Merge pull request github#12298 from MathiasVP/join-order-metric-query-with-more-rows

QL: Output more rows in the join order query

Author: tausbn

ql/ql/src/codeql_ql/StructuredLogs.qll

@@ -273,6 +273,8 @@ module KindPredicatesLog {
       then result = this.getPredicateName()
       else result = "<Summary event>"
     }
+
+    Array getRA(string ordering) { result = this.getObject("ra").getArray(ordering) }
   }
 
   class SentinelEmpty extends SummaryEvent {
@@ -288,8 +290,12 @@ module KindPredicatesLog {
 
     string getRAReference() { result = this.getString("raReference") }
 
+    Array getCounts() { result = this.getArray("counts") }
+
     float getCount(int i) { result = getRanked(this.getArray("counts"), i) }
 
+    Array getDuplicationPercentage() { result = this.getArray("duplicationPercentages") }
+
     float getDuplicationPercentage(int i) {
       result = getRanked(this.getArray("duplicationPercentages"), i)
     }
@@ -327,6 +333,12 @@ module KindPredicatesLog {
     string getPosition() { result = this.getString("position") }
 
     Predicate getPredicate() { result = getPredicateFromPosition(this.getPosition()) }
+
+    /**
+     * Gets the RA for this event. Unlike recursive predicates, a COMPUTE_SIMPLE
+     * event only has one pipeline ordering (and it's named "pipeline").
+     */
+    Array getRA() { result = this.getObject("ra").getArray("pipeline") }
   }
 
   /** Gets the `index`'th event that's evaluated by `recursive`. */

ql/ql/src/queries/performance/LargeJoinOrder.ql

@@ -16,14 +16,37 @@ import KindPredicatesLog
 float getBadness(ComputeSimple simple) {
   exists(float maxTupleCount, float resultSize, float largestDependency, float denom |
     resultSize = simple.getResultSize() and
-    maxTupleCount = max(simple.getPipelineRun().getCount(_)) and
+    extractInformation(simple, maxTupleCount, _, _, _) and
     largestDependency = max(simple.getDependencies().getADependency().getResultSize()) and
     denom = resultSize.maximum(largestDependency) and
     denom > 0 and // avoid division by zero (which would create a NaN result).
     result = maxTupleCount / denom
   )
 }
 
-from ComputeSimple evt, float f
-where f = getBadness(evt) and f > 1.5
-select evt, f order by f desc
+pragma[nomagic]
+predicate hasPipelineRun(ComputeSimple simple, PipeLineRun run) { run = simple.getPipelineRun() }
+
+predicate extractInformation(
+  ComputeSimple simple, float maxTupleCount, Array tuples, Array duplicationPercentages, Array ra
+) {
+  exists(PipeLineRun run |
+    hasPipelineRun(simple, run) and
+    maxTupleCount = max(run.getCount(_)) and
+    tuples = run.getCounts() and
+    duplicationPercentages = run.getDuplicationPercentage() and
+    ra = simple.getRA()
+  )
+}
+
+from
+  ComputeSimple evt, float badness, float maxTupleCount, Array tuples, Array duplicationPercentages,
+  Array ra, int index
+where
+  badness = getBadness(evt) and
+  badness > 1.5 and
+  extractInformation(evt, maxTupleCount, tuples, duplicationPercentages, ra)
+select evt.getPredicateName() as predicate_name, badness, index,
+  tuples.getFloat(index) as tuple_count,
+  duplicationPercentages.getFloat(index) as duplication_percentage, ra.getString(index) order by
+    badness desc