Better java app id resolving (#311)

* Better java app id resolving

* CR fix

Author: liorgorb

gprofiler/merge.py

@@ -13,7 +13,7 @@
 from gprofiler.containers_client import ContainerNamesClient
 from gprofiler.gprofiler_types import ProcessToStackSampleCounters, StackToSampleCount
 from gprofiler.log import get_logger_adapter
-from gprofiler.metadata.application_identifiers import get_application_name
+from gprofiler.metadata import application_identifiers
 from gprofiler.metadata.metadata_type import Metadata
 from gprofiler.system_metrics import Metrics
 
@@ -256,7 +256,7 @@ def concatenate_profiles(
 
     for pid, stacks in process_profiles.items():
         container_name = _get_container_name(pid, container_names_client, add_container_names)
-        application_name = get_application_name(pid) if identify_applications else None
+        application_name = application_identifiers.get_application_name(pid) if identify_applications else None
         if application_name is not None:
             application_name = f"appid: {application_name}"
         prefix = (container_name + ";") if add_container_names else ""

gprofiler/metadata/application_identifiers.py

@@ -11,7 +11,10 @@
 from granulate_utils.linux.ns import resolve_host_path, resolve_proc_root_links
 from psutil import NoSuchProcess, Process
 
+from gprofiler.exceptions import CalledProcessError
 from gprofiler.log import get_logger_adapter
+from gprofiler.profilers.java import jattach_path
+from gprofiler.utils import run_process
 
 _logger = get_logger_adapter(__name__)
 
@@ -244,11 +247,16 @@ def get_application_name(self, process: Process) -> Optional[str]:
         if not any("libjvm.so" in m.path for m in process.memory_maps()):
             return None
 
-        jar_arg = _get_cli_arg_by_name(process.cmdline(), "-jar")
-        if jar_arg is _NON_AVAILABLE_ARG:
-            return None
+        try:
+            java_properties = run_process([jattach_path(), str(process.pid), "properties"]).stdout.decode()
+            for line in java_properties.splitlines():
+                if line.startswith("sun.java.command"):
+                    app_id = line[line.find("=") + 1 :]
+                    return f"java: {app_id} ({_append_file_to_proc_wd(process, app_id)})"
+        except CalledProcessError as e:
+            _logger.warning(f"Couldn't get Java properties for process {process.pid}: {e.stderr}")
 
-        return f"java: {jar_arg} ({_append_file_to_proc_wd(process, jar_arg)})"
+        return None
 
 
 # Please note that the order matter, because the FIRST matching identifier will be used.

gprofiler/profilers/java.py

@@ -35,11 +35,11 @@
 from packaging.version import Version
 from psutil import Process
 
+from gprofiler import merge
 from gprofiler.exceptions import CalledProcessError
 from gprofiler.gprofiler_types import ProcessToStackSampleCounters, StackToSampleCount
 from gprofiler.kernel_messages import get_kernel_messages_provider
 from gprofiler.log import get_logger_adapter
-from gprofiler.merge import parse_one_collapsed
 from gprofiler.profilers.profiler_base import ProcessProfilerBase
 from gprofiler.profilers.registry import ProfilerArgument, register_profiler
 from gprofiler.utils import (
@@ -795,7 +795,7 @@ def _profile_ap_process(self, ap_proc: AsyncProfiledProcess, comm: str) -> Stack
             return self._profiling_error_stack("error", "process exited before reading the output", comm)
         else:
             logger.info(f"Finished profiling process {ap_proc.process.pid}")
-            return parse_one_collapsed(output, comm)
+            return merge.parse_one_collapsed(output, comm)
 
     def _check_hotspot_error(self, ap_proc: AsyncProfiledProcess) -> None:
         pid = ap_proc.process.pid

gprofiler/profilers/perf.py

@@ -9,10 +9,10 @@
 from threading import Event
 from typing import List, Optional
 
+from gprofiler import merge
 from gprofiler.exceptions import StopEventSetException
 from gprofiler.gprofiler_types import ProcessToStackSampleCounters
 from gprofiler.log import get_logger_adapter
-from gprofiler.merge import merge_global_perfs
 from gprofiler.profilers.profiler_base import ProfilerBase
 from gprofiler.profilers.registry import ProfilerArgument, register_profiler
 from gprofiler.utils import run_process, start_process, wait_event, wait_for_file_by_prefix
@@ -202,7 +202,7 @@ def snapshot(self) -> ProcessToStackSampleCounters:
         for perf in self._perfs:
             perf.switch_output()
 
-        return merge_global_perfs(
+        return merge.merge_global_perfs(
             self._perf_fp.wait_and_script() if self._perf_fp is not None else None,
             self._perf_dwarf.wait_and_script() if self._perf_dwarf is not None else None,
         )

gprofiler/profilers/python.py

@@ -17,6 +17,7 @@
 from granulate_utils.python import _BLACKLISTED_PYTHON_PROCS, DETECTED_PYTHON_PROCESSES_REGEX
 from psutil import NoSuchProcess, Process
 
+from gprofiler import merge
 from gprofiler.exceptions import (
     CalledProcessError,
     CalledProcessTimeoutError,
@@ -25,7 +26,6 @@
 )
 from gprofiler.gprofiler_types import ProcessToStackSampleCounters, StackToSampleCount, nonnegative_integer
 from gprofiler.log import get_logger_adapter
-from gprofiler.merge import parse_many_collapsed, parse_one_collapsed_file
 from gprofiler.metadata.py_module_version import get_modules_versions
 from gprofiler.metadata.system_metadata import get_arch
 from gprofiler.profilers.profiler_base import ProcessProfilerBase, ProfilerBase, ProfilerInterface
@@ -147,7 +147,7 @@ def _profile_process(self, process: Process) -> StackToSampleCount:
                 raise
 
             logger.info(f"Finished profiling process {process.pid} with py-spy")
-            parsed = parse_one_collapsed_file(Path(local_output_path), comm)
+            parsed = merge.parse_one_collapsed_file(Path(local_output_path), comm)
             if self.add_versions:
                 parsed = _add_versions_to_process_stacks(process, parsed)
             return parsed
@@ -368,7 +368,7 @@ def snapshot(self) -> ProcessToStackSampleCounters:
         finally:
             # always remove, even if we get read/decode errors
             collapsed_path.unlink()
-        parsed = parse_many_collapsed(collapsed_text)
+        parsed = merge.parse_many_collapsed(collapsed_text)
         if self.add_versions:
             parsed = _add_versions_to_stacks(parsed)
         return parsed

gprofiler/profilers/ruby.py

@@ -10,10 +10,10 @@
 
 from psutil import Process
 
+from gprofiler import merge
 from gprofiler.exceptions import ProcessStoppedException, StopEventSetException
 from gprofiler.gprofiler_types import StackToSampleCount
 from gprofiler.log import get_logger_adapter
-from gprofiler.merge import parse_one_collapsed_file
 from gprofiler.profilers.profiler_base import ProcessProfilerBase
 from gprofiler.profilers.registry import register_profiler
 from gprofiler.utils import pgrep_maps, random_prefix, removed_path, resource_path, run_process
@@ -75,7 +75,7 @@ def _profile_process(self, process: Process) -> StackToSampleCount:
                 raise StopEventSetException
 
             logger.info(f"Finished profiling process {process.pid} with rbspy")
-            return parse_one_collapsed_file(Path(local_output_path), process_comm(process))
+            return merge.parse_one_collapsed_file(Path(local_output_path), process_comm(process))
 
     def _select_processes_to_profile(self) -> List[Process]:
         return pgrep_maps(r"(?:^.+/ruby[^/]*$)")

lint.sh

@@ -11,6 +11,6 @@ if [[ "$1" = "--ci" ]]; then
 fi
 
 isort --settings-path .isort.cfg --skip granulate-utils .
-black --line-length 120 $check_arg --exclude granulate-utils .
+black --line-length 120 $check_arg --exclude "granulate-utils|\.venv" .
 flake8 --config .flake8 .
 mypy .