feat: exporter prometheus_client /metrics (JEP-0013 Phase 2)

docs/source/contributing/jeps/JEP-0013-observability-telemetry-logs.md

@@ -1601,6 +1601,17 @@ all subsequent phases have E2E coverage from the start.
   - document tested pairs; W3C Trace Context in gRPC remains
   best-effort across Python and Go (no OTel SDK requirement to
   propagate `traceparent` where needed).
+- **`operation` as a Prometheus label (Phase 2):** exporter series use
+  `@export` method names (`on`, `off`, `flash`, …) as the `operation`
+  label. The set is finite per process (loaded drivers), and unknown
+  methods are not recorded. If cardinality grows in the field, evaluate
+  moving `operation` from a series label to an exemplar key (see
+  *Cardinality guidelines*).
+- **Per-chunk `jumpstarter_stream_bytes_total` increments (Phase 2):**
+  `copy_stream` calls `Counter.inc()` once per chunk. That is in-process
+  and uses independent tx/rx label sets, so Phase 2 accepts it. If flash
+  or storage throughput regresses, batch byte counts and flush
+  periodically instead of incrementing on every chunk.
 
 ## Rejected Alternatives
 
@@ -1669,6 +1680,11 @@ all subsequent phases have E2E coverage from the start.
 - Event retention: Loki retention policy (per-tenant, per-stream retention
   classes) for annotated log events (**DD-2**); whether Jumpstarter should
   document recommended retention defaults or leave this to operators.
+- Whether the exporter `operation` metric label should stay a bounded
+  series dimension or move to exemplars if cardinality grows (see
+  *Risks*).
+- Whether `jumpstarter_stream_bytes_total` should batch per-chunk
+  increments if flash performance degrades (see *Risks*).
 
 ## Future Possibilities
 

python/packages/jumpstarter-cli/jumpstarter_cli/run.py

@@ -11,6 +11,8 @@
 from jumpstarter_cli_common.config import opt_config
 from jumpstarter_cli_common.exceptions import handle_exceptions
 
+from jumpstarter.metrics import start_metrics_server
+
 logger = logging.getLogger(__name__)
 
 
@@ -70,7 +72,15 @@ def _reap_zombie_processes(capture_child=None):
         logger.warning(f"PARENT: Error during zombie reaping: {e}")
 
 
-def _handle_child(config, parsed_bind=None, tls_insecure=False, tls_cert=None, tls_key=None, passphrase=None):  # noqa: C901
+def _handle_child(  # noqa: C901
+    config,
+    parsed_bind=None,
+    tls_insecure=False,
+    tls_cert=None,
+    tls_key=None,
+    passphrase=None,
+    metrics_bind_address=":0",
+):
     """Handle child process with graceful shutdown."""
     async def serve_with_graceful_shutdown():  # noqa: C901
         received_signal = 0
@@ -99,46 +109,54 @@ async def signal_handler():
             # Start signal handler immediately
             signal_tg.start_soon(signal_handler)
 
-            if parsed_bind is not None:
-                host, port = parsed_bind
-                tls_credentials = None
-                if tls_insecure:
-                    if passphrase:
-                        click.echo(
-                            "WARNING: --passphrase has no effect without TLS; "
-                            "the passphrase will be transmitted in plaintext",
-                            err=True,
-                        )
-                elif tls_cert and tls_key:
-                    tls_credentials = _tls_server_credentials(tls_cert, tls_key)
-
-                interceptors = None
-                if passphrase:
-                    from jumpstarter.exporter.auth import PassphraseInterceptor
-                    interceptors = [PassphraseInterceptor(passphrase)]
-
-                exporter_exit_code = None
-                async with config.create_exporter(standalone=True) as exporter:
-                    try:
-                        await exporter.serve_standalone_tcp(
-                            host, port,
-                            tls_credentials=tls_credentials,
-                            interceptors=interceptors,
-                        )
-                    except* Exception as excgroup:
-                        _handle_exporter_exceptions(excgroup)
-                    exporter_exit_code = exporter.exit_code
-            else:
-                # Create exporter and run it (controller mode)
-                exporter_exit_code = None
-                async with config.create_exporter() as exporter:
-                    try:
-                        await exporter.serve()
-                    except* Exception as excgroup:
-                        _handle_exporter_exceptions(excgroup)
+            listen_addr, shutdown_metrics = start_metrics_server(metrics_bind_address)
+            if listen_addr:
+                logger.info("Serving metrics server at http://%s/metrics", listen_addr)
 
-                    # Check if exporter set an exit code (e.g., from hook failure with on_failure='exit')
-                    exporter_exit_code = exporter.exit_code
+            try:
+                if parsed_bind is not None:
+                    host, port = parsed_bind
+                    tls_credentials = None
+                    if tls_insecure:
+                        if passphrase:
+                            click.echo(
+                                "WARNING: --passphrase has no effect without TLS; "
+                                "the passphrase will be transmitted in plaintext",
+                                err=True,
+                            )
+                    elif tls_cert and tls_key:
+                        tls_credentials = _tls_server_credentials(tls_cert, tls_key)
+
+                    interceptors = None
+                    if passphrase:
+                        from jumpstarter.exporter.auth import PassphraseInterceptor
+                        interceptors = [PassphraseInterceptor(passphrase)]
+
+                    exporter_exit_code = None
+                    async with config.create_exporter(standalone=True) as exporter:
+                        try:
+                            await exporter.serve_standalone_tcp(
+                                host, port,
+                                tls_credentials=tls_credentials,
+                                interceptors=interceptors,
+                            )
+                        except* Exception as excgroup:
+                            _handle_exporter_exceptions(excgroup)
+                        exporter_exit_code = exporter.exit_code
+                else:
+                    # Create exporter and run it (controller mode)
+                    exporter_exit_code = None
+                    async with config.create_exporter() as exporter:
+                        try:
+                            await exporter.serve()
+                        except* Exception as excgroup:
+                            _handle_exporter_exceptions(excgroup)
+
+                        # Check if exporter set an exit code (e.g., from hook failure with on_failure='exit')
+                        exporter_exit_code = exporter.exit_code
+            finally:
+                if shutdown_metrics is not None:
+                    shutdown_metrics()
 
             # Cancel the signal handler after exporter completes
             signal_tg.cancel_scope.cancel()
@@ -204,7 +222,13 @@ def parent_signal_handler(signum, _):
 
 
 def _serve_with_exc_handling(
-    config, parsed_bind=None, tls_insecure=False, tls_cert=None, tls_key=None, passphrase=None
+    config,
+    parsed_bind=None,
+    tls_insecure=False,
+    tls_cert=None,
+    tls_key=None,
+    passphrase=None,
+    metrics_bind_address=":0",
 ):
     max_rapid_failures = config.failure_detection.max_rapid_failures
     rapid_failure_window = config.failure_detection.rapid_failure_window
@@ -253,7 +277,15 @@ def _serve_with_exc_handling(
                 rapid_failure_count = 0
         else:
             os.setsid() # Become group leader so all spawned subprocesses are reached by parent's signals
-            _handle_child(config, parsed_bind, tls_insecure, tls_cert, tls_key, passphrase)
+            _handle_child(
+                config,
+                parsed_bind,
+                tls_insecure,
+                tls_cert,
+                tls_key,
+                passphrase,
+                metrics_bind_address,
+            )
             sys.exit(1) # should never happen
 
 
@@ -294,8 +326,27 @@ def _serve_with_exc_handling(
     default=False,
     help="Exit after the current lease ends instead of waiting for a new one.",
 )
+@click.option(
+    "--metrics-bind-address",
+    "metrics_bind_address",
+    default=":0",
+    show_default=True,
+    help=(
+        "Address for HTTP GET /metrics (Prometheus/OpenMetrics). "
+        "Default :0 binds an ephemeral loopback port. Use 0 to disable."
+    ),
+)
 @handle_exceptions
-def run(config, listener_bind, tls_insecure, tls_cert, tls_key, passphrase, exit_on_lease_end):
+def run(
+    config,
+    listener_bind,
+    tls_insecure,
+    tls_cert,
+    tls_key,
+    passphrase,
+    exit_on_lease_end,
+    metrics_bind_address,
+):
     """Run an exporter locally."""
     if listener_bind is not None and config is None:
         raise click.UsageError("--exporter-config (or --exporter) is required when using --tls-grpc-listener")
@@ -313,4 +364,12 @@ def run(config, listener_bind, tls_insecure, tls_cert, tls_key, passphrase, exit
     if exit_on_lease_end:
         config.exit_on_lease_end = True
     parsed_bind = _parse_listener_bind(listener_bind) if listener_bind is not None else None
-    return _serve_with_exc_handling(config, parsed_bind, tls_insecure, tls_cert, tls_key, passphrase)
+    return _serve_with_exc_handling(
+        config,
+        parsed_bind,
+        tls_insecure,
+        tls_cert,
+        tls_key,
+        passphrase,
+        metrics_bind_address,
+    )