-
Type:
Task
-
Resolution: Fixed
-
Priority:
Major - P3
-
Affects Version/s: None
-
Component/s: None
-
None
-
Query Execution
-
Fully Compatible
-
35
Given a simple timeseries collection:
./build/install/bin/mongod --setParameter featureFlagSbeFull=true > db.coll.drop() > db.createCollection("coll", {timeseries: {timeField: "t"}}) > db.coll.insert({t: new Date(), a: 4, b: 7, c: 5}) > db.coll.insert({t: new Date(), a: 2, b: 9, c: 8}) > db.coll.insert({t: new Date(), a: 3, b: 1, c: 6})
the goals of this task are to make the following query work in block-processing mode and to write a jstest that verifies it works:
db.coll.aggregate([{$project: {a: 1}}, {$group: {_id: null, x: {$avg : "$a"}}}], {allowDiskUse: false})
Note that block processing is not yet supported with allowDiskUse: true, which has been the default since 6.0. This is being addressed in SERVER-84781.
Before this task the SBE plan for this query is
"slotBasedPlan" : { "slots" : "$$RESULT=s14 env: { s5 = Nothing (nothing) }", "stages" : " [3] project [s14 = newBsonObj(\"_id\", s6, \"x\", s13)] [3] project [s13 = if (s12 == 0ll) then null else (doubleDoubleSumFinalize(s11) / s12) ] [3] group [s6] [s11 = aggDoubleDoubleSum(s8), s12 = sum( if ((typeMatch(s8, 1088) ?: true) || !(isNumber(s8))) then 0ll else 1ll )] spillSlots[s9, s10] mergingExprs[aggMergeDoubleDoubleSums(s9), sum(s10)] [3] block_to_row blocks[s3, s4] row[s7, s8] [3] project [s6 = null] [2] ts_bucket_to_cellblock s1 pathReqs[s3 = ProjectPath(Get(_id)/Id), s4 = ProjectPath(Get(a)/Id)] [1] scan s1 s2 none none none none none none lowPriority [] @\"136d26ac-0c66-4e1c-8eb6-ad0b089ce557\" true false " }
where the sum of aggregated numeric values is in s11 and the count of these is in s12. If the denominator is non-zero, these are divided to produce the average in s13, else the average (s13) is set to null.
The integration test written for this ticket should verify both correctness and that $avg actually uses the block processing code path where expected.