetch_findings, fetch_sbom, etc.)", + "gridPos": { "x": 0, "y": 10, "w": 12, "h": 10 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT toStartOfInterval(timestamp, INTERVAL 5 MINUTE) AS ts, name AS server_fn, quantile(0.95)(durationNano) / 1e6 AS p95_ms FROM signoz_traces.distributed_signoz_index_v3 WHERE serviceName = 'compliance-dashboard' AND timestamp >= now() - INTERVAL 3 HOUR GROUP BY ts, server_fn ORDER BY ts" + } + ] + }, + "panelType": "time_series", + "unit": "ms" + }, + { + "id": "panel-server-fn-calls", + "type": "graph", + "title": "Server Function Call Volume", + "description": "Number of calls per server function over time", + "gridPos": { "x": 12, "y": 10, "w": 12, "h": 10 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT toStartOfInterval(timestamp, INTERVAL 5 MINUTE) AS ts, name AS server_fn, count() AS calls FROM signoz_traces.distributed_signoz_index_v3 WHERE serviceName = 'compliance-dashboard' AND timestamp >= now() - INTERVAL 3 HOUR GROUP BY ts, server_fn ORDER BY ts" + } + ] + }, + "panelType": "time_series", + "unit": "" + }, + { + "id": "panel-server-fn-errors", + "type": "table", + "title": "Server Function Errors (Last 24h)", + "description": "Server functions with the most errors", + "gridPos": { "x": 0, "y": 20, "w": 12, "h": 8 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT name AS server_fn, count() AS total_calls, countIf(hasError = true) AS errors, round(errors * 100.0 / total_calls, 2) AS error_pct, quantile(0.95)(durationNano) / 1e6 AS p95_ms FROM signoz_traces.distributed_signoz_index_v3 WHERE serviceName = 'compliance-dashboard' AND timestamp >= now() - INTERVAL 24 HOUR GROUP BY server_fn HAVING total_calls > 0 ORDER BY errors DESC LIMIT 20" + } + ] + }, + "panelType": "table" + }, + { + "id": "panel-slow-requests", + "type": "table", + "title": "Slowest Requests (Last 1h)", + "description": "Top 20 slowest individual requests", + "gridPos": { "x": 12, "y": 20, "w": 12, "h": 8 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT name AS operation, traceID, durationNano / 1e6 AS duration_ms, toString(timestamp) AS time, hasError FROM signoz_traces.distributed_signoz_index_v3 WHERE serviceName = 'compliance-dashboard' AND parentSpanID = '' AND timestamp >= now() - INTERVAL 1 HOUR ORDER BY durationNano DESC LIMIT 20" + } + ] + }, + "panelType": "table" + }, + { + "id": "row-page-navigation", + "type": "row", + "title": "Page Navigation Patterns", + "collapsed": true + }, + { + "id": "panel-page-hits", + "type": "graph", + "title": "Page Load Distribution", + "description": "Which dashboard pages are hit most frequently (by server function patterns)", + "gridPos": { "x": 0, "y": 29, "w": 12, "h": 8 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT CASE WHEN name LIKE '%repositor%' THEN 'Repositories' WHEN name LIKE '%finding%' THEN 'Findings' WHEN name LIKE '%sbom%' THEN 'SBOM' WHEN name LIKE '%dast%' THEN 'DAST' WHEN name LIKE '%graph%' THEN 'Graph' WHEN name LIKE '%chat%' THEN 'Chat' WHEN name LIKE '%scan%' THEN 'Scan Runs' WHEN name LIKE '%overview%' OR name LIKE '%stats%' THEN 'Overview' ELSE 'Other' END AS page, count() AS hits FROM signoz_traces.distributed_signoz_index_v3 WHERE serviceName = 'compliance-dashboard' AND timestamp >= now() - INTERVAL 24 HOUR GROUP BY page ORDER BY hits DESC" + } + ] + }, + "panelType": "bar" + }, + { + "id": "panel-agent-connectivity", + "type": "graph", + "title": "Agent API Connectivity", + "description": "Success vs failure rate of dashboard→agent API calls", + "gridPos": { "x": 12, "y": 29, "w": 12, "h": 8 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT toStartOfInterval(timestamp, INTERVAL 5 MINUTE) AS ts, countIf(hasError = false) AS success, countIf(hasError = true) AS failed FROM signoz_traces.distributed_signoz_index_v3 WHERE serviceName = 'compliance-dashboard' AND timestamp >= now() - INTERVAL 6 HOUR GROUP BY ts ORDER BY ts" + } + ] + }, + "panelType": "time_series", + "unit": "" + }, + { + "id": "row-logs", + "type": "row", + "title": "Logs", + "collapsed": false + }, + { + "id": "panel-log-rate", + "type": "graph", + "title": "Log Volume by Severity", + "description": "Log entries per severity level over time", + "gridPos": { "x": 0, "y": 38, "w": 12, "h": 8 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT toStartOfInterval(timestamp, INTERVAL 5 MINUTE) AS ts, severity_text AS level, count() AS cnt FROM signoz_logs.distributed_logs_v2 WHERE resources_string['service.name'] = 'compliance-dashboard' AND timestamp >= now() - INTERVAL 6 HOUR GROUP BY ts, level ORDER BY ts" + } + ] + }, + "panelType": "time_series", + "unit": "" + }, + { + "id": "panel-recent-errors", + "type": "list", + "title": "Recent Error Logs", + "description": "Last 50 ERROR-level log entries from the dashboard service", + "gridPos": { "x": 12, "y": 38, "w": 12, "h": 8 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT toString(timestamp) AS time, body AS message, trace_id FROM signoz_logs.distributed_logs_v2 WHERE resources_string['service.name'] = 'compliance-dashboard' AND severity_text = 'ERROR' AND timestamp >= now() - INTERVAL 24 HOUR ORDER BY timestamp DESC LIMIT 50" + } + ] + }, + "panelType": "list" + }, + { + "id": "panel-warn-messages", + "type": "table", + "title": "Top Warning Messages (24h)", + "description": "Most frequent WARN messages — catches connectivity issues, deserialization failures, etc.", + "gridPos": { "x": 0, "y": 46, "w": 24, "h": 8 }, + "query": { + "clickhouse_sql": [ + { + "name": "A", + "query": "SELECT body AS message, count() AS occurrences, max(toString(timestamp)) AS last_seen FROM signoz_logs.distributed_logs_v2 WHERE resources_string['service.name'] = 'compliance-dashboard' AND severity_text = 'WARN' AND timestamp >= now() - INTERVAL 24 HOUR GROUP BY message ORDER BY occurrences DESC LIMIT 30" + } + ] + }, + "panelType": "table" + } + ], + "variables": { + "time_range": { + "type": "custom", + "label": "Time Range", + "default": "1h", + "options": ["15m", "30m", "1h", "3h", "6h", "12h", "24h", "7d"] + } + } +}