[xcl-lens] Fix filters

projects/xcl-lens/src/xcl_lens/parser/rccl.py

@@ -70,7 +70,8 @@ class RcclLogParser:
 
         self._report_sys()
         self._report_user_envs()
-        self._report_gdr_info()
+        self._report_net_ib_info()
+        self._report_gdr_rw_info()
         self._report_graph_info()
         self._report_channel_transport_info()
         self._report_collective_transfers()
@@ -83,8 +84,8 @@ class RcclLogParser:
         # Preferred format:
         #   <host>:<pid>:<tid> [rank] NCCL INFO/WARN/ERROR <content>
         # where <host> itself does NOT contain ':' (so we always stop at first colon)
-        match = re.search(
-            r"^([^:\s]+):\d+:\d+\s+\[(\d+)\]\s+NCCL\s+(?:INFO|WARN|ERROR)\s+(.*)",
+        match = re.match(
+            r"([^:\s]+):\d+:\d+\s+\[(\d+)\]\s+NCCL\s+(?:INFO|WARN|ERROR)\s+(.*)",
             line,
         )
         if match:
@@ -136,11 +137,10 @@ class RcclLogParser:
                 )
         print()
 
-    def _report_gdr_info(self):
-        """Parse and print GPU Direct RDMA (GDR) related information."""
-        print("===> GDR Info:\n")
+    def _report_net_ib_info(self):
+        """Parse and print NET/IB GPU Direct RDMA HCA information."""
+        print("===> NET/IB Info:\n")
 
-        # Part 1: NET/IB : GPU Direct RDMA Enabled for HCA <hca_no> '<hca_id>'
         ib_rows = []
         pattern_ib = re.compile(r"NET/IB\s+:\s+GPU Direct RDMA Enabled for HCA\s+(\d+)\s+'([^']+)'")
         for (host, rank, content), _ in self.log_entries.items():
@@ -157,7 +157,6 @@ class RcclLogParser:
                     }
                 )
 
-        print("  NET/IB : GPU Direct RDMA Enabled for:\n")
         if ib_rows:
             df_ib = pd.DataFrame(ib_rows)
             df_ib.drop_duplicates(inplace=True)
@@ -168,7 +167,10 @@ class RcclLogParser:
         else:
             print("  (No data found)\n")
 
-        # Part 2: GPU Direct RDMA Enabled for GPU <gpu> / HCA <hca_no> (distance <expr>), read <0|1>
+    def _report_gdr_rw_info(self):
+        """Parse and print GPU Direct RDMA read/write information."""
+        print("===> GDR R/W Info:\n")
+
         gpu_rows = []
         pattern_gpu = re.compile(
             r"GPU Direct RDMA Enabled for GPU\s+(\S+)\s*/\s*"
@@ -198,7 +200,6 @@ class RcclLogParser:
                     }
                 )
 
-        print("  GPU Direct RDMA Enabled for GPU:\n")
         if gpu_rows:
             df_gpu = pd.DataFrame(gpu_rows)
             df_gpu.drop_duplicates(inplace=True)
@@ -390,7 +391,7 @@ class RcclLogParser:
             return
 
         df = pd.DataFrame(data)
-        df.sort_values(by=["host", "rank", "channel", "sender", "receiver"], inplace=True)
         df.drop_duplicates(inplace=True)
+        df.sort_values(by=["host", "rank", "channel", "sender", "receiver"], inplace=True)
         print(df.to_string(index=False))
         print()

projects/xcl-lens/src/xcl_lens/runner.py

@@ -40,7 +40,13 @@ def run_with_input(
 
     # Case 3: Execute as command
     return _execute_command(
-        summary=summary, verbose=verbose, rank=rank, log_prefix=log_prefix, cmd=command
+        summary=summary,
+        verbose=verbose,
+        hosts=hosts,
+        ranks=ranks,
+        rank=rank,
+        log_prefix=log_prefix,
+        cmd=command,
     )
 
 
@@ -84,14 +90,23 @@ def _process_files(
     return 0
 
 
-def _execute_command(*, summary: bool, verbose: bool, rank: int, log_prefix: str, cmd: list[str]):
+def _execute_command(
+    *,
+    summary: bool,
+    verbose: bool,
+    hosts: list[str] | None,
+    ranks: list[str] | None,
+    rank: int,
+    log_prefix: str,
+    cmd: list[str],
+):
     env = os.environ.copy()
     env["NCCL_DEBUG"] = "INFO"
     env["NCCL_DEBUG_SUBSYS"] = "ALL"
 
     print(f"{log_prefix} [Wrapper] Running command: {' '.join(cmd)}")
 
-    parser = RcclLogParser()
+    parser = RcclLogParser(verbose=verbose, hosts=hosts, ranks=ranks)
     process = subprocess.Popen(
         cmd,
         env=env,
@@ -110,6 +125,6 @@ def _execute_command(*, summary: bool, verbose: bool, rank: int, log_prefix: str
     process.wait()
 
     if rank == 0:
-        parser.report(verbose=verbose)
+        parser.report()
 
     return process.returncode