get_causal_chains
Analyze CUDA and host events to identify causal chains explaining GPU latency, providing severity, root cause, and recommendations for debugging.
Instructions
Analyze CUDA + host events and return causal chains with severity, root cause, and recommendations. AI-first: TSC-compressed by default.
Input Schema
TableJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
| since | No | time range, e.g. 1m, 5m. Default: all data (0 = no time filter) | |
| pid | No | filter by single process ID. 0 = all. Deprecated: use pids. | |
| pids | No | filter by process ID(s). Takes precedence over pid. | |
| tsc | No | telegraphic compression (default: true) |
Output Schema
TableJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
No arguments | |||
Implementation Reference
- The `get_causal_chains` tool is called via the `MCPClient.call` method in the `MCPClient` class. It is registered as a tool method for convenience within the investigation scripts.
def get_causal_chains(self, since: str = "10m") -> dict: """Get causal chains (120s timeout — replay is expensive on large DBs).""" return self.call("get_causal_chains", {"since": since, "tsc": False}, timeout=120)