Pass jfr, gc and jit telemetry devices to the provisioned pipelines

docs/reference/telemetry.md

@@ -214,32 +214,32 @@ Records Solr version, JVM version, and hardware info as run metadata. Called onc
 
 Enables Java Flight Recorder. Injects `-XX:StartFlightRecording=...` into `SOLR_OPTS`.
 
-**Pipeline:** `docker` or `from-distribution` only
+**Pipeline:** `from-distribution` or `from-sources` (Docker support is tracked in [#138](https://github.com/apache/solr-orbit/issues/138))
 **Enable:** `--telemetry jfr`
 
-The flight recording is written to `profile.jfr` in the benchmark log directory. Requires OpenJDK 11 or later. To use a custom recording template, pass `--telemetry-params recording-template:/path/to/template.jfc`.
+The flight recording is written to `profile.jfr` in the `telemetry` directory of the test run (`~/.solr-orbit/benchmarks/test-runs/<test-run-id>/telemetry/`). Requires OpenJDK 11 or later. To use a custom recording template, pass `--telemetry-params recording-template:/path/to/template.jfc`.
 
 ---
 
 ### gc
 
 Enables GC logging (Java 9+ `-Xlog:` format). Injects the GC log flags into `SOLR_OPTS`.
 
-**Pipeline:** `docker` or `from-distribution` only
+**Pipeline:** `from-distribution` or `from-sources` (Docker support is tracked in [#138](https://github.com/apache/solr-orbit/issues/138))
 **Enable:** `--telemetry gc`
 
-GC output is written to `gc.log` in the benchmark log directory. The default log configuration is `gc*=info,safepoint=info,age*=trace`. Override with `--telemetry-params gc-log-config:gc*=debug`.
+GC output is written to `gc.log` in the `telemetry` directory of the test run (`~/.solr-orbit/benchmarks/test-runs/<test-run-id>/telemetry/`). The default log configuration is `gc*=info,safepoint=info,age*=trace`. Override with `--telemetry-params gc-log-config:gc*=debug`.
 
 ---
 
 ### jit
 
 Enables JIT compiler logging. Injects `-XX:+LogCompilation` and related flags into `SOLR_OPTS`.
 
-**Pipeline:** `docker` or `from-distribution` only
+**Pipeline:** `from-distribution` or `from-sources` (Docker support is tracked in [#138](https://github.com/apache/solr-orbit/issues/138))
 **Enable:** `--telemetry jit`
 
-JIT output is written to `jit.log` in the benchmark log directory.
+JIT output is written to `jit.log` in the `telemetry` directory of the test run (`~/.solr-orbit/benchmarks/test-runs/<test-run-id>/telemetry/`).
 
 ---
 
@@ -280,7 +280,7 @@ Add JVM profiling devices on provisioned pipelines:
 
 ```bash
 solr-orbit run \
-  --pipeline docker \
+  --pipeline from-distribution \
   --distribution-version 9.10.1 \
   --workload nyc_taxis \
   --telemetry gc,jfr

solrorbit/test_run_orchestrator.py

@@ -38,7 +38,7 @@
 
 from solrorbit import actor, config, doc_link, \
     worker_coordinator, exceptions, builder, metrics, \
-        publisher, workload, version, PROGRAM_NAME
+        publisher, workload, version, paths, telemetry, PROGRAM_NAME
 from solrorbit.builder import cluster_config as cc_module
 from solrorbit.builder.supplier import SourceRepository, Builder
 from solrorbit.builder.solr_provisioner import SolrProvisioner, SolrDockerLauncher
@@ -462,6 +462,25 @@ def _load_cluster_config(cfg):
     return instance
 
 
+def _create_jvm_telemetry_devices(cfg):
+    """
+    Build the JVM telemetry devices (``jfr``, ``gc``, ``jit``) enabled with ``--telemetry``.
+
+    ``SolrProvisioner`` adds the ``instrument_java_opts()`` of every device it is given to
+    ``SOLR_OPTS``, so only the enabled ones are returned. Their output goes to a
+    ``telemetry`` directory inside the test run directory.
+    """
+    enabled_devices = cfg.opts("telemetry", "devices", mandatory=False, default_value=[]) or []
+    telemetry_params = cfg.opts("telemetry", "params", mandatory=False, default_value={}) or {}
+    log_root = os.path.join(paths.test_run_root(cfg), "telemetry")
+    devices = [
+        telemetry.FlightRecorder(telemetry_params, log_root, java_major_version=None),
+        telemetry.Gc(telemetry_params, log_root, java_major_version=None),
+        telemetry.JitCompiler(log_root, telemetry_params),
+    ]
+    return [device for device in devices if device.command in enabled_devices]
+
+
 def solr_from_sources(cfg):
     """
     Clone/update the Solr source tree, build a distribution with Gradle, provision
@@ -517,7 +536,8 @@ def solr_from_sources(cfg):
     cc_instance = _load_cluster_config(cfg)
     solr_modules = cfg.opts("solr", "modules", mandatory=False, default_value="")
     provisioner = SolrProvisioner(cache_dir=os.path.join(base_dir, "cache"), port=port,
-                                  cluster_config=cc_instance, solr_modules=solr_modules)
+                                  cluster_config=cc_instance, solr_modules=solr_modules,
+                                  telemetry_devices=_create_jvm_telemetry_devices(cfg))
     try:
         provisioner.start(solr_root, mode="cloud")
         set_default_hosts(cfg, host="127.0.0.1", port=port)
@@ -562,7 +582,8 @@ def solr_from_distribution(cfg):
     cc_instance = _load_cluster_config(cfg)
     solr_modules = cfg.opts("solr", "modules", mandatory=False, default_value="")
     provisioner = SolrProvisioner(cache_dir=cache_dir, port=port, cluster_config=cc_instance,
-                                  solr_modules=solr_modules)
+                                  solr_modules=solr_modules,
+                                  telemetry_devices=_create_jvm_telemetry_devices(cfg))
     _tarball = provisioner.download(version_str)
     solr_root = provisioner.install(version_str, install_dir)
 

tests/test_execution_orchestrator_test.py

@@ -32,6 +32,7 @@
 import pytest
 
 from solrorbit import config, exceptions, test_run_orchestrator
+from solrorbit.builder.solr_provisioner import SolrProvisioner
 from solrorbit.utils import opts
 
 
@@ -151,3 +152,81 @@ def test_check_cloud_mode_raises_for_standalone_solr():
         admin_client.return_value.is_cloud_mode.return_value = False
         with pytest.raises(exceptions.SystemSetupError, match="not running in SolrCloud"):
             test_run_orchestrator._check_cloud_mode(cfg)
+
+
+TEST_RUN_ID = "28a032d1-0b03-4579-ad2a-c65316f126e9"
+
+
+def _provisioned_cfg(root_dir, devices, params=None):
+    cfg = config.Config()
+    cfg.add(config.Scope.benchmark, "system", "test_run.id", TEST_RUN_ID)
+    cfg.add(config.Scope.benchmark, "node", "root.dir", str(root_dir))
+    cfg.add(config.Scope.benchmark, "telemetry", "devices", devices)
+    cfg.add(config.Scope.benchmark, "telemetry", "params", params or {})
+    cfg.add(config.Scope.benchmark, "builder", "distribution.version", "9.10.1")
+    cfg.add(config.Scope.benchmark, "solr", "install_dir", str(root_dir / "install"))
+    cfg.add(config.Scope.benchmark, "solr", "cache_dir", str(root_dir / "cache"))
+    return cfg
+
+
+def test_create_jvm_telemetry_devices_returns_only_enabled_devices(tmp_path):
+    cfg = _provisioned_cfg(tmp_path, ["gc", "jfr", "segment-stats"])
+
+    devices = test_run_orchestrator._create_jvm_telemetry_devices(cfg)
+
+    assert sorted(device.command for device in devices) == ["gc", "jfr"]
+    expected_log_root = os.path.join(str(tmp_path), "test-runs", TEST_RUN_ID, "telemetry")
+    assert all(device.log_root == expected_log_root for device in devices)
+
+
+def test_create_jvm_telemetry_devices_returns_nothing_without_telemetry(tmp_path):
+    cfg = _provisioned_cfg(tmp_path, [])
+
+    assert test_run_orchestrator._create_jvm_telemetry_devices(cfg) == []
+
+
+def test_jvm_telemetry_devices_add_their_flags_to_solr_opts(tmp_path):
+    cfg = _provisioned_cfg(tmp_path, ["gc", "jfr", "jit"], params={"gc-log-config": "gc*=debug"})
+    devices = test_run_orchestrator._create_jvm_telemetry_devices(cfg)
+
+    provisioner = SolrProvisioner(cache_dir=str(tmp_path / "cache"), telemetry_devices=devices)
+    solr_opts = provisioner._build_env()["SOLR_OPTS"]
+
+    log_root = os.path.join(str(tmp_path), "test-runs", TEST_RUN_ID, "telemetry")
+    assert f"-Xlog:gc*=debug:file={os.path.join(log_root, 'gc.log')}" in solr_opts
+    assert f"filename={os.path.join(log_root, 'profile.jfr')}" in solr_opts
+    assert f"-XX:LogFile={os.path.join(log_root, 'jit.log')}" in solr_opts
+
+
+@mock.patch("solrorbit.test_run_orchestrator.run_test")
+@mock.patch("solrorbit.test_run_orchestrator.set_default_hosts")
+@mock.patch("solrorbit.test_run_orchestrator._load_cluster_config", return_value=None)
+@mock.patch("solrorbit.test_run_orchestrator.SolrProvisioner")
+def test_from_distribution_passes_jvm_telemetry_devices(provisioner_class, load_cluster_config, set_default_hosts, run_test, tmp_path):
+    cfg = _provisioned_cfg(tmp_path, ["gc", "jit"])
+
+    test_run_orchestrator.solr_from_distribution(cfg)
+
+    devices = provisioner_class.call_args.kwargs["telemetry_devices"]
+    assert sorted(device.command for device in devices) == ["gc", "jit"]
+    run_test.assert_called_once_with(cfg, external=True)
+
+
+@mock.patch("solrorbit.test_run_orchestrator.run_test")
+@mock.patch("solrorbit.test_run_orchestrator.set_default_hosts")
+@mock.patch("solrorbit.test_run_orchestrator._load_cluster_config", return_value=None)
+@mock.patch("solrorbit.test_run_orchestrator.SolrProvisioner")
+@mock.patch("solrorbit.test_run_orchestrator.tarfile.open")
+@mock.patch("solrorbit.test_run_orchestrator.glob.glob", return_value=["/src/solr-10.0.0-SNAPSHOT.tgz"])
+@mock.patch("solrorbit.test_run_orchestrator.Builder")
+@mock.patch("solrorbit.test_run_orchestrator.SourceRepository")
+def test_from_sources_passes_jvm_telemetry_devices(source_repository, bldr, glob_glob, tarfile_open, provisioner_class,
+                                                   load_cluster_config, set_default_hosts, run_test, tmp_path):
+    tarfile_open.return_value.__enter__.return_value.getnames.return_value = ["solr-10.0.0-SNAPSHOT/bin/solr"]
+    cfg = _provisioned_cfg(tmp_path, ["jfr"])
+
+    test_run_orchestrator.solr_from_sources(cfg)
+
+    devices = provisioner_class.call_args.kwargs["telemetry_devices"]
+    assert [device.command for device in devices] == ["jfr"]
+    run_test.assert_called_once_with(cfg, external=True)