holmesgpt 0.13.2__py3-none-any.whl → 0.18.4__py3-none-any.whl

holmes/plugins/toolsets/grafana/toolset_grafana_dashboard.jinja2

@@ -0,0 +1,49 @@
+---
+  ## **IMPORTANT: Handling Grafana Dashboard URLs**
+
+  **When the user provides a Grafana dashboard URL (e.g., http://some-domain.com/d/09ec8aa1e996d6ffcd6817bbaff4db1b/kubernetes-api-server):**
+
+  1. **DO NOT use the internet/fetch_webpage tool** - Grafana URLs should always be handled by the grafana dashboards toolset
+  2. **Extract the dashboard parameters from the URL:**
+     - Pattern: `/d/{uid}/{dashboard-name}`
+     - Example: `/d/09ec8aa1e996d6ffcd6817bbaff4db1b/kubernetes-api-server` → UID is `09ec8aa1e996d6ffcd6817bbaff4db1b`
+  3. **Use `grafana_get_dashboard_by_uid` with the extracted UID** to fetch the dashboard definition
+  4. **Extract queries from the dashboard panels** and execute them with the appropriate toolset (Prometheus, Loki, etc.)
+
+  ### **Example Workflow:**
+  User: "Look at this graph: http://localhost:3000/d/abc123/my-dashboard?from=now-1h&to=now"
+  ↓
+  1. Recognize this is a Grafana URL (contains /d/)
+  2. Extract UID: abc123
+  3. Call grafana_get_dashboard_by_uid(uid="abc123")
+  4. Analyze dashboard panels and their queries
+  5. Execute relevant queries using the time range from the URL (from=now-1h, to=now)
+
+  **Key Point:** Always prefer the Grafana toolset for any URL pointing to a Grafana instance. This gives you access to the dashboard structure, panel queries, and metadata - not just HTML content.
+
+
+## **Instruction: Prometheus Dashboard Query Execution**
+
+**When running Prometheus queries from Grafana dashboards:**
+
+### **Key Rules:**
+1. **Always run as range queries** (not instant queries) to match dashboard behavior
+2. **Use dashboard's time range** from JSON: `"time": {"from": "now-1h", "to": "now"}`
+3. **Substitute variables:**
+   - `$__rate_interval` → `1m` (1-hour range) or `5m` (3+ hour range)
+   - `$cluster` → omit if empty
+4. **Copy query exactly** including `ceil()`, `sort_desc()`, etc.
+
+### **Example:**
+```
+Dashboard: "now-1h" + query with [$__rate_interval]
+↓
+Execute: start="-3600" + query with [1m]
+```
+
+### **Ignore:**
+- `"refresh": "10s"` (UI setting, not query-related)
+
+**This ensures you see the same time series data and historical spikes that users see on their dashboards.**
+
+---

holmes/plugins/toolsets/grafana/toolset_grafana_tempo.jinja2

@@ -1,12 +1,247 @@
-Use Tempo when investigating latency or performance issues. Tempo provides traces information for application running on the cluster.
+Grafana Tempo provides distributed tracing data through its REST API. Each tool maps directly to a specific Tempo API endpoint.
+
 Assume every application provides tempo traces.
-1. Start by identifying an initial filter to use. This can be a pod name, a deployment name or a service name
-2. Call fetch_tempo_traces_comparative_sample first when investigating performance issues via traces. This tool provides comprehensive analysis for identifying patterns. For other issues not related to performance, you can start with fetch_tempo_traces.
-3. Use `fetch_tempo_traces` setting the appropriate query params
-    - Use the min_duration filter to ensure you get traces that trigger the alert when you are investigating a performance issue
-    - If possible, use start and end date to narrow down your search.
-        - Use fetch_finding_by_id if you are provided with a finding/alert id. It will contain details about when the alert was triggered
-    - Use at least one of the following argument to ensure you get relevant traces: `service_name`, `pod_name` or `deployment_name`.
-4. When you have a specific trace ID to investigate, use `fetch_tempo_trace_by_id` to get detailed information about that trace.
-5. Look at the duration of each span in any single trace and deduce any issues.
-6. ALWAYS fetch the logs for a pod once you identify a span that is taking a long time. There may be an explanation for the slowness in the logs.
+
+## API Endpoints and Tool Mapping
+
+1. **Trace Search** (GET /api/search)
+   - `tempo_search_traces_by_query`: Use with 'q' parameter for TraceQL queries
+   - `tempo_search_traces_by_tags`: Use with 'tags' parameter for logfmt queries
+
+2. **Trace Details** (GET /api/v2/traces/{trace_id})
+   - `tempo_query_trace_by_id`: Retrieve full trace data
+
+3. **Tag Discovery**
+   - `tempo_search_tag_names` (GET /api/v2/search/tags): List available tags
+   - `tempo_search_tag_values` (GET /api/v2/search/tag/{tag}/values): Get values for a tag
+
+4. **TraceQL Metrics**
+   - `tempo_query_metrics_instant` (GET /api/metrics/query): Single value computation
+   - `tempo_query_metrics_range` (GET /api/metrics/query_range): Time series data
+
+## Usage Workflow
+
+### 1. Discovering Available Data
+Start by understanding what tags and values exist:
+- Use `tempo_search_tag_names` to discover available tags
+- Use `tempo_search_tag_values` to see all values for a specific tag (e.g., service names)
+
+### 2. Searching for Traces
+
+**TraceQL Search (recommended):**
+Use `tempo_search_traces_by_query` with TraceQL syntax for powerful filtering.
+
+**TraceQL Capabilities:**
+TraceQL can select traces based on the following:
+- **Span and resource attributes** - Filter by any attribute on spans or resources
+- **Timing and duration** - Filter by trace/span duration
+- **Basic aggregates** - Use aggregate functions to compute values across spans
+
+**Supported Aggregate Functions:**
+- `count()` - Count the number of spans matching the criteria
+- `avg(attribute)` - Calculate average of a numeric attribute across spans
+- `min(attribute)` - Find minimum value of a numeric attribute
+- `max(attribute)` - Find maximum value of a numeric attribute
+- `sum(attribute)` - Sum values of a numeric attribute across spans
+
+**Aggregate Function Usage:**
+Aggregates are used with the pipe operator `|` to filter traces based on computed values across their spans.
+
+**Aggregate Examples:**
+- `{ span.http.status_code = 200 } | count() > 3` - Find traces with more than 3 spans having HTTP 200 status
+- `{ } | sum(span.bytesProcessed) > 1000000000` - Find traces where total processed bytes exceed 1 GB
+- `{ status = error } | by(resource.service.name) | count() > 1` - Find services with more than 1 error
+
+**Select Function:**
+- `{ status = error } | select(span.http.status_code, span.http.url)` - Select specific attributes from error spans
+
+**TraceQL Query Structure:**
+TraceQL queries follow the pattern: `{span-selectors} | aggregate`
+
+**TraceQL Query Examples (from official docs):**
+
+1. **Find traces of a specific operation:**
+   ```
+   {resource.service.name = "frontend" && name = "POST /api/orders"}
+   ```
+   ```
+   {
+     resource.service.namespace = "ecommerce" &&
+     resource.service.name = "frontend" &&
+     resource.deployment.environment = "production" &&
+     name = "POST /api/orders"
+   }
+   ```
+
+2. **Find traces with a particular outcome:**
+   ```
+   {
+     resource.service.name="frontend" &&
+     name = "POST /api/orders" &&
+     status = error
+   }
+   ```
+   ```
+   {
+     resource.service.name="frontend" &&
+     name = "POST /api/orders" &&
+     span.http.status_code >= 500
+   }
+   ```
+
+3. **Find traces with a particular behavior:**
+   ```
+   {span.service.name="frontend" && name = "GET /api/products/{id}"} && {span.db.system="postgresql"}
+   ```
+
+4. **Find traces across environments:**
+   ```
+   { resource.deployment.environment = "production" } && { resource.deployment.environment = "staging" }
+   ```
+
+5. **Structural operators (advanced):**
+   ```
+   { resource.service.name="frontend" } >> { status = error }  # Frontend spans followed by errors
+   { } !< { resource.service.name = "productcatalogservice" }  # Traces without productcatalog as child
+   { resource.service.name = "productcatalogservice" } ~ { resource.service.name="frontend" }  # Sibling spans
+   ```
+
+6. **Additional operator examples:**
+   ```
+   { span.http.method = "GET" && status = ok } && { span.http.method = "DELETE" && status != ok }  # && for multiple conditions
+   ```
+
+   ```
+   { resource.deployment.environment =~ "prod-.*" && span.http.status_code = 200 }  # =~ regex match
+   { span.http.method =~ "DELETE|GET" }  # Regex match multiple values
+   { trace:rootName !~ ".*perf.*" }  # !~ negated regex
+   { resource.cloud.region = "us-east-1" } || { resource.cloud.region = "us-west-1" }  # || OR operator
+   ```
+
+   ```
+   { span.http.status_code >= 400 && span.http.status_code < 500 }  # Client errors (4xx)
+   { span.http.url = "/path/of/api" } >> { span.db.name = "db-shard-001" }  # >> descendant
+   { span.http.status_code = 200 } | select(resource.service.name)  # Select specific attributes
+   ```
+
+**Common Attributes to Query:**
+- `resource.service.name` - Service name
+- `resource.k8s.*` - Kubernetes metadata (pod.name, namespace.name, deployment.name, etc.)
+- `span.http.*` - HTTP attributes (status_code, method, route, url, etc.)
+- `name` - Span name
+- `status` - Span status (error, ok)
+- `duration` - Span duration
+- `kind` - Span kind (server, client, producer, consumer, internal)
+
+**Tag-based Search (legacy):**
+Use `tempo_search_traces_by_tags` with logfmt format when you need min/max duration filters:
+- Example: `service.name="api" http.status_code="500"`
+- Supports `min_duration` and `max_duration` parameters
+
+### 3. Analyzing Specific Traces
+When you have trace IDs from search results:
+- Use `tempo_query_trace_by_id` to get full trace details
+- Examine spans for errors, slow operations, and bottlenecks
+
+### 4. Computing Metrics from Traces
+**TraceQL metrics** compute aggregated metrics from your trace data, helping you answer critical questions like:
+- How many database calls across all systems are downstream of your application?
+- What services beneath a given endpoint are failing?
+- What services beneath an endpoint are slow?
+
+TraceQL metrics parse your traces in aggregate to provide RED (Rate, Error, Duration) metrics from trace data.
+
+**Supported Functions:**
+- `rate` - Calculate rate of spans/traces
+- `count_over_time` - Count spans/traces over time
+- `sum_over_time` - Sum span attributes
+- `avg_over_time` - Average of span attributes
+- `max_over_time` - Maximum value over time
+- `min_over_time` - Minimum value over time
+- `quantile_over_time` - Calculate quantiles
+- `histogram_over_time` - Generate histogram data
+- `compare` - Compare metrics between time periods
+
+**Modifiers:**
+- `topk` - Return top N results
+- `bottomk` - Return bottom N results
+
+**TraceQL Metrics Query Examples:**
+
+1. **rate** - Calculate error rate by service and HTTP route:
+   ```
+   { resource.service.name = "foo" && status = error } | rate() by (span.http.route)
+   ```
+
+2. **count_over_time** - Count spans by HTTP status code:
+   ```
+   { name = "GET /:endpoint" } | count_over_time() by (span.http.status_code)
+   ```
+
+3. **sum_over_time** - Sum HTTP response sizes by service:
+   ```
+   { name = "GET /:endpoint" } | sum_over_time(span.http.response.size) by (resource.service.name)
+   ```
+
+4. **avg_over_time** - Average duration by HTTP status code:
+   ```
+   { name = "GET /:endpoint" } | avg_over_time(duration) by (span.http.status_code)
+   ```
+
+5. **max_over_time** - Maximum response size by HTTP target:
+   ```
+   { name = "GET /:endpoint" } | max_over_time(span.http.response.size) by (span.http.target)
+   ```
+
+6. **min_over_time** - Minimum duration by HTTP target:
+   ```
+   { name = "GET /:endpoint" } | min_over_time(duration) by (span.http.target)
+   ```
+
+7. **quantile_over_time** - Calculate multiple percentiles (99th, 90th, 50th) with exemplars:
+   ```
+   { span:name = "GET /:endpoint" } | quantile_over_time(duration, .99, .9, .5) by (span.http.target) with (exemplars=true)
+   ```
+
+8. **histogram_over_time** - Build duration histogram grouped by custom attribute:
+   ```
+   { name = "GET /:endpoint" } | histogram_over_time(duration) by (span.foo)
+   ```
+
+9. **compare** - Compare error spans against baseline (10 attributes):
+   ```
+   { resource.service.name="a" && span.http.path="/myapi" } | compare({status=error}, 10)
+   ```
+
+10. **Using topk modifier** - Find top 10 endpoints by request rate:
+   ```
+   { resource.service.name = "foo" } | rate() by (span.http.url) | topk(10)
+   ```
+
+**Choosing Between Instant and Range Queries:**
+
+**Instant Metrics** (`tempo_query_metrics_instant`) - Returns a single aggregated value for the entire time range. Use this when:
+- You need a total count or sum across the whole period
+- You want a single metric value (e.g., total error count, average latency)
+- You don't need to see how the metric changes over time
+- You're computing a KPI or summary statistic
+
+**Time Series Metrics** (`tempo_query_metrics_range`) - Returns values at regular intervals controlled by the 'step' parameter. Use this when:
+- You need to graph metrics over time or analyze trends
+- You want to see patterns, spikes, or changes in metrics
+- You're troubleshooting time-based issues
+- You need to correlate metrics with specific time periods
+
+## Special workflow for performance issues
+When investigating performance issues in kubernetes via traces, call tempo_fetch_traces_comparative_sample. This tool provides comprehensive analysis for identifying patterns.
+
+## Important Notes
+- TraceQL is the modern query language - prefer it over tag-based search
+- TraceQL metrics are computed from trace data, not traditional Prometheus metrics
+- TraceQL metrics is an experimental feature that computes RED (Rate, Error, Duration) metrics from trace data
+- Common attributes to use in queries: resource.service.name, span.http.route, span.http.status_code, span.http.target, status, name, duration
+- All timestamps can be Unix epoch seconds or RFC3339 format
+- Use time filters (start/end) to improve query performance
+- To get information about Kubernetes resources try these first: resource.service.name, resource.k8s.pod.name, resource.k8s.namespace.name, resource.k8s.deployment.name, resource.k8s.node.name, resource.k8s.container.name
+- TraceQL and TraceQL metrics language are complex. If you get empty data, try to simplify your query and try again!
+- IMPORTANT: TraceQL is not the same as 'TraceQL metrics' - Make sure you use the correct syntax and functions