HPCC-30606 Changes following review

Signed-off-by: Shamser Ahmed <shamser.ahmed@lexisnexis.com>
hpcc-systems · Oct 31, 2023 · 20ea0dc · 20ea0dc
1 parent edcce6d
commit 20ea0dc
Showing 1 changed file with 22 additions and 23 deletions.
diff --git a/common/wuanalysis/anarule.cpp b/common/wuanalysis/anarule.cpp
@@ -129,13 +129,13 @@ class IoSkewRule : public AActivityRule
     {
         stat_type ioAvg = activity.getStatRaw(stat, StAvgX);
         stat_type ioMaxSkew = activity.getStatRaw(stat, StSkewMax);
+        unsigned actkind = activity.getAttr(WaKind);
         if (ioMaxSkew > options.queryOption(watOptSkewThreshold))
         {
             stat_type timeMaxLocalExecute = activity.getStatRaw(StTimeLocalExecute, StMaxX);
             stat_type timeAvgLocalExecute = activity.getStatRaw(StTimeLocalExecute, StAvgX);
 
             stat_type cost;
-            unsigned actkind = activity.getAttr(WaKind);
             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
@@ -144,34 +144,33 @@ class IoSkewRule : public AActivityRule
             }
             else
             {
-                bool fileSkew = false;
-                if (stat == StTimeDiskWriteIO)
+                bool sizeSkew = false;
+                bool numRowsSkew = false;
+                IWuEdge *wuEdge = nullptr;
+                if ((stat==StTimeDiskWriteIO) || (actkind==TAKspillwrite))
                 {
-                    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 (activity.getStatRaw(StSizeDiskWrite)>options.queryOption(watOptSkewThreshold))
+                        sizeSkew = true;
+                    IWuEdge *wuEdge = activity.queryInput(0);
                 }
-                if (stat == StTimeDiskReadIO)
+                else if ((stat == StTimeDiskReadIO) || (actkind==TAKspillread))
                 {
-                    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;
-                    }
+                    if (activity.getStatRaw(StSizeDiskRead)>options.queryOption(watOptSkewThreshold))
+                        sizeSkew = true;
+                    IWuEdge *wuEdge = activity.queryOutput(0);
                 }
+                if (wuEdge && wuEdge->getStatRaw(StNumRowsProcessed, StSkewMax)>options.queryOption(watOptSkewThreshold))
+                    numRowsSkew = true;
                 cost = (timeMaxLocalExecute - timeAvgLocalExecute);
-                if (fileSkew)
-                    result.set(ANA_IOSKEW_RECORDS_ID, cost, "Significant skew in datafile partitioning is causing uneven %s time", category);
+                if (sizeSkew)
+                {
+                    if (numRowsSkew)
+                        result.set(ANA_IOSKEW_RECORDS_ID, cost, "Significant skew in number of records is causing uneven %s time", category);
+                    else
+                        result.set(ANA_IOSKEW_RECORDS_ID, cost, "Significant skew in record sizes 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);
+                    result.set(ANA_IOSKEW_RECORDS_ID, cost, "Significant skew in IO performance is causing uneven %s time", category);
             }
             updateInformation(result, activity);
             return true;