summary-warnings for normalize step; extend summary warnings for errors (#57)

* summary-warnings for normalize step; extend summary warnings for errors

Signed-off-by: Lars Schneidenbach <schneidenbach@us.ibm.com>

Author: lasch

src/aiu_trace_analyzer/pipeline/normalize.py

@@ -5,9 +5,9 @@
 import re
 
 import aiu_trace_analyzer.logger as aiulog
+from aiu_trace_analyzer.types import TraceEvent, GlobalIngestData, TraceWarning
 from aiu_trace_analyzer.pipeline.context import AbstractContext
 from aiu_trace_analyzer.pipeline.hashqueue import AbstractHashQueueContext
-from aiu_trace_analyzer.types import TraceEvent, GlobalIngestData
 from aiu_trace_analyzer.pipeline.tools import FlexEventMapToTS
 
 
@@ -59,7 +59,20 @@ class NormalizationContext(AbstractHashQueueContext):
 
     def __init__(self, soc_frequency: float, ignore_crit: bool = False,
                  filterstr: str = "") -> None:
-        super().__init__()
+        super().__init__(warnings=[
+            TraceWarning(
+                name="long_dur",
+                text="OVC: Detected {d[count]} event(s) with long duration and"
+                     " thus potential undetected overflow in TSx counter.",
+                data={"count": 0}
+            ),
+            TraceWarning(
+                name="ts_seq_err",
+                text="OVC: local_correction fix has missed a spot in TS-sequence of {d[count]} event(s).",
+                data={"count": 0},
+                is_error=True
+            )
+        ])
         self.soc_frequency = soc_frequency
         self.frequency_minmax = (1e99, 0.0, 0, 0.0, 0.0)
         self.OVERFLOW_TIME_SPAN_US = float(1 << 32) / self.soc_frequency
@@ -195,9 +208,7 @@ def tsx_32bit_local_correction(self, event: TraceEvent) -> dict:
                 prev = curr
 
             if event["dur"] > self.OVERFLOW_TIME_SPAN_US:
-                aiulog.log(aiulog.WARN,
-                           "OVC: Detected event with long duration and"
-                           " thus potential undetected overflow in TSx counter.")
+                self.warnings["long_dur"].update()
 
             if "Cmpt Exec" not in event["name"]:
                 return args
@@ -249,7 +260,7 @@ def tsx_32bit_global_correction(self, qid, event: TraceEvent) -> dict:
                 curr = int(args[ts], 0)
                 curr += (ovc * 1 << 32)
                 if curr < prev:
-                    aiulog.log(aiulog.ERROR, "attempt of local_correction fix has missed a spot in TS-sequence.")
+                    self.warnings["ts_seq_err"].update()
                     if not self.ignore_crit:
                         assert curr >= prev, "local_correction of TS-sequence incomplete."
                 args[ts] = str(curr)

src/aiu_trace_analyzer/types.py

@@ -180,24 +180,28 @@ def get_dialect(cls, jobhash: int) -> InputDialect:
 class TraceWarning:
     """
     Keep track of warnings to allow accumulated warning at the end of a run
-    Usage:
+    Example:
 
         w = TraceWarning(
             name="MyWarning",
             text="This stage has detected {d[count]} issues with max {d[max]}.",
             data={"count": 0, "max": 0.0},
             update_fn={"count": int.__add__, "max": max},
-            autolog=True
+            autolog=True,
+            is_error=True,
         )
 
         name:      a key that can be used to manage multiple warnings in e.g. a dictionary
         text:      the warning text with variables (always us 'd' as the dictionary name)
         data:      dictionary with entries that match the text variables
         update_fn: functions to run when the update function is called with data
-        autolog:   automatically print the warning at destruction time
+        autolog:   automatically print the warning at destruction time (default)
+        is_error:  print the summary warning as ERROR level and not as WARN (default)
 
         Whenever a warning should be added:
+
             w.update({"count": 1, "max": 100.0})
+
         this calls the preset update_fn for each item to update the values
         if update_fn is e.g. int.__add__, then the new_val entry for count will be increased by 1
 
@@ -213,14 +217,16 @@ def __init__(
             text: str,
             data: dict[str, any],
             update_fn: dict[str, callable] = {},
-            auto_log: bool = True):
+            auto_log: bool = True,
+            is_error: bool = False):
         self.occurred = False
         self.name = name
         # format-string with {d[key]} placeholders
         self.text: str = text
         self.args_list: dict[str, any] = {k: v for k, v in data.items()}
         self.update_fn: dict[str, callable] = {k: v for k, v in update_fn.items()}
         self.auto_log = auto_log
+        self.warn_level = aiulog.WARN if not is_error else aiulog.ERROR
 
         text_keys = re.findall(r"{d\[([.\w]+)\]}", self.text)
         if len(text_keys) != len(self.args_list):
@@ -249,7 +255,7 @@ def __init__(
 
     def __del__(self) -> None:
         if self.auto_log is True and self.has_warning():
-            aiulog.log(aiulog.WARN, self)
+            aiulog.log(self.warn_level, self)
 
     def get_name(self) -> str:
         return self.name