Fix NIXL bench output parsing in case of noisy output

Author: amaslenn

src/cloudai/workloads/common/nixl.py

@@ -127,6 +127,13 @@ def extract_nixlbench_data(stdout_file: Path) -> pd.DataFrame:
             continue
         parts = line.split()
         if header_present and (len(parts) == 6 or len(parts) == 10):
+            try:
+                int(parts[0])  # block size
+                int(parts[1])  # batch size
+            except ValueError:
+                # doesn't look like a data line, skip
+                continue
+
             if len(parts) == 6:
                 data.append([parts[0], parts[1], parts[2], parts[-1]])
             else:

tests/report_generation_strategy/test_nixl_bench_report.py

@@ -75,6 +75,41 @@ def test_nixl_bench_report_parsing(tmp_path: Path, sample: str, exp_latency: lis
     assert df["bw_gb_sec"].tolist() == exp_bw
 
 
+def test_nixlbench_report_parsing__noisy_output(tmp_path: Path):
+    sample = """
+----------------------------------------------------------------------------------------------------------------------------------------------------------------
+
+Block Size (B)      Batch Size     B/W (GB/Sec)   Avg Lat. (us)  Avg Prep (us)  P99 Prep (us)  Avg Post (us)  P99 Post (us)  Avg Tx (us)    P99 Tx (us)
+----------------------------------------------------------------------------------------------------------------------------------------------------------------
+[1757492049.543051] [ptyche0321:627036:0]   +--------------------------------+---------------------------------------------------------+
+[1757492049.543052] [ptyche0321:627036:0]   | ucp_context_0 intra-node cfg#2 | active message by ucp_am_send*(egr) from host memory    |
+[1757492049.543053] [ptyche0321:627036:0]   +--------------------------------+-------------------------------------+-------------------+
+[1757492049.543053] [ptyche0321:627036:0]   |                        0..8184 | short                               | tcp/enP6p3s0f1np1 |
+[1757492049.543053] [ptyche0321:627036:0]   |                      8185..inf | multi-frag copy-in                  | tcp/enP6p3s0f1np1 |
+[1757492049.543054] [ptyche0321:627036:0]   +--------------------------------+-------------------------------------+-------------------+
+4096                1              0.198413       20.6           13.0           13.0           10.6           16.0           10.1           14.0
+8192                1              0.381410       21.5           9.0            9.0            10.9           17.0           10.6           15.0
+16384               1              0.786957       20.8           7.0            7.0            10.5           15.0           10.3           16.0
+32768               1              1.580617       20.7           8.0            8.0            10.6           13.0           10.1           15.0
+65536               1              3.151882       20.8           6.0            6.0            10.5           16.0           10.3           15.0
+131072              1              6.133162       21.4           6.0            6.0            10.9           14.0           10.5           16.0
+262144              1              12.458916      21.0           12.0           12.0           10.6           16.0           10.4           15.0
+524288              1              24.594299      21.3           7.0            7.0            10.8           15.0           10.5           14.0
+1048576             1              49.639065      21.1           6.0            6.0            10.6           16.0           10.5           13.0
+2097152             1              90.493545      23.2           6.0            6.0            12.0           40.0           11.0           13.0
+4194304             1              184.397175     22.7           6.0            6.0            11.5           38.0           11.1           19.0
+8388608             1              367.256639     22.8           6.0            6.0            10.8           39.0           11.9           22.0
+16777216            1              681.912650     24.6           5.0            5.0            11.4           40.0           13.1           17.0
+33554432            1              1205.204798    27.8           6.0            6.0            11.3           40.0           16.3           18.0
+67108864            1              1636.801561    41.0           5.0            5.0            11.7           36.0           29.2           38.0
+[1757492049.615857] [ptyche0321:627915:0]   |                      8177..inf | multi-frag copy-in                                    | tcp/enP6p3s0f1np1 |
+[1757492049.615862] [ptyche0321:627915:0]   +--------------------------------+-------------------------------------------------------+-------------------+
+"""
+    (tmp_path / "stdout.txt").write_text(sample)
+    df = extract_nixlbench_data(tmp_path / "stdout.txt")
+    assert df.shape == (15, 4)
+
+
 class TestWasRunSuccessful:
     def test_no_file(self, nixl_tr: TestRun):
         assert not nixl_tr.test.test_definition.was_run_successful(nixl_tr).is_successful