Merge pull request #35 from amd/alex_dmesg_logs

Dmesg log collection

Author: alexandraBara

nodescraper/plugins/inband/dmesg/collector_args.py

@@ -28,4 +28,11 @@
 
 
 class DmesgCollectorArgs(CollectorArgs):
+    """Collector args
+
+    Args:
+        CollectorArgs (CollectorArgs): specific dmesg collector args
+    """
+
+    collect_rotated_logs: bool = False
     skip_sudo: bool = False

nodescraper/plugins/inband/dmesg/dmesg_collector.py

@@ -23,9 +23,11 @@
 # SOFTWARE.
 #
 ###############################################################################
+import re
 from typing import Optional
 
 from nodescraper.base import InBandDataCollector
+from nodescraper.connection.inband import TextFileArtifact
 from nodescraper.enums import EventCategory, EventPriority, ExecutionStatus, OSFamily
 from nodescraper.models import TaskResult
 
@@ -42,6 +44,113 @@ class DmesgCollector(InBandDataCollector[DmesgData, DmesgCollectorArgs]):
 
     DMESG_CMD = "dmesg --time-format iso -x"
 
+    DMESG_LOGS_CMD = (
+        r"ls -1 /var/log/dmesg* 2>/dev/null | grep -E '^/var/log/dmesg(\.[0-9]+(\.gz)?)?$' || true"
+    )
+
+    def _shell_quote(self, s: str) -> str:
+        """Single quote fix
+
+        Args:
+            s (str): path to be converted
+
+        Returns:
+            str: path to be returned
+        """
+        return "'" + s.replace("'", "'\"'\"'") + "'"
+
+    def _nice_dmesg_name(self, path: str) -> str:
+        """Map path to filename
+
+        Args:
+            path (str): file path
+
+        Returns:
+            str: new local filename
+        """
+        prefix = "rotated_"
+        base = path.rstrip("/").rsplit("/", 1)[-1]
+
+        if base == "dmesg":
+            return f"{prefix}dmesg_log.log"
+
+        m = re.fullmatch(r"dmesg\.(\d+)\.gz", base)
+        if m:
+            return f"{prefix}dmesg.{m.group(1)}.gz.log"
+
+        m = re.fullmatch(r"dmesg\.(\d+)", base)
+        if m:
+            return f"{prefix}dmesg.{m.group(1)}.log"
+
+        middle = base[:-3] if base.endswith(".gz") else base
+        return f"{prefix}{middle}.log"
+
+    def _collect_dmesg_rotations(self):
+        """Collect dmesg logs"""
+        list_res = self._run_sut_cmd(self.DMESG_LOGS_CMD, sudo=True)
+        paths = [p.strip() for p in (list_res.stdout or "").splitlines() if p.strip()]
+        if not paths:
+            self._log_event(
+                category=EventCategory.OS,
+                description="No /var/log/dmesg files found (including rotations).",
+                data={"list_exit_code": list_res.exit_code},
+                priority=EventPriority.WARNING,
+            )
+            return 0
+
+        collected_logs, failed_logs = [], []
+        for p in paths:
+            qp = self._shell_quote(p)
+            if p.endswith(".gz"):
+                cmd = f"gzip -dc {qp} 2>/dev/null || zcat {qp} 2>/dev/null"
+                res = self._run_sut_cmd(cmd, sudo=True, log_artifact=False)
+                if res.exit_code == 0 and res.stdout is not None:
+                    fname = self._nice_dmesg_name(p)
+                    self.logger.info("Collected dmesg log: %s", fname)
+                    self.result.artifacts.append(
+                        TextFileArtifact(filename=fname, contents=res.stdout)
+                    )
+                    collected_logs.append(
+                        {"path": p, "as": fname, "bytes": len(res.stdout.encode("utf-8", "ignore"))}
+                    )
+                else:
+                    failed_logs.append(
+                        {"path": p, "exit_code": res.exit_code, "stderr": res.stderr, "cmd": cmd}
+                    )
+            else:
+                cmd = f"cat {qp}"
+                res = self._run_sut_cmd(cmd, sudo=True, log_artifact=False)
+                if res.exit_code == 0 and res.stdout is not None:
+                    fname = self._nice_dmesg_name(p)
+                    self.logger.info("Collected dmesg log: %s", fname)
+                    self.result.artifacts.append(
+                        TextFileArtifact(filename=fname, contents=res.stdout)
+                    )
+                    collected_logs.append(
+                        {"path": p, "as": fname, "bytes": len(res.stdout.encode("utf-8", "ignore"))}
+                    )
+                else:
+                    failed_logs.append(
+                        {"path": p, "exit_code": res.exit_code, "stderr": res.stderr, "cmd": cmd}
+                    )
+
+        if collected_logs:
+            self._log_event(
+                category=EventCategory.OS,
+                description="Collected dmesg rotated files",
+                data={"collected": collected_logs},
+                priority=EventPriority.INFO,
+            )
+            self.result.message = self.result.message or "dmesg rotated files collected"
+
+        if failed_logs:
+            self._log_event(
+                category=EventCategory.OS,
+                description="Some dmesg files could not be collected.",
+                data={"failed": failed_logs},
+                priority=EventPriority.WARNING,
+            )
+
     def _get_dmesg_content(self) -> str:
         """run dmesg command on system and return output
 
@@ -79,6 +188,8 @@ def collect_data(
             return self.result, None
 
         dmesg_content = self._get_dmesg_content()
+        if args.collect_rotated_logs:
+            self._collect_dmesg_rotations()
 
         if dmesg_content:
             dmesg_data = DmesgData(dmesg_content=dmesg_content)

test/unit/plugin/test_dmesg_collector.py

@@ -23,6 +23,8 @@
 # SOFTWARE.
 #
 ###############################################################################
+import types
+
 import pytest
 
 from nodescraper.connection.inband.inband import CommandArtifact
@@ -141,3 +143,136 @@ def test_data_model():
     assert dmesg_data2.dmesg_content == (
         "2023-06-01T01:00:00,685236-05:00 test message1\n2023-06-01T02:30:00,685106-05:00 test message2"
     )
+
+
+class DummyRes:
+    def __init__(self, command="", stdout="", exit_code=0, stderr=""):
+        self.command = command
+        self.stdout = stdout
+        self.exit_code = exit_code
+        self.stderr = stderr
+
+
+def get_collector(monkeypatch, run_map, system_info, conn_mock):
+    c = DmesgCollector(
+        system_info=system_info,
+        system_interaction_level=SystemInteractionLevel.INTERACTIVE,
+        connection=conn_mock,
+    )
+    c.result = types.SimpleNamespace(artifacts=[], message=None)
+    c._events = []
+
+    def _log_event(**kw):
+        c._events.append(kw)
+
+    def _run_sut_cmd(cmd, *args, **kwargs):
+        return run_map(cmd, *args, **kwargs)
+
+    monkeypatch.setattr(c, "_log_event", _log_event, raising=True)
+    monkeypatch.setattr(c, "_run_sut_cmd", _run_sut_cmd, raising=True)
+    return c
+
+
+def test_collect_rotations_good_path(monkeypatch, system_info, conn_mock):
+    ls_out = (
+        "\n".join(
+            [
+                "/var/log/dmesg_log",
+                "/var/log/dmesg.1",
+                "/var/log/dmesg.2.gz",
+                "/var/log/dmesg.10.gz",
+            ]
+        )
+        + "\n"
+    )
+
+    def run_map(cmd, **kwargs):
+        if cmd.startswith("ls -1 /var/log/dmesg"):
+            return DummyRes(command=cmd, stdout=ls_out, exit_code=0)
+        if cmd.startswith("cat '"):
+            if "/var/log/dmesg.1'" in cmd:
+                return DummyRes(command=cmd, stdout="dmesg.1 content\n", exit_code=0)
+            if "/var/log/dmesg_log'" in cmd:
+                return DummyRes(command=cmd, stdout="dmesg content\n", exit_code=0)
+        if "gzip -dc" in cmd and "/var/log/dmesg.2.gz" in cmd:
+            return DummyRes(command=cmd, stdout="gz2 content\n", exit_code=0)
+        if "gzip -dc" in cmd and "/var/log/dmesg.10.gz" in cmd:
+            return DummyRes(command=cmd, stdout="gz10 content\n", exit_code=0)
+        return DummyRes(command=cmd, stdout="", exit_code=1, stderr="unexpected")
+
+    c = get_collector(monkeypatch, run_map, system_info, conn_mock)
+
+    c._collect_dmesg_rotations()
+
+    names = {a.filename for a in c.result.artifacts}
+    assert names == {
+        "rotated_dmesg_log.log",
+        "rotated_dmesg.1.log",
+        "rotated_dmesg.2.gz.log",
+        "rotated_dmesg.10.gz.log",
+    }
+
+    descs = [e["description"] for e in c._events]
+    assert "Collected dmesg rotated files" in descs
+
+
+def test_collect_rotations_no_files(monkeypatch, system_info, conn_mock):
+    def run_map(cmd, **kwargs):
+        if cmd.startswith("ls -1 /var/log/dmesg"):
+            return DummyRes(command=cmd, stdout="", exit_code=0)
+        return DummyRes(command=cmd, stdout="", exit_code=1)
+
+    c = get_collector(monkeypatch, run_map, system_info, conn_mock)
+
+    c._collect_dmesg_rotations()
+
+    assert c.result.artifacts == []
+
+    events = c._events
+    assert any(
+        e["description"].startswith("No /var/log/dmesg files found")
+        and e["priority"].name == "WARNING"
+        for e in events
+    )
+
+
+def test_collect_rotations_gz_failure(monkeypatch, system_info, conn_mock):
+    ls_out = "/var/log/dmesg.2.gz\n"
+
+    def run_map(cmd, **kwargs):
+        if cmd.startswith("ls -1 /var/log/dmesg"):
+            return DummyRes(command=cmd, stdout=ls_out, exit_code=0)
+        if "gzip -dc" in cmd and "/var/log/dmesg.2.gz" in cmd:
+            return DummyRes(command=cmd, stdout="", exit_code=1, stderr="gzip: not found")
+        return DummyRes(command=cmd, stdout="", exit_code=1)
+
+    c = get_collector(monkeypatch, run_map, system_info, conn_mock)
+
+    c._collect_dmesg_rotations()
+
+    assert c.result.artifacts == []
+
+    fail_events = [
+        e for e in c._events if e["description"] == "Some dmesg files could not be collected."
+    ]
+    assert fail_events, "Expected a failure event"
+    failed = fail_events[-1]["data"]["failed"]
+    assert any(item["path"].endswith("/var/log/dmesg.2.gz") for item in failed)
+
+
+def test_collect_data_integration(monkeypatch, system_info, conn_mock):
+    def run_map(cmd, **kwargs):
+        if cmd == DmesgCollector.DMESG_CMD:
+            return DummyRes(command=cmd, stdout="DMESG OUTPUT\n", exit_code=0)
+        if cmd.startswith("ls -1 /var/log/dmesg"):
+            return DummyRes(command=cmd, stdout="/var/log/dmesg\n", exit_code=0)
+        if cmd.startswith("cat '") and "/var/log/dmesg'" in cmd:
+            return DummyRes(command=cmd, stdout="dmesg file content\n", exit_code=0)
+        return DummyRes(command=cmd, stdout="", exit_code=1)
+
+    c = get_collector(monkeypatch, run_map, system_info, conn_mock)
+
+    result, data = c.collect_data()
+
+    assert isinstance(data, DmesgData)
+    assert data.dmesg_content == "DMESG OUTPUT\n"