tinkermonkey/rounds-explore

Rounds: Explore Telemetry

Explore OTEL data held in the configured telemetry backend (SigNoz by default) using the rounds adapter interfaces. No LLM budget is consumed — these are direct queries.

Usage

/rounds-explore [optional natural-language description of what to look for]

If $ARGUMENTS is provided, use it to guide which queries to run first (e.g. "look for database timeout errors in the last 2 hours" or "show me what the checkout service has been doing"). If no arguments are given, start with a broad recent-activity scan.

---

Available Python Commands

All three commands run via python -m rounds.main cli-run <command> '<json>' from /workspace/rounds.

1. Search Logs — keyword and metadata filters

cd /workspace/rounds && python -m rounds.main cli-run search-logs '{
  "query": "timeout",
  "since_minutes": 120,
  "services": ["checkout-service"],
  "limit": 50
}'

Parameters:

• query (string, default "") — keyword to match in log bodies; empty string returns all logs
• since_minutes (int, default 60) — lookback window from now
• until_minutes (int, optional) — upper bound in minutes ago (creates a closed time window)
• services (list of strings, optional) — filter by service name
• limit (int, default 50) — max results

Response shape:

{
  "status": "success",
  "operation": "search-logs",
  "query": "timeout",
  "count": 3,
  "logs": [
    {
      "timestamp": "2024-01-15T10:30:00+00:00",
      "severity": "ERROR",
      "body": "Connection timeout after 30s",
      "trace_id": "abc123...",
      "span_id": "def456...",
      "attributes": { "service.name": "checkout-service", "http.status_code": "504" }
    }
  ]
}

Log exploration patterns:

Search for a specific error keyword:

python -m rounds.main cli-run search-logs '{"query": "NullPointerException", "since_minutes": 60}'

Search for database-related logs across all services:

python -m rounds.main cli-run search-logs '{"query": "database", "since_minutes": 240, "limit": 100}'

Look at all recent logs from one service (no keyword filter):

python -m rounds.main cli-run search-logs '{"query": "", "since_minutes": 30, "services": ["api-gateway"], "limit": 100}'

Narrow a time window (between 2 and 1 hours ago):

python -m rounds.main cli-run search-logs '{"query": "error", "since_minutes": 120, "until_minutes": 60}'

---

2. Search Spans — metadata filters

cd /workspace/rounds && python -m rounds.main cli-run search-spans '{
  "since_minutes": 60,
  "services": ["payment-service"],
  "has_error": true,
  "limit": 50
}'

Parameters:

• since_minutes (int, default 60) — lookback window from now
• until_minutes (int, optional) — upper bound in minutes ago from now
• services (list of strings, optional) — filter by service name
• operation (string, optional) — operation name substring filter (e.g. "POST /checkout")
• has_error (bool, optional) — true for error spans only, false for healthy spans, omit for all
• limit (int, default 50) — max results

Response shape:

{
  "status": "success",
  "operation": "search-spans",
  "count": 2,
  "spans": [
    {
      "trace_id": "abc123def456...",
      "span_id": "789xyz...",
      "service": "payment-service",
      "operation": "POST /charge",
      "duration_ms": 1234.5,
      "has_error": true,
      "timestamp": "2024-01-15T10:30:00+00:00",
      "attributes": { "http.status_code": "500", "rpc.method": "Charge" }
    }
  ]
}

Span exploration patterns:

Find all error spans in the last hour:

python -m rounds.main cli-run search-spans '{"since_minutes": 60, "has_error": true, "limit": 100}'

Find slow operations matching a name pattern:

python -m rounds.main cli-run search-spans '{"since_minutes": 30, "operation": "checkout", "limit": 50}'

Survey recent activity in a service (all spans, no error filter):

python -m rounds.main cli-run search-spans '{"since_minutes": 15, "services": ["inventory-service"], "limit": 100}'

The trace_id values in results can be passed directly to get-trace-tree or investigate-trace to drill deeper.

---

3. Get Trace Tree — full transaction hierarchy

cd /workspace/rounds && python -m rounds.main cli-run get-trace-tree '{
  "trace_id": "abc123def456789012345678901234ab"
}'

Parameters:

• trace_id (string, required) — 32-character hex OpenTelemetry trace ID

Response shape:

{
  "status": "success",
  "operation": "get-trace-tree",
  "trace_id": "abc123...",
  "error_span_count": 1,
  "tree": {
    "span_id": "root-span-id",
    "service": "api-gateway",
    "operation": "POST /checkout",
    "duration_ms": 1500.0,
    "status": "error",
    "attributes": { "http.method": "POST", "http.url": "/checkout" },
    "events": [],
    "children": [
      {
        "span_id": "child-span-id",
        "service": "payment-service",
        "operation": "ProcessPayment",
        "duration_ms": 800.0,
        "status": "ok",
        "attributes": {},
        "events": [],
        "children": []
      }
    ]
  }
}

Tree exploration patterns:

Get the full transaction tree for a trace ID found via search-spans:

python -m rounds.main cli-run get-trace-tree '{"trace_id": "abc123def456789012345678901234ab"}'

The tree is returned as a nested structure. Walk it to:

• Identify which service originated the request (root span)
• See the full call chain (children recursively)
• Find all error spans ("status": "error") and their positions in the tree
• Measure per-span durations to spot bottlenecks
• Inspect span attributes for HTTP status codes, DB queries, RPC methods

After reviewing the tree, follow up with get-correlated-logs or trigger LLM analysis:

# LLM-powered code-flow explanation (uses diagnosis budget):
python -m rounds.main cli-run investigate-trace '{"trace_id": "abc123..."}'

---

Typical Exploratory Workflow

1. Broad scan — find recent error spans:

   python -m rounds.main cli-run search-spans '{"since_minutes": 60, "has_error": true, "limit": 50}'
   

2. Narrow by service or operation — pick an interesting service from step 1:

   python -m rounds.main cli-run search-spans '{"since_minutes": 60, "services": ["payment-service"], "has_error": true}'
   

3. Look for correlated log messages — search logs around the same time:

   python -m rounds.main cli-run search-logs '{"query": "payment", "since_minutes": 60, "services": ["payment-service"]}'
   

4. Build the full transaction tree — take a trace_id from step 2:

   python -m rounds.main cli-run get-trace-tree '{"trace_id": "<trace_id_from_step_2>"}'
   

5. Correlate logs with that trace — use the existing correlated-logs path:

   python -m rounds.main cli-run search-logs '{"query": "", "since_minutes": 10, "services": ["payment-service"]}'
   

6. Trigger LLM analysis — once you have a trace ID worth diagnosing:

   python -m rounds.main cli-run investigate-trace '{"trace_id": "<trace_id>"}'
   

---

Implementation Notes

• All three commands call into ManagementService → TelemetryPort (SigNoz adapter by default)
• search_logs uses SigNoz /api/v3/query_range with dataSource=logs and a body contains filter
• search_spans uses /api/v3/query_range with dataSource=traces and flexible AND-combined filters
• get_trace_tree calls GET /api/v1/traces/{id} and builds the full SpanNode hierarchy
• No LLM calls are made; these are pure telemetry queries
• Source: rounds/adapters/telemetry/signoz.py — search_logs, search_spans methods
• Source: rounds/adapters/cli/commands.py — search_logs, search_spans, get_trace_tree handlers