Read job id from env

composer/diagnostics/timing.py

@@ -8,6 +8,7 @@
 """
 
 from contextlib import asynccontextmanager, contextmanager
+import os
 from logging import Logger
 import time
 from contextvars import ContextVar
@@ -75,6 +76,10 @@ class PhaseRecord:
     final_link: str | None = None # URL (cloud) or local results directory (local) of the last prover run.
     token_usage_by_model: dict[str, TokenTotals] = field(default_factory=dict)
 
+def run_id_generator() -> str:
+    if (res := os.getenv("AUTOPROVER_JOB_ID")):
+        return res
+    return uuid.uuid4().hex
 
 @dataclass
 class RunSummary:
@@ -85,7 +90,7 @@ class RunSummary:
     _active_prover_by_task: dict[str, tuple[float, int]] = field(default_factory=dict, repr=False)
     prover_reported_ms_total: int = 0  # Run-wide prover-REPORTED runtime, summed over every prover run. Distinct from prover_total_s, which is composer's client-side wall-clock (and so includes cloud queue / polling / result download).
     _active_prover_reported_by_task: dict[str, int] = field(default_factory=dict, repr=False)  # Maps task_id -> prover-reported ms accumulated while the task is in flight.
-    run_id: str = field(default_factory=lambda: uuid.uuid4().hex)  # Maps task_id -> (prover_s_accum, prover_calls) recorded while task is in flight.
+    run_id: str = field(default_factory=run_id_generator)  # Maps task_id -> (prover_s_accum, prover_calls) recorded while task is in flight.
     _latest_link_by_task: dict[str, str] = field(default_factory=dict, repr=False)  # Maps task_id -> link for the most recent prover run.
     token_usage_by_model: dict[str, TokenTotals] = field(default_factory=dict)  # Maps model_name -> accumulated raw token counts across the whole run.
     _active_tokens_by_task: dict[str, dict[str, TokenTotals]] = field(default_factory=dict, repr=False)  # Maps task_id -> {model_name -> token counts} accumulated while the task is in flight.

composer/pipeline/cli.py

@@ -155,6 +155,7 @@ async def cli_pipeline[P: enum.Enum, H](
     model = get_model()
     text_log, events_log = setup_autoprove_logging(project_root, thread_id)
     print(f"autoprove logs: {text_log}\n           events: {events_log}", file=sys.stderr)
+    print(f"Selected run id: {summary.run_id}")
     install_run_summary(summary)
 
     disc_cache_ns: tuple[str, ...] | None = (