|
While working on SERVER-53987 I came across these failures in my patch isolated to the cst_jscore_passthrough suite. I've disabled these tests in cst_jscore_passthrough.yml for the time being.
Looking at the timeseries_bucket_limit_count.js failure, the query is returning incorrect results. We expect 1100 documents but only get 2.
assert: [1100] != [2] are not equal : [ { "x" : 0, "_id" : 0 }, { "x" : 1000, "_id" : 1000 } ]
|
doassert@src/mongo/shell/assert.js:20:14
|
assert.eq@src/mongo/shell/assert.js:179:9
|
runTest@jstests/core/timeseries/timeseries_bucket_limit_count.js:52:5
|
@jstests/core/timeseries/timeseries_bucket_limit_count.js:93:1
|
@jstests/core/timeseries/timeseries_bucket_limit_count.js:13:2
|
I printed the explain() output, but couldn't see anything out of the ordinary:
{
|
"stages" : [
|
{
|
"$cursor" : {
|
"queryPlanner" : {
|
"plannerVersion" : 1,
|
"namespace" : "timeseries_bucket_limit_count.system.buckets.t_1",
|
"indexFilterSet" : false,
|
"parsedQuery" : {
|
},
|
"queryHash" : "8B3D4AB8",
|
"planCacheKey" : "8B3D4AB8",
|
"maxIndexedOrSolutionsReached" : false,
|
"maxIndexedAndSolutionsReached" : false,
|
"maxScansToExplodeReached" : false,
|
"winningPlan" : {
|
"stage" : "COLLSCAN",
|
"direction" : "forward"
|
},
|
"rejectedPlans" : [ ]
|
}
|
}
|
},
|
{
|
"$_internalUnpackBucket" : {
|
"exclude" : [ ],
|
"timeField" : "time"
|
}
|
},
|
{
|
"$project" : {
|
"_id" : true,
|
"x" : true
|
}
|
},
|
{
|
"$sort" : {
|
"sortKey" : {
|
"_id" : 1
|
}
|
}
|
},
|
{
|
"$project" : {
|
"_id" : true,
|
"x" : true
|
}
|
}
|
],
|
"serverInfo" : {
|
"host" : "ip-10-122-10-73",
|
"port" : 20000,
|
"version" : "0.0.0",
|
"gitVersion" : "unknown"
|
},
|
"command" : {
|
"aggregate" : "system.buckets.t_1",
|
"pipeline" : [
|
{
|
"$_internalUnpackBucket" : {
|
"timeField" : "time",
|
"exclude" : [ ]
|
}
|
},
|
{
|
"$sort" : {
|
"_id" : 1
|
}
|
},
|
{
|
"$project" : {
|
"x" : 1
|
}
|
}
|
],
|
"cursor" : {
|
},
|
"collation" : {
|
"locale" : "simple"
|
}
|
},
|
"ok" : 1
|
}
|
I suspect this may be an issue with the CST parser as it's only failing in the cst_jscore_passthrough suite for the all feature flags variant. I wonder if removing the _id index table on time-series collections triggered some unexpected behaviour.
|