Make NodeJS profiling attachable

Dockerfile

@@ -21,6 +21,10 @@ ARG AP_BUILDER_ALPINE=@sha256:69704ef328d05a9f806b6b8502915e6a0a4faa4d72018dc423
 ARG BURN_BUILDER_GOLANG=@sha256:f7d3519759ba6988a2b73b5874b17c5958ac7d0aa48a8b1d84d66ef25fa345f1
 # gprofiler - ubuntu 20.04
 ARG GPROFILER_BUILDER_UBUNTU=@sha256:cf31af331f38d1d7158470e095b132acd126a7180a54f263d386da88eb681d93
+# node-package-builder-musl alpine
+ARG NODE_PACKAGE_BUILDER_MUSL=@sha256:69704ef328d05a9f806b6b8502915e6a0a4faa4d72018dc42343f511490daf8a
+# node-package-builder-glibc - ubuntu:20.04
+ARG NODE_PACKAGE_BUILDER_GLIBC=@sha256:cf31af331f38d1d7158470e095b132acd126a7180a54f263d386da88eb681d93
 
 # pyspy & rbspy builder base
 FROM rust${RUST_BUILDER_VERSION} AS pyspy-rbspy-builder-common
@@ -148,6 +152,22 @@ COPY scripts/async_profiler_build_shared.sh .
 COPY scripts/async_profiler_build_musl.sh .
 RUN ./async_profiler_build_shared.sh /tmp/async_profiler_build_musl.sh
 
+# node-package-builder-musl
+FROM alpine${NODE_PACKAGE_BUILDER_MUSL} AS node-package-builder-musl
+WORKDIR /tmp
+COPY scripts/node_builder_musl_env.sh .
+RUN ./node_builder_musl_env.sh
+COPY scripts/build_node_package.sh .
+RUN ./build_node_package.sh
+
+# node-package-builder-glibc
+FROM ubuntu${NODE_PACKAGE_BUILDER_GLIBC} AS node-package-builder-glibc
+WORKDIR /tmp
+COPY scripts/node_builder_glibc_env.sh .
+RUN ./node_builder_glibc_env.sh
+COPY scripts/build_node_package.sh .
+RUN ./build_node_package.sh
+
 # burn
 FROM golang${BURN_BUILDER_GOLANG} AS burn-builder
 WORKDIR /tmp
@@ -191,6 +211,8 @@ COPY --from=async-profiler-builder-glibc /tmp/async-profiler/build/async-profile
 COPY --from=async-profiler-builder-glibc /tmp/async-profiler/build/libasyncProfiler.so gprofiler/resources/java/glibc/libasyncProfiler.so
 COPY --from=async-profiler-builder-musl /tmp/async-profiler/build/libasyncProfiler.so gprofiler/resources/java/musl/libasyncProfiler.so
 COPY --from=async-profiler-builder-glibc /tmp/async-profiler/build/fdtransfer gprofiler/resources/java/fdtransfer
+COPY --from=node-package-builder-musl /tmp/module_build gprofiler/resources/node/module/musl
+COPY --from=node-package-builder-glibc /tmp/module_build gprofiler/resources/node/module/glibc
 
 COPY --from=rbspy-builder /tmp/rbspy/rbspy gprofiler/resources/ruby/rbspy
 

README.md

@@ -75,6 +75,7 @@ Profiling using eBPF incurs lower overhead & provides kernel & native stacks.
 * `--nodejs-mode`: Controls which profiler is used for NodeJS.
      * `none` - (default) no profiler is used.
      * `perf` - augment the system profiler (`perf`) results with jitdump files generated by NodeJS. This requires running your `node` processes with `--perf-prof` (and for Node >= 10, with `--interpreted-frames-native-stack`). See this [NodeJS page](https://nodejs.org/en/docs/guides/diagnostics-flamegraph/) for more information.
+     * `attach-maps` - Generates perf map using [node-linux-perf module](https://github.com/mmarchini-oss/node-linux-perf). This module is injected at runtime. Requires entrypoint of application to be CommonJS script. (Doesn't work for ES modules)
 
 ### System profiling options
 
@@ -367,6 +368,7 @@ Alongside `perf`, gProfiler invokes runtime-specific profilers for processes bas
   * Uses [Granulate's fork](https://github.com/Granulate/rbspy) of the [rbspy](https://github.com/rbspy/rbspy) profiler.
 * NodeJS (version >= 10 for functioning `--perf-prof`):
   * Uses `perf inject --jit` and NodeJS's ability to generate jitdump files. See [NodeJS profiling options](#nodejs-profiling-options).
+  * Can also generate perf maps at runtime.
 * .NET runtime
   * Uses dotnet-trace.
 

dev-requirements.txt

@@ -10,3 +10,4 @@ types-PyYAML==6.0.3
 types-pkg-resources==0.1.3
 types-protobuf==3.19.22
 types-toml==0.10.8
+types-retry==0.9.9

gprofiler/main.py

@@ -433,9 +433,10 @@ def parse_cmd_args() -> configargparse.Namespace:
         "--nodejs-mode",
         dest="nodejs_mode",
         default="disabled",
-        choices=["perf", "disabled", "none"],
-        help="Select the NodeJS profiling mode: perf (run 'perf inject --jit' on perf results, to augment them"
-        " with jitdump files of NodeJS processes, if present) or disabled (no runtime-specific profilers for NodeJS)",
+        choices=["attach-maps", "perf", "disabled", "none"],
+        help="Select the NodeJS profiling mode: attach-maps (generates perf-maps at runtime),"
+        " perf (run 'perf inject --jit' on perf results, to augment them with jitdump files"
+        " of NodeJS processes, if present) or disabled (no runtime-specific profilers for NodeJS)",
     )
 
     nodejs_options.add_argument(
@@ -601,6 +602,7 @@ def parse_cmd_args() -> configargparse.Namespace:
     args = parser.parse_args()
 
     args.perf_inject = args.nodejs_mode == "perf"
+    args.perf_node_attach = args.nodejs_mode == "attach-maps"
 
     if args.upload_results:
         if not args.server_token:
@@ -617,8 +619,8 @@ def parse_cmd_args() -> configargparse.Namespace:
     if args.perf_mode in ("dwarf", "smart") and args.frequency > 100:
         parser.error("--profiling-frequency|-f can't be larger than 100 when using --perf-mode 'smart' or 'dwarf'")
 
-    if args.nodejs_mode == "perf" and args.perf_mode not in ("fp", "smart"):
-        parser.error("--nodejs-mode perf requires --perf-mode 'fp' or 'smart'")
+    if args.nodejs_mode in ("perf", "attach-maps") and args.perf_mode not in ("fp", "smart"):
+        parser.error("--nodejs-mode perf or attach-maps requires --perf-mode 'fp' or 'smart'")
 
     return args
 

gprofiler/metadata/application_metadata.py

@@ -4,16 +4,14 @@
 #
 
 import functools
-from subprocess import CompletedProcess
 from threading import Event, Lock
 from typing import Any, Dict, Optional
 
-from granulate_utils.linux.ns import get_process_nspid, run_in_ns
 from granulate_utils.linux.process import is_process_running, read_process_execfn
 from psutil import NoSuchProcess, Process
 
 from gprofiler.log import get_logger_adapter
-from gprofiler.utils import run_process
+from gprofiler.metadata.versions import get_exe_version
 
 logger = get_logger_adapter(__name__)
 
@@ -38,21 +36,7 @@ def _clear_cache(self) -> None:
                     del self._cache[process]
 
     def get_exe_version(self, process: Process, version_arg: str = "--version", try_stderr: bool = False) -> str:
-        """
-        Runs {process.exe()} --version in the appropriate namespace
-        """
-        exe_path = f"/proc/{get_process_nspid(process.pid)}/exe"
-
-        def _run_get_version() -> "CompletedProcess[bytes]":
-            return run_process([exe_path, version_arg], stop_event=self._stop_event, timeout=self._GET_VERSION_TIMEOUT)
-
-        cp = run_in_ns(["pid", "mnt"], _run_get_version, process.pid)
-        stdout = cp.stdout.decode().strip()
-        # return stderr if stdout is empty, some apps print their version to stderr.
-        if try_stderr and not stdout:
-            return cp.stderr.decode().strip()
-
-        return stdout
+        return get_exe_version(process, self._stop_event, self._GET_VERSION_TIMEOUT, version_arg, try_stderr)
 
     @functools.lru_cache(4096)
     def get_exe_version_cached(self, process: Process, version_arg: str = "--version", try_stderr: bool = False) -> str:

gprofiler/metadata/versions.py

@@ -0,0 +1,35 @@
+#
+# Copyright (c) Granulate. All rights reserved.
+# Licensed under the AGPL3 License. See LICENSE.md in the project root for license information.
+#
+from subprocess import CompletedProcess
+from threading import Event
+
+from granulate_utils.linux.ns import get_process_nspid, run_in_ns
+from psutil import Process
+
+from gprofiler.utils import run_process
+
+
+def get_exe_version(
+    process: Process,
+    stop_event: Event,
+    get_version_timeout: int,
+    version_arg: str = "--version",
+    try_stderr: bool = False,
+) -> str:
+    """
+    Runs {process.exe()} --version in the appropriate namespace
+    """
+    exe_path = f"/proc/{get_process_nspid(process.pid)}/exe"
+
+    def _run_get_version() -> "CompletedProcess[bytes]":
+        return run_process([exe_path, version_arg], stop_event=stop_event, timeout=get_version_timeout)
+
+    cp = run_in_ns(["pid", "mnt"], _run_get_version, process.pid)
+    stdout = cp.stdout.decode().strip()
+    # return stderr if stdout is empty, some apps print their version to stderr.
+    if try_stderr and not stdout:
+        return cp.stderr.decode().strip()
+
+    return stdout