HPCC-30606 Fix "significant skew in records warning"

Signed-off-by: Shamser Ahmed <[email protected]>
hpcc-systems · Oct 26, 2023 · edcce6d · edcce6d
1 parent 8b3ec0a
commit edcce6d
Showing 1 changed file with 34 additions and 4 deletions.
diff --git a/common/wuanalysis/anarule.cpp b/common/wuanalysis/anarule.cpp
@@ -135,14 +135,44 @@ class IoSkewRule : public AActivityRule
             stat_type timeAvgLocalExecute = activity.getStatRaw(StTimeLocalExecute, StAvgX);
 
             stat_type cost;
-            //If one node didn't spill then it is possible the skew caused all the lost time
             unsigned actkind = activity.getAttr(WaKind);
-            if ((actkind==TAKspillread||actkind==TAKspillwrite) && activity.getStatRaw(stat, StMinX) == 0)
+            if ((actkind==TAKspillread||actkind==TAKspillwrite) && (activity.getStatRaw(stat, StMinX) == 0))
+            {
+                //If one node didn't spill then it is possible the skew caused all the lost time
                 cost = timeMaxLocalExecute;
+                result.set(ANA_IOSKEW_RECORDS_ID, cost, "Uneven worker spilling is causing uneven %s time", category);
+            }
             else
+            {
+                bool fileSkew = false;
+                if (stat == StTimeDiskWriteIO)
+                {
+                    IWuEdge *inputEdge;
+                    unsigned i=0;
+                    while (!fileSkew && (inputEdge=activity.queryInput(i++)))
+                    {
+                        //output IO is likely to be skewed because # rows written by the workers are skewed
+                        if (inputEdge->getStatRaw(StNumRowsProcessed, StSkewMax)>options.queryOption(watOptSkewThreshold))
+                            fileSkew = true;
+                    }
+                }
+                if (stat == StTimeDiskReadIO)
+                {
+                    IWuEdge *outputEdge;
+                    unsigned i=0;
+                    while (!fileSkew && (outputEdge=activity.queryOutput(i++)))
+                    {
+                        //input IO is likely to be skewed because # rows read by the workers are skewed
+                        if (outputEdge->getStatRaw(StNumRowsProcessed, StSkewMax)>options.queryOption(watOptSkewThreshold))
+                            fileSkew = true;
+                    }
+                }
                 cost = (timeMaxLocalExecute - timeAvgLocalExecute);
-
-            result.set(ANA_IOSKEW_RECORDS_ID, cost, "Significant skew in records causes uneven %s time", category);
+                if (fileSkew)
+                    result.set(ANA_IOSKEW_RECORDS_ID, cost, "Significant skew in datafile partitioning is causing uneven %s time", category);
+                else
+                    result.set(ANA_IOSKEW_RECORDS_ID, cost, "Significant skew in IO performance may be causing uneven %s time", category);
+            }
             updateInformation(result, activity);
             return true;
         }